The way I see it, artificial intelligence (or AI), really leapt into the zeitgeist in late-2022 or early-2023 with the public introduction of DALL-E2 and ChatGPT. Both are provided by OpenAI and are software products that use AI to generate art and writing, respectively (and often at astounding quality). Since then, developments in AI have progressed at a breathtaking pace.
Meanwhile, the latest earnings season for the US stock market – for the third quarter of 2023 – is coming to its tail-end. I thought it would be useful to collate some of the interesting commentary I’ve come across in earnings conference calls, from the leaders of technology companies that I follow or have a vested interest in, on the topic of AI and how the technology could impact their industry and the business world writ large. This is an ongoing series. For the older commentary:
With that, here are the latest commentary, in no particular order:
Airbnb’s management sees generative AI as an opportunity to reimagine the company’s product and transform Airbnb into the ultimate travel agent
First, I think that we are thinking about generative AI as an opportunity to reimagine much of our product category and product catalog. So if you think about how you can sell a lot of different types of products and new offerings, generative AI could be really, really powerful. It can match you in a way you’ve never seen before. So imagine Airbnb being almost like the ultimate travel agent as an app. We think this can unlock opportunities that we’ve never seen.
Airbnb’s management believes that digital-first travel companies will benefit from AI faster than physical-first travel companies
So Airbnb and OTAs are probably going to benefit more quickly from AI than, say, a hotel will just because Airbnb and OTAs are more digital. And so the transformation will happen at the digital surface sooner.
Airbnb’s management believes that Airbnb’s customer service can improve significantly by placing an AI agent between a traveller and her foreign host
One of the areas that we’re specifically going to benefit is customer service. Right now, customer service in Airbnb is really, really hard, especially compared to hotels. The problem is, imagine you have a Japanese host booking with — hosting a German guest and there’s a problem, and you have these 2 people speaking different languages calling customer service, there’s a myriad of issues, there’s no front desk, we can’t go on-premise. We don’t understand the inventory, and we need to try to adjudicate an issue based on 70 different policies that can be up to 100 pages long. AI can literally start to solve these problems where agents can supervise a model that can — in second, come up with a better resolution and provide front desk level support in nearly every community in the world.
Airbnb’s management believes that AI can lead to a fundamentally different search experience for travellers
But probably more importantly, Kevin, is what we can do by reimagining the search experience. Travel search has not really changed much in 25 years since really Expedia, Hotels.com, it’s pretty much the same as it’s been. And Airbnb, we fit that paradigm. There’s a search box, you enter a date location, you refine your results and you book something. And it really hasn’t changed much for a couple of decades. I think now with AI, there can be entirely different booking models. And I think this is like a Cambrian moment for like the Internet or mobile for travel where suddenly an app could actually learn more about you. They could ask you questions and they could offer you a significantly greater personalized service. Before the Internet, there were travel agents, and they actually used to learn about you. And then travel got unbundled, it became self-service and it became all about price. But we do think that there’s a way that travel could change and AI could lead the way with that.
Airbnb’s management believes that all travel apps will eventually trend towards being an AI travel agent
And I generally think for sure, as Airbnb becomes a little more of a so-called like AI travel agent, which is what I think all travel apps will trend towards to some extent.
Alphabet’s management has learnt a lot from trials of Search Generative Experience (SGE), and the company has added new capabilities (videos and images); Search Generative Experience has positive user feedback and strong adoption
This includes our work with the Search Generative Experience, which is our experiment to bring generative AI capabilities into Search. We have learned a lot from people trying it, and we have added new capabilities like incorporating videos and images into responses and generating imagery. We have also made it easier to understand and debug generated code. Direct user feedback has been positive with strong growth in adoption.
SGE allows Alphabet to serve a wider range of information needs and provide more links; ads will continue to be relevant in SGE and users actually find ads useful in SGE; Alphabet wants to experiment with SGE-native ad formats
With generative AI applied to Search, we can serve a wider range of information needs and answer new types of questions, including those that benefit from multiple perspectives. We are surfacing more links with SGE and linking to a wider range of sources on the results page, creating new opportunities for content to be discovered. Of course, ads will continue to play an important role in this new Search experience. People are finding ads helpful here as they provide useful options to take action and connect with businesses. We’ll experiment with new formats native to SGE that use generative AI to create relevant, high-quality ads customized to every step of the Search journey.
Alphabet’s management thinks SGE could be a subscription service; it’s still very early days in the roll-out of SGE and management wants to get the user experience correct (Alphabet has gone through similar transitions before, so management is confident about this)
And I do think over time, there will be newer paths, just like we have done on YouTube. I think with the AI work, there are subscription models as a possible path as well. And obviously, all of the AI investments we are doing applies across Cloud, too, and I’m pretty optimistic about what’s ahead there as well…
…On the first part about SGE, we are still in very, very early days in terms of how much we have rolled it out, but we have definitely gotten it out to enough people both geographically across user segments and enough to know that the product is working well, it improves the experience and — but there are areas to improve, which we are fine-tuning. Our true north here is getting at the right user experience we want to, and I’m pretty comfortable seeing the trajectory. And we’ve always worked through these transitions, be it from desktop to mobile or from now mobile to AI and then to experience. And so it’s nothing new.
Alphabet is making it easier for people to identify AI-generated content through digital watermarks
One area we are focused on is making sure people can more easily identify when they are encountering AI-generated content online. Using new technology powered by Google DeepMind SynthID, images generated by Vertex AI can be watermarked in a way that is invisible to the human eye without reducing the image quality. Underlying all this work is the foundational research done by our teams at Google DeepMind and Google Research.
Alphabet’s management is committed to changing Alphabet’s cost base to accommodate AI investments; Alphabet has, for a long time, driven its cost curves down spectacularly, and management is confident that it will be the same for the current build-out of AI infrastructure
As we expand access to our new AI services, we continue to make meaningful investments in support of our AI efforts. We remain committed to durably reengineering our cost base in order to help create capacity for these investments in support of long-term sustainable financial value. Across Alphabet, teams are looking at ways to operate as effectively as possible focused on their biggest priorities…
…When I looked at the strength of the work we have done across our infrastructure as a company, our technical infrastructure as a company, and various given stages, at a given moment in time when we adopted new generations of technology, we have looked at the cost of it. But then the curves, the efficiency curves, we have driven on top of it has always been phenomenal to see. And I see the current moment as no different. Already through this year, we are driving significant efficiencies both in our models, in training costs and serving costs and our ability to adapt what’s needed to the right use case.
Alphabet has new tools (including those powered by AI) that make it easier for (1) creators to produce content for Youtube’s various formats, (2) creators to connect with advertisers, and (3) advertisers drive higher ROI on advertising
At Made On YouTube in September, we announced new tools that make it easier to create engaging content. Dream Screen is an experimental feature that allows creators to add AI-generated video or image backgrounds to Shorts. And YouTube Create is a new mobile app with a suite of production tools for editing Shorts, longer videos or both…
…AI will do wonders for creation and storytelling. From Dream Screen and YouTube Create, which Sundar talked about, to features that audit up content in multiple languages, flip interim existing assets, remix and clip videos and more, we’re just getting started. We’re also helping brands break through its speed and scale across the funnel to drive results. Spotlight Moments launched last week. It uses AI to identify trending content around major cultural moments for brand sponsorship opportunities. There’s video reach campaigns, which are expanding to in-feed and Shorts, and will be generally available in November. AI is helping advertisers find as many people as possible and their ideal audience for the lowest possible price. Early tests are delivering 54% more reach at 42% lower cost. And then with video view campaigns, AI is serving skippable ads across in-stream, in-feed and Shorts and helping advertisers earn the maximum number of views at the lowest possible cost. So far, they’re driving 40% more views on average versus in-stream alone. Then for YouTube and other feed-based services, there’s our new demand-gen campaign, which launched in April, rolled out worldwide last week and was designed for the needs of today’s social marketers to engage people as they stream, scroll and connect. It combines video and image ads in one campaign with access to 3 billion users across YouTube and Google and the ability to optimize and measure across the funnel using Google AI. Demand gen is already driving successful brands like Samsung and Toyota.
Alphabet’s management believes that Google Cloud offers optimised infrastructure for AI training and inference, and more than 50% of all generative AI start-ups are using Google Cloud; Alphabet’s TPUs (tensor processing units) are winning customers; Google Cloud’s Vertex AI platform offers more than 100 AI models and the number of active generative AI projects built on Vertex AI grew by seven times sequentially
We offer advanced AI optimized infrastructure to train and serve models at scale. And today, more than half of all funded generative AI start-ups are Google Cloud customers. This includes AI21 Labs, Contextual, Elemental Cognition, Writer and more. We continue to provide the widest choice of accelerator options. Our A3 VMs [virtual machines] powered by NVIDIA’s H100 GPU are generally available, and we are winning customers with Cloud TPU v5e, our most cost efficient and versatile accelerator to date. On top of our infrastructure, our Vertex AI platform helps customers build, deploy and scale AI-powered applications. We offer more than 100 models, including popular third-party and open source models, as well as tools to quickly build Search in conversation use cases. From Q2 to Q3, the number of active generative AI projects on Vertex AI grew by 7x, including Highmark Health, which is creating more personalized member materials.
Duet AI, Alphabet’s AI assistant, is built on Google’s large foundation models and is used by large companies to boost developer productivity and smaller companies to help with data analytics; more than 1 million testers have used Duet AI in Google Workspace
Duet AI was created using Google’s leading large foundation models and is specially trained to help users to be more productive on Google Cloud. We continue expanding its capabilities and integrating it across a wide range of cloud products and services. With Duet AI, we are helping leading brands like PayPal and Deutsche Bank boost developer productivity, and we are enabling retailers like Aritzia and Gymshark to gain new insights for better and faster business results…
…In Workspace, thousands of companies and more than 1 million trusted testers have used Duet AI. They are writing and refining content in Gmail and Docs, creating original images from text within slides, organizing data and sheets and more.
Alphabet’s new consumer hardware products have an AI chip – Tensor G3 – built in them
Our portfolio of Pixel products are brought to life, thanks to our combination of foundational technologies AI, Android and Google Tensor. Google Tensor G3 is the third generation of our tailor-built chip. It’s designed to power transformative experiences by bringing the latest in Google AI research directly to our newest phones.
Gemini is the foundation of the next-generation AI models that Google Deepmind will be releasing throughout 2024; Gemini will be multi-modal and will be used internally across all of Alphabet’s products as well as offered externally via Vertex
On Gemini, obviously, it’s effort from our combined Google DeepMind team. I’m very excited at the progress there as we’re working through getting the model ready. To me, more importantly, we are just really laying the foundation of what I think of as the next-generation series of models we’ll be launching throughout 2024. The pace of innovation is extraordinarily impressive to see. We are creating it from the ground up to be multimodal, highly efficient tool and API integrations and, more importantly, laying the platform to enable future innovations as well. And we are developing Gemini in a way that it is going to be available at various sizes and capabilities, and we’ll be using it immediately across all our products internally as well as bringing it out to both developers and cloud customers through Vertex. So I view it as a journey, and each generation is going to be better than the other. And we are definitely investing, and the early results are very promising.
Alphabet’s AI tools are very well received by advertisers and nearly 80% of advertisers use at least one AI-powered search ads product
Our AI tools are very well received, AI, gen AI are top of mind for everybody, really. There’s a ton of excitement, lots of questions about it. Many understand the value. Nearly 80% of our advertisers already use at least one AI-powered search ads product. And yes, we’re hearing a lot of good feedback on, number one, our ads AI Essentials, which are really helping to unlock the power of AI and set up for durable ROI growth on the advertiser side, this is — those are products like the foundation for data and measurement, things like Google Tech, consent mode and so on; and obviously, Search and PMax, we talked about it; and then all the gen AI products, all those different ones. So there’s a whole lot of interest in those products, yes.
Anthropic, a high-profile AI startup, recently chose AWS as its primary cloud provider, and Anthropic will work with Amazon to further develop Amazon’s Trainium (for training AI models) and Inferentia (for AI inference work) chips; Amazon’s management believes the collaboration with Anthropic will help Amazon bring further price performance advantages to Trainium and Inferentia
Recently, we announced the leading LLM maker Anthropic chose AWS as its primary cloud provider. And we’ll use Trainium training and Inferentia to build, trade and deploy future LLMs. As part of this partnership, AWS and Anthropic will collaborate on the future development of training and inferential technology. We believe this collaboration will be helpful in continuing to accelerate the price performance advantages that Trainium and Inferentia deliver for customers.
Perplexity is another AI startup that chose to run their models with Trainium and Inferentia
We are also seeing success with generative AI start-ups like Perplexity AI who chose to go all in with AWS, including running future models in Trainium and Inferentia.
Amazon’s management believes that Amazon’s Trainium and Inferentia chips are very attractive to people in the industry because they offer better price-performance characteristics and they can meet demand; Anthropic and Perplexity’s decisions to go with Trainium and Inferentia are statements to that effect
I would also say our chips, Trainium and Inferentia, as most people know, there’s a real shortage right now in the industry and chips, it’s really hard to get the amount of GPUs that everybody wants. And so it’s just another reason why Trainium and Inferentia are so attractive to people. They have better price performance characteristics than the other options out there, but also the fact that you can get access to them. And we’ve done a I think, a pretty good job providing supply there and ordering meaningfully in advance as well. And so you’re seeing very large LLM providers make big bets on those chips. I think anthropic deciding to train their future LLM model on Trainium and using Inferentia as well is really a statement. And then you look at the really hot start-up perplexity.ai, who also just made a decision to do all their Trainium and Inferentia on top of Trainium and Inferentia. So those are two examples.
Amazon recently announced the general availability of Amazon Bedrock (AWS’s LLMs-as-a-service), which gives access to a variety of 3rd-party large language models (LLMs) as well as Amazon’s own LLM called Titan; Meta’s Llama-2 LLM will also be on Bedrock, the first time it is available through a fully-managed service
In the middle layer, which we think of as large language models as a service, we recently introduced general availability for Amazon Bedrock, which offers customers access to leading LLMs from third-party providers like anthropics, stability AI, coherent AI 21 as well as from Amazon’s own LLM called Titan, where customers can take those models, customize them using their own data, but without leaking that data back into the generalized LLM have access to the same security, access control and features that they run the rest of their applications with in AWS all through a managed service. In the last couple of months, we’ve announced the imminent addition of Meta’s Llama 2 model to Bedrock the first time it’s being made available through a fully managed service.
Amazon’s management believes that Bedrock helps customers experiment rapidly with different LLMs and is the easiest way to build and scale enterprise-ready generative AI applications; customer reaction to Bedrock has been very positive;
Also through our expanded collaboration with Anthropic, customers will gain access to future anthropic models through bedrock with exclusive early access to unique features model customization and the ability to fine-tune the models. And Bedrock has added several new compelling features, including the ability to create agents which can be programmed to accomplish tasks like answering questions or automating workflows. In these early days of generative AI, companies are still learning which models they want to use, which models they use for what purposes and which model sizes they should use to get the latency and cost characteristics they desire. In our opinion, the only certainty is that there will continue to be a high rate of change. Bedrock helps customers with this fluidity, allowing them to rapidly experiment with move between model types and sizes and enabling them to pick the right tool for the right job. The customer reaction to Bedrock has been very positive and the general availability is buoyed that further. Bedrock is the easiest way to build and scale enterprise-ready generative AI applications and a real game changer for developers and companies trying to get value out of this new technology…
Bedrock’s ability to let customers conduct fast experiments is very useful because customers sometimes get surprised at the true costs of running certain AI models
Because what happens is you try a model, you test the model, you like the results of the model and then you plug it into your application and what a lot of companies figure out quickly is that using the really large — the large models and the large sizes ends up often being more expensive than what they anticipated and what they want to spend on that application. And sometimes too much latency in getting the answers as it shovels through the really large models. And so customers are experimenting with lots of different types of models and then different model sizes to get the cost and latency characteristics that they need for different use cases. It’s one of the things that I think is so useful about Bedrock is that customers are trying so many variants right now but to have a service that not only lets you leverage lots of third party as well as Amazon large language miles, but also lots of different sizes and then makes the transition of moving those workloads easy between them is very advantageous.
Amazon Code Whisperer, AWS’s coding companion, has a lot of early traction and has become more powerful recently by having the capability to be customised on a customer’s own code base (a first-of-its kind feature)
Generative AI coding companion Amazon Code Whisper has gotten a lot of early traction and got a lot more powerful recently with the launch of its new customization capability. The #1 enterprise request for coding companions has been wanting these companions to be familiar with customers’ proprietary code bases is not just having code companions trained on open source code. Companies want the equivalent of a long-time senior engineer who knows their code base well. That’s what Code Whisper just launched, another first of its kind out there in its current forum and customers are excited about it.
Amazon’s management believes that customers want to bring AI models to their data, not the other way around – and this is an advantage for AWS as customers’ data resides within AWS
It’s also worth remembering that customers want to bring the models to their data, not the other way around. And much of that data resides in AWS as the clear market segment leader in cloud infrastructure.
There are many companies that are building generative AI apps on AWS and this number is growing fast
The number of companies building generative AI apps and AWS is substantial and growing very quickly, including Adidas, Booking.com, Bridgewater, Clarient, GoDaddy, Lexus Nexus, Merck, Royal Philips and United Airlines, name a few
Generative AI’s growth rate within AWS is very fast – even faster than Amazon’s management expected – and management believes that the absolute amount of generative AI business within AWS compares very favourably with other cloud providers
I could see it also just the growth rate for us in generative AI is very fast. Again, I have seen a lot of different numbers publicly. It’s real hard to measure an apples-to-apples. But in our best estimation, our — the amount of growth we’re seeing in the absolute amount of generative AI business we’re seeing compares very favorably with anything else I’ve seen externally.
Generative AI is already a pretty significant business for AWS, but it’s still early days
What I would tell you is that we have been surprised at the pace of growth in generative AI. Our generative AI business is growing very, very quickly, as I mentioned earlier. And almost by any measure, it’s a pretty significant business for us already. And yet I would also say that companies are still in the relatively early stages.
All of Amazon’s significant businesses are working on generative AI applications, with examples including using generative AI to (1) help consumers discover products, (2) forecast inventory in various locations, (3) help 3rd-party sellers create new product pages, (4) help advertisers with image generation for ads, and (5) improve Alexa
Beyond AWS, all of our significant businesses are working on generative AI applications to transform their customer experiences. There are too many for me to name on this call, but a few examples include, in our stores business, we’re using generative AI to help people better discover products they want to more easily access the information needed to make decisions. We use generative AI models to forecast inventory we need in our various locations and to derive optimal last mile transportation routes for drivers to employ. We’re also making it much easier for our third-party sellers to create new product pages by entering much less information and getting the models to the rest. In advertising, we just launched a generative AI image generation tool, where all brands need to do is upload a product photo and description to quickly create unique lifestyle images that will help customers discover products they love. And in Alexa, we built a much more expansive LLM and previewed the early version of this. Apart from being a more intelligent version of herself, Alexa’s new conversational AI capabilities include the ability to make multiple requests at once as well as more natural and conversational requests without having to use specific phrases.
Amazon’s management still believes in the importance of building the world’s best personal assistant and they thinksAlexa could be one of these assistants
We continue to be convicted that the vision of being the world’s best personal assistant is a compelling and viable one and that Alexa has a good chance to be one of the long-term winners in this arena.
While Amazon’s management is pulling back Amazon’s capital expenditure on other areas, they are increasing capital expenditure for AI-related infrastructure
For the full year 2023, we expect capital investments to be approximately $50 billion compared to $59 billion in 2022. We expect fulfillment and transportation CapEx to be down year-over-year partially offset by increased infrastructure CapEx, support growth of our AWS business, including additional investments related to generative AI and large language model efforts.
Apple’s management sees AI and machine learning as fundamental technologies to the company and they’re integrated in virtually every product that Apple ships
If you kind of zoom out and look at what we’ve done on AI and machine learning and how we’ve used it, we view AI and machine learning as fundamental technologies, and they’re integral to virtually every product that we ship.
Apple’s AI-powered features include Personal Voice in iOS17, and fall detection, crash detection, and ECG on the Apple Watch; Apple’s management does not want to label Apple’s AI-powered features with “AI” – instead the features are labelled as consumer benefits
And so just recently, when we shipped iOS 17, it had features like Personal Voice and Live Voicemail. AI is at the heart of these features. And then you can go all the way to then life-saving features on the Watch and the phone like fall detection, crash detection, ECG on the watch. These would not be possible without AI. And so we don’t label them as such, if you will. We label them as to what their consumer benefit is, but the fundamental technology behind it is AI and machine learning.
Apple is investing in generative AI but management has no details to share yet
In terms of generative AI, we have — obviously, we have work going on. I’m not going to get into details about what it is because as you know, we really don’t do that. But you can bet that we’re investing, we’re investing quite a bit. We are going to do it responsibly. And it will — you will see product advancements over time where those technologies are at the heart of them.
From the vantage point of Arista Networks’ management, Oracle has become an important AI data centre company
Our historic classification of our Cloud Titan customers has been based on industry definition of customers with or likely to attain greater than 1 million installed compute service. Looking ahead, we will combine Cloud and AI customer spend into one category called Cloud and AI Titan sector. And as a result of this combination, Oracle OCI becomes a new member of the sector, while Apple shift to cloud specialty providers…
…So I think OCI has become a meaningful top-tier cloud customer and they belong in the cloud tightening category and in addition to their AI investments as well. So for reasons of classification and definition, the change is very warranted. And yes, they happened to be a good customer of Arista, that’s nice as well.
Arista Networks’ management has observed that its large customers have different needs when it comes to AI and non-AI networking technologies
During the past year, our Cloud Titan customers have been planning a different mix of AI networking and classic cloud networking for their compute and storage clusters.
Arista Networks’ management believes that the company’s recent deal with a public sector organisation to provide Ethernet networking technology for the organisation’s AI initiative is an example of why Ethernet is important in AI
Our next [ one ] showcases our expansion of Arista in the public sector with their AI initiative. This grant-funded project utilizes Arista simplified operational models with CloudVision. New AI workloads require high scale, high ratings, high bandwidth and low latency as well as a need for granular visibility. This build out of a single EVPN-VXLAN based 400-gig fabric is based on deep buffers fines and underscores the importance of a lossless architecture for AI networking.
Arista Networks’ management is seeing its customers prioritise AI in their data centre spending right now, but demand for other forms of data centre-related spending will follow
We’ve always looked at that the cloud network as a front end and the back end. And as we said last year, many of our cloud customers are favoring spending more on the back end with AI, which doesn’t mean they stop spending on front end, but they’re clearly prioritized and doubled down on AI this year. My guess is as we look at the next few years, they’ll continue to double down on AI. But you cannot build an AI bank cluster without thinking of the front end. So we’ll see a full cycle here, while today the focus is greatly on AI and the back end of the network. In the future, we expect to see more investments in the front end as well.
Arista Networks’ management sees AI networking as being dominated by Infiniband today- with some room for a combination of Infiniband and Ethernet – but they still believe that AI networking will trend toward Ethernet over time, with 2025 being a potential inflection point
Today if I look at the 5 major designs for AI networking, one of them is still very InfiniBand dominated, all the others we’re looking at is — are adopting on dual strategy of both Ethernet and InfiniBand. So I think AI networking is going to become more and more favorable to Ethernet, particularly with the Ultra Ethernet Consortium and the work they’re doing to define a spec, you’re going to see more products based on UEC. You’re going to see more of a connection between the back end and the front-end using IP as a singular protocol. And so we’re feeling very encouraged that especially in 2025, there will be a lot of production rollout of back end and, of course, front end based on Ethernet.
Arista Networks’ management sees networking spend as contributing to 10%-15% of the total cost of an AI data centre
Coming back to this networking spend versus the rest of the GPUs and et cetera, I would say it started to get higher and higher with 100-gig, 400-gig and 800-gig, where the optics and the switches are more than 10%, perhaps even 15% in some cases, 20, a lot of its governed by the cables and optics too. But the percentage hasn’t changed a lot in high-speed networking. In other words, it’s not too different between 10, 100, 200, 400 and 800. So we — you’ll continue to see that 10% to 15% range.
Arista Networks’ management sees diversified activity when it comes to the development of AI data centres
[Question] And just what you’re seeing in terms of other people kind of building out some of these AI clusters, if you classify some of those customers as largely focused on back end today, and those represent opportunities going forward? Or just kind of what the discussion is outside of the Cloud Titans amongst some of these other guys that are building very large networks?
[Answer] The Tier 2 cloud providers are doing exactly what the Tier 1 is doing just at a smaller scale. So the activity is out there. Many companies are trying to build these clusters, maybe not hundreds of thousands GPUs but thousands of GPUs together in their real estate if they can get them. But the designs that we’re working on with them, the type of sort of features, fine-tuning is actually very, very similar to the cloud, just at a smaller scale. So we’re very happy with that activity and this is across the board. It’s very positive to see this in the ecosystem that it’s not limited just 4 or 5 customers.
Arista Networks’ management is observing that data centre companies are facing a shortage of GPUs (graphics processing units) and they are trying to develop AI with smaller GPU clusters
I think they’re also waiting for GPUs like everyone else is. So there’s that common problem that we’re not the only one with lead time issues. But to clarify the comment on scale, Anshul and I are also seeing some very interesting enterprise projects against smaller scale. So a lot of customers are trying AI for small clusters, not too different from what we saw with HPC clusters back in the day.
Arista Networks’ management believes that good networking technology for AI requires not just good silicon, but the right software, so they are not concerned about Arista Networks’ suppliers moving up the stack
It’s not just the merchant silicon but how you can enable the merchant silicon with the right software and drivers, and this is an area that really Arista excels, and if you just have chips, you can’t build the system. But our system-wide features, whether it’s a genetic load balancing, or latency analyzer to really improve the job completion time and deal with that frequent communication and generative AI is also fundamentally important…
… [Question] So I think there was a mention on merchant silicon earlier in the Q&A. And one of your merchant silicon partners has actually moved up the stack towards the service provider routing. I’m just curious if there’s any intention on going after that piece if that chip is made available to you?
[Answer] I believe you are referring to the latest announcement Broadcom on their 25.60 Jericho chip that was announced recently.
[Question] Yes, the Qumran3D.
[Answer] Qumran3D, exactly. So it’s the same family, same features. And as you know, we’ve been a great partner of Broadcom for a long time, and we will continue to build new products. This is not a new entry, so to speak. We’ve been building these products that can be used on switches our orders for a while, and that bandwidth just doubled going to now 25.6. So you can expect some products from us in the future with those variants as well. But really — nothing really changed…
…And the investment we have made in our routing stack over the last 10 years, I want to say, has just gotten better and stronger. Power in the Internet, power in the cloud, power in the AI, these are hard problems. And they require thousands of engineers of investment to build the right VXLAN, BGP routing, EVPN, et cetera. So it’s not just a chip. It’s how we name the chip to do these complicated routing algorithms.
AI is becoming a really important component of Arista Networks’ customers
We’re simply seeing AI is going to become such an important component of all our cloud titans that is now a combined vertical.
Datadog’s management is excited about generative AI and large language models and they believe that the adoption of AI will lead to additional growth in cloud workloads
Finally, we continue to be excited about the opportunity in generative AI and Large Language Models. First, we believe adopting NextGen AI will require the use of cloud and other modern technologies and drive additional growth in cloud workloads.
Datadog is building LLM observability products
So we are continuing to invest by integrating with more components at every layer of the new AI stack and by developing our own LLM observability products.
Datadog’s management is seeing adoption of AI across many of its customers, but the activity is concentrated in AI-native customers
And while we see signs of AI adoption across large parts of our customer base, in the near term, we continue to see AI-related usage manifest itself most accurately with next-gen AI native customers who contributed about 2.5% of our ARR this quarter.
Datadog is adding value to its own platform using AI with one example being Bits AI, Datadog’s test-and-analysis tool
Besides observing the AI stack, we also expect to keep adding value to our own platform using AI. Datadog’s unified platform and purely SaaS model, combined with strong multiproduct adoption by our customers generates a large amount of deep and precise observability data. We believe combining AI capabilities with this broad data set will allow us to deliver differentiated value to customers. And we are working to productise differentiated value through recently announced capabilities such as our Bits AI assistant, AI generated synthetic test and AI-led air analysis and resolution, and we expect to deliver many more related innovation to customers over time.
Datadog’s management is seeing that AI-native customers are using Amazon’s AWS whereas the larger enterprises that are using AI are using Microsoft’s Azure
Interestingly enough, the — when we look at our cohort of customers that are that we consider to be AI native and built largely on AI in all AI providers, they tend to be on different clouds. What we see is that the majority of those companies actually have a lot of their usage on AWS. Today, the larger part of the usage or the larger of these customers are on Azure. So we see really several different adoption trends there that I think are interesting to the broader market.
Datadog’s management is seeing broad usage of AI across Datadog’s customers, but the customers are adopting AI only at low volumes
Whereas we see broad usage of AI functionality across the customer base, but at low volumes, and it corresponds to the fact that for most customers or most enterprises really, they’re still in the early stages of developing and shipping applications. So for now, the usage is concentrated among the model providers.
Datadgo’s management sees a lot of opportunity for Datadog as AI usage proliferates – for example, management believes that the widespread use of AI will result in the creation of a lot of code and these code will need to be monitored
So on the DevSecOps side, I think it’s too early to tell how much the revenue opportunity there is in the tooling specific lab there. When you think of the whole spectrum of tools, the closer you get to the developer side to how are is to monetize and the further you get towards operations and infrastructure, the easier it is to monetize. You can ship things that are very useful and very accretive to our platform because they get you a lot of users, a lot of attention and a lot of stickiness that are harder to monetize. So we’ll see where on the spectrum that is. What we know, though, is that broader Generative AI up and down the stack from the components themselves, the GPUs all the way up to the models and the various things that are used to orchestrate them and store the data and move the data around all of that is going to generate a lot of opportunity for us. We said right now, it’s conciliated among the AI native largely model providers. But we see that it’s going to broaden and concern a lot more of our customers down the road…
…So in general, the more complexity there is, the more useful observability, the more you see his value from writing code to actually understanding it and observing it. So to caricature if you — if you spend a whole year writing 5 lines of code that are really very deep, you actually know those 5 lines pretty well, maybe you don’t observe because you’ll see you understand exactly how they work and what’s going on with them. On the other hand, if thanks to all the major advances of technology and all of the very super source AI and you can just very quickly generate thousands of lines of code, ship them and start operating them, you actually have no idea how these work and what they do. And you need a lot of tooling observability to actually understand that and keep driving that and secure it and do everything you need to do with it over time. So we think that overall, this increases in productivity are going to favor observability.
Datadog’s management is also trying to guess how transformative AI will be, but there are signs that AI’s impact will be truly huge
In terms of the future growth of AI, look, I think like everyone, we’re trying to guess how transformative it’s going to be. It looks like it’s going to be pretty is, if you judge from just internally, how much of that technology we are adopting a how much is the productivity impact, it seems to be having.
AI-related use cases are still just a small fraction of the overall usage of Datadog’s products, but Datadog’s management thinks that AI will drive a lot of the company’s growth in the future
So again, today, we only see a tiny bit of it, which is early adoption by model providers and a lot of companies that are trying to scale up and experiment and figure out who it applies to their businesses and what they can ship to use the technology. But we think it’s going to drive a lot of growth in the years to come.
Datadog’s management can’t tell when Datadog’s broader customer base will start ramping up AI workloads but they are experimenting; most of the innovation happening right now is concentrated among the model providers
[Question] Olivier, you called out the 2.5 points from AI native customers a few times, but you’ve also said that the broader customer base should start adding AI workloads to our platform over time. When do you think that actually takes place and the broader customer base starts to impact that AI growth in more earnest?
[Answer] We don’t know. And I think it’s too early to tell. For one part, there’s some uncertainty in terms of — these customers are being to figure out what it is they are going to ship to their own customers. I think everybody is trying to learn that right now and experiment it. And — but the other part is also that right now, the innovation is largely concentrated among the model providers. And so it’s rational right now for most customers to rely on those instead of they’re deploying their own infrastructure. Again, we think it’s slightly going to change. We see a lot of demand in interest in other ways to host models and run models and customers and all those things like that. But today, that’s the — these are the trends of the market today basically.
Etsy’s management is improving the company’s search function by combining humans and machine learning technology to better identify the quality of each product listing on the Etsy platform
We’re moving beyond relevance to the next frontier of search focused on better identifying the quality of each Etsy listing, utilizing humans and ML technology so that from a highly relevant result set, we bring the very best of Etsy to the top, personalized to what we understand of your tastes and preferences. For example, from the start of the year, we’re tracking to a ninefold increase in the number of human-curated listings on Etsy to over 1.5 million listings by year-end. We’re also utilizing ML models designed to determine the visual appeal of items and incorporating that information into our search algorithms.
Etsy’s management is using generative AI to improve the Etsy search-experience when buyers enter open-ended queries, which helps build purchase-frequency
There’s also a huge opportunity to evolve the Etsy experience so that we show buyers a more diverse set of options when they search for open-ended head query items such as back-to-school. On the left of this slide, you can see an example of how a search for back-to-school items looks on Etsy. We generally show multiple very similar versions of customized pencils, stickers, lawn signs and so on, all mixed together. This is suboptimal as it offers buyers only a few main ideas on the first page of search and requires a ton of cognitive load to distinguish between virtually identical items. We’ve recently launched a variety of experiments with the help of Gen AI to evolve these types of head query searches. As we move into 2024, when a buyer searches for broad queries, we expect to be able to show a far more diverse and compelling set of ideas, all beautifully curated by organizing search results into a number of ideas for you that are truly different and helping to elevate the very best items within each of these ideas, we can take a lot of the hard work out of finding exactly the perfect item. And help build frequency as we highlight the wide range of merchandise available on Etsy.
Etsy’s management is using machine learning to identify product-listings that are not conforming to the company’s product policies, and listing-takedowns are already up 140% year-on-year
We’ve hired a lot of people, and we also have been investing a lot in machine learning and machine learning is really helping us to be able to identify among the 120 million listings on Etsy, those that may not conform with our policy. Takedowns are up 140% year-over-year.
Fiverr’s management has developed Fiverr Neo, a generative AI tool that helps customers scope their projects better and match them with suitable freelance talent, just like a human recruiter would, just better; management believes that Fiverr Neo will help save customers time when they are looking for freelance talent
The vision for Fiverr Neo is quite wild – we imagine Neo will serve as a personalized recruiting expert that can help our customers more accurately scope their projects and get matched with freelance talent, just like a human recruiter, only with more data and more brain power. What we have done so far is leverage the existing LLM engines to allow customers to express their project needs in natural language, which Neo will synthesize and define the scope before matching the client with a short list of choices pulled from the entire Fiverr freelancer database. It’s a substantial step forward from the existing experience and streamlines the time the customer needs to make an informed decision.
Fiverr’s management used a combination of Fiverr’s own software and LLMs from other companies to build Fiverr Neo
So there’s a lot of learning as we build this product. And what we’re doing is really a hybrid of technologies. Some of them are being developed by us. Some are off the shelf, most of the leading companies that are developing LLM, which have partnered with us. And we’re putting this to the maximum. I think a lot of these systems are not yet optimized for large scale and high performance but we find our own ways of developing a lot of this technology to provide a very smooth experience to our customers.
Fiverr Neo is still new, but users are already experiencing more accurate matches
In terms of Fiverr neo, we’re very pleased with the rollout. Obviously, very, very young product, but we’re seeing over 100,000 users that are trying the product. And what we’re seeing from their experience is that we’re able to provide more accurate matches, which is basically what we wanted to do and have a higher engagement and satisfaction levels, which we’re very happy with and the beginning of a repeat usage of the product.
Fiverr’s management thinks that AI has a positive impact on the product categories that Fiverr can introduce to its marketplace and management is ensuring that Fiverr’s catalog will contain any new skills that the AI-age will require; management thinks that a lot of AI hype at the beginning of the year has died down and the world is looking for killer AI applications
So I did address this also in how we think about next year and the fact that AI both impact the efficiency of how we work allows us to do pretty incredible things in our product. It also has an impact — positive impact on the categories that we can introduce. So again, we’re not getting into specific category breakdown. But what we’re seeing on the buyer side, I think we’ve introduced these categories, these categories continue growing. I think that a lot of the height that surrounded AI in the beginning of the year subsided and right now, it’s really looking for the killer applications that could be developed with AI, and we’re developing some of them and our customers are as well. So these are definitely areas where we continue seeing growth, but not just that, but we continue investing in the catalog side to ensure that the new types of skills that pop up are going to be addressed on the Fiverr market base.
Mastercard’s management is using AI to improve the company’s fraud-related solutions and has signed agreements in Argentina, Saudi Arabia, and Nigeria in this area
AI also continues to play a critical role powering our products and fueling our network intelligence. We’re scaling our AI-powered transaction fraud monitoring solution, which delivers real-time predictive scores based on a unique blend of customer and network level insights. This powerful solution gives our customers the ability to take preventive action before the transaction is authorized. This quarter alone, we signed agreements in Argentina, Saudi Arabia and Nigeria with financial institutions and fintechs who will benefit from early fraud detection and with merchants who will experience less friction and higher approval rates.
MercadoLibre’s management is very excited about AI and how it can help MercadoLibre improve the user experience and its business operations
As you know, we don’t guide, but there are many exciting things going on, particularly, obviously, AI. That hopefully will enable us to provide our users a better experience, enable us to launch innovative ideas, and also scale and gain efficiencies, whether it is in customer service, or whether it is in fraud prevention or whether it is in the way our developers, 15,000 developers, go about developing and performing quality control, et cetera. So obviously, looking forward for the next 3 years, I think that’s a key thing to look into.
MercadoLibre’s management is working on using AI to improve the company’s product-search function and they are happy with the progress so far
Last question in terms of AI and search, we are working on that. I mean we are putting a lot of effort into building solutions around AI. I think we don’t have much to disclose as of now, but search, reviews, questions and answers, buy box and products, as Marcos was saying, copilot for our developer. We’re looking at the broad range of AI uses for MercadoLibre to boost consumer demand and efficiency. And we’re happy with the progress that we have so far, but not much to be said yet.
MercadoLibre’s management has been using AI for many years in fraud prevention and credit scoring for the company’s services
We have been using AI for a long time now for many, many years, both in terms of fraud prevention and credit scoring. Both 2 instances, they are pretty much use cases which are ideal for AI, because we have, in the case of fraud prevention, millions of transactions every day and with a clear outcome, either fraud or not fraud. So with the right variables, we can build a very strong model that has predicted and have really best-in-class fraud prevention. And with that knowledge and given the experience we have been building on credits, we have also been — built our credit scoring models leveraging the AI.
The next-generation Ray-Ban Meta smart glasses has embedded AI
The next generation of Ray-Ban Meta smart glasses, which are the first smart glasses with our Meta AI built in.
Meta Platforms’ management thinks glasses are an ideal form-factor for an AI device as it can see exactly what you see and hear what you hear
And in many ways, glasses are the ideal form factor for an AI device because they enable your AI assistant to see what you see and hear what you hear.
Llama 2 is now the leading open source AI model with >30 million downloads last month
We’re also building foundation models like Llama 2, which we believe is now the leading open source model with more than 30 million Llama downloads last month.
Beyond generative AI, Meta Platforms’ management is using recommendation AI systems for the company’s Feeds, Reels, ads, and integrity systems and these AI systems are very important to the company; AI feed recommendations led to increases in time spent on Facebook (7%) and Instagram (6%)
Beyond that, there was also a different set of sophisticated recommendation AI systems that powers our Feeds, Reels, ads and integrity systems. And this technology has less hype right now than generative AI but it is also very important and improving very quickly. AI-driven feed recommendations continue to grow their impact on incremental engagement. This year alone, we’ve seen a 7% increase in time spent on Facebook and a 6% increase on Instagram as a result of recommendation improvements.
Meta Platforms’ AI tools for advertisers has helped drive its Advantage+ advertising product to reach a US$10 billion revenue run-rate, with more than 50% of the company’s advertisers using Advantage+ creative tools
Our AI tools for advertisers are also driving results with Advantage+ shopping campaigns reaching a $10 billion run rate and more than half of our advertisers using our Advantage+ creative tools to optimize images and text and their ads creative.
AI-recommended content has become increasingly incremental to engagement on Meta Platforms’ properties
AI-recommended content from unconnected accounts and feed continues to become increasingly incremental to engagement, including in the U.S. and Canada. These gains are being driven by improvements to our recommendation systems, and we see additional opportunities to advance our systems even further in the future as we deploy more advanced models.
Meta Platforms’ management believes that the company’s Business AIs can easily help businesses set up AIs to communicate with consumers at very low cost, which is important in developed economies where cost of labour is high (businesses in developing economies tend to hire humans to communicate with consumers)
Now I think that this is going to be a really big opportunity for our new Business AIs that I talked about earlier that we hope will enable any business to easily set up an AI that people can message to help with commerce and support. Today, most commerce and messaging is in countries where the cost of labor is low enough that it makes sense for businesses to have people corresponding with customers over text. And in those countries like Thailand or Vietnam, there’s a huge amount of commerce that happens in this way. But in lots of parts of the world, the cost of labor is too expensive for this to be viable. But with business AIs, we have the opportunity to bring down that cost and expand commerce and messaging into larger economies across the world. So making business AIs work for more businesses is going to be an important focus for us into 2024.
Meta Platforms’ management has started testing the company’s AI capabilities with a few partners in business messaging
We’ve recently started testing AI capabilities with a few partners and we’ll take our time to get the experience right, but we believe this will be a big unlock for business messaging in the future.
Meta Platforms’ management still believes in the benefits of open-sourcing Meta’s AI models: It increases adoption (which benefits the company as the security features and cost-efficiency of the models improves) and talent is more attracted to Meta Platforms
We have a pretty long history of open sourcing parts of our infrastructure that are not kind of the direct product code. And a lot of the reason why we do this is because it increases adoption and creates a standard around the industry, which often drives forward innovation faster so we benefit and our products benefit as well as there’s more scrutiny on kind of security and safety-related things so we think that there’s a benefit there.
And sometimes, more companies running models or infrastructure can make it run more efficiently, which helps reduce our costs as well, which is something that we’ve seen with open compute. So I think that there’s a good chance that, that happens here over time. And obviously, our CapEx expenses are a big driver of our costs, so any aid in innovating on efficiency is sort of a big thing there.
The other piece is just that over time with our AI efforts, we’ve tried to distinguish ourselves as being a place that does work that will be shared with the industry and that attracts a lot of the best people to come work here. So a lot of people want to go to the place to work where their work is going to touch most people. One way to do that is by building products that billions of people use. But if you’re really a focused engineer or researcher in this area, you also want to build the thing that’s going to be the standard for the industry. So that’s pretty exciting and it helps us do leading work.
Meta Platforms’ management thinks the AI characters that the company introduced recently could lead to a new kind of medium and art form and ultimately drive increasing engagement for users of the company’s social apps
We’re designing these to make it so that they can help facilitate and encourage interactions between people and make things more fun by making it so you can drop in some of these AIs into group chats and things like that just to make the experiences more engaging. So this should be incremental and create additional engagement. The AIs also have profiles in Instagram and Facebook and can produce content, and over time, going to be able to interact with each other. And I think that’s going to be an interesting dynamic and an interesting, almost a new kind of medium and art form. So I think that will be an interesting vector for increasing engagement and entertainment as well.
Meta Platforms’ management thinks that generative AI is a really exciting technology and that it changes everything and although it’s hard to predict what generative AI’s impact is going to be on how individuals use Meta’s services, they still thinks it’s worth investing in it;In terms of how big this is going to be, it’s hard to predict because I don’t think that anyone has built what we’re building here. I mean, there’s some analogy is like what OpenAI is doing with ChatGPT, but that’s pretty different from what we’re trying to do. Maybe the Meta AI part of what we’re doing overlaps with the type of work that they’re doing, but the AI characters piece, there’s a consumer part of that, there’s a business part, there’s a creators part. I’m just not sure that anyone else is doing this. And when we’re working on things like Stories and Reels, there were some market precedents before that. Here, there’s technology which is extremely exciting. But I think part of what leading in an area and developing a new thing means is you don’t quite know how big it’s going to be. But what I predict is that I do think that the fundamental technology around generative AI is going to transform meaningfully how people use each of the different apps that we build…
…So I think you’re basically seeing that there are going to be — this is a very broad and exciting technology. And frankly, I think that this is partially why working in the technology industry is so awesome, right, is that every once in a while, something comes along like this, that like changes everything and just makes everything a lot better and your ability to just be creative and kind of rethink the things that you’re doing to be better for all the people you serve…
…But yes, it’s hard sitting here now to be able to predict like the metrics are going to be around, like what’s the balance of messaging between AIs and people or what the balance and Feeds between AI content and people content or anything like that. But I mean, I’m highly confident that this is going to be a thing and I think it’s worth investing in.
Meta Platforms’ management believes that generative AI will have a big impact on the digital advertising industry
It’s going to change advertising in a big way. It’s going to make it so much easier to run ads. Businesses that basically before would have had to create their own creative or images now won’t have to do that. They’ll be able to test more versions of creative, whether it’s images or eventually video or text. That’s really exciting, especially when paired with the recommendation AI.
Microsoft’s management is making AI real for everyone through the introduction of Copilots
With Copilots, we are making the age of AI real for people and businesses everywhere. We are rapidly infusing AI across every layer of the tech stack and for every role and business process to drive productivity gains for our customers.
Microsoft’s management believes that Azure has the best AI infrastructure for both training and inference
We have the most comprehensive cloud footprint with more than 60 data center regions worldwide as well as the best AI infrastructure for both training and inference. And we also have our AI services deployed in more regions than any other cloud provider.
Azure AI provides access to models from OpenAI and open-sourced models (including Meta’s) and 18,000 organisations now use Azure OpenAI
Azure AI provides access to best-in-class frontier models from OpenAI and open-source models, including our own as well as from Meta and Hugging Face, which customers can use to build their own AI apps while meeting specific cost, latency and performance needs. Because of our overall differentiation, more than 18,000 organizations now use Azure OpenAI service, including new to Azure customers.
GitHub Copilot increases developer productivity by up to 55%; there are more than 1 million paid Copilot users and more than 37,000 organisations that subscribe to Copilot for business (up 40% sequentially)
With GitHub Copilot, we are increasing developer productivity by up to 55% while helping them stay in the flow and bringing the joy back to coding. We have over 1 million paid Copilot users and more than 37,000 organizations that subscribe to Copilot for business, up 40% quarter-over-quarter, with significant traction outside the United States.
Microsoft’s management is using AI to improve the healthcare industry: Dragon Ambient Experience (from the Nuance acquisition) has been used in more than 10 million patient interactions to-date to automatically document the interactions, andDAX Copilot can draft clinical notes in seconds, saving 40 minutes of documentation time daily for physicians
In health care, our Dragon Ambient Experience solution helps clinicians automatically document patient interactions at the point of care. It’s been used across more than 10 million interactions to date. And with DAX Copilot, we are applying generative models to draft high-quality clinical notes in seconds, increasing physician productivity and reducing burnout. For example, Atrium Health, a leading provider in Southeast United States, credits DAX Copilot with helping its physicians each save up to 40 minutes per day in documentation time.
Microsoft’s management has infused Copilot across Microsoft’s work-productivity products and tens of thousands of users are already using Copilot in early access
Copilot is your everyday AI assistant, helping you be more creative in Word, more analytical in Excel, more expressive in PowerPoint, more productive in Outlook and more collaborative in Teams. Tens of thousands of employees at customers like Bayer, KPMG, Mayo Clinic, Suncorp and Visa, including 40% of the Fortune 100, are using Copilot as part of our early access program.
Users find Copilot amazing and have enjoyed similar productivity gains as developers did with Github Copilot
Customers tell us that once they use Copilot, they can’t imagine work without it, and we are excited to make it generally available for enterprise customers next week. This quarter, we also introduced a new hero experience in Copilot, helping employees tap into their entire universe of work, data and knowledge using chat. And the new Copilot Lab helps employees build their own work habits for this era of AI by helping them turn good prompts into great ones…
…And in fact, the interesting thing is it’s not any one tool, right, which is the feedback even sort of is very clear that it’s the all up. You just keep hitting the Copilot button across every surface, right, whether it’s in Word to create documents, in Excel to do analysis or PowerPoint or Outlook or Teams. Like clearly, the Teams Meeting, which is an intelligent recap, right? It’s not just a dumb transcript. It’s like having a knowledge base of all your meetings that you can query and add to essentially the knowledge terms of your enterprise. And so we are seeing broad usage across and the interesting thing is by different functions, whether it’s in finance or in sales by roles. We have seen productivity gains like we saw with developers in GitHub Copilot.
At the end of the day, Microsoft management is still grounded about the rate of adoption of Copilot in Office, since it is an enterprise product
And of course, this is an enterprise product. I mean at the end of the day, we are grounded on enterprise cycle times in terms of adoption and ramp. And it’s incrementally priced. So therefore, that all will apply still. But at least for something completely new, to have this level of usage already and this level of excitement is something we’re very, very pleased with.
Microsoft’s management recently introduced Security Copilot, the world’s first generative AI cybersecurity product, and it is seeing high demand
We see high demand for Security Copilot, the industry’s first and most advanced generative AI product, which is now seamlessly integrated with Microsoft 365 Defender. Dozens of organizations, including Bridgewater, Fidelity National Financial and Government of Alberta, have been using Copilot in preview and early feedback has been positive.
Bing users have engaged in over 1.9 billion chats and Bing has a new personalised answers feature, and better support for DALL-E-3 (more than 1.8 billion images have been created with DALL-E-3 to-date)
Bing users have engaged in more than 1.9 billion chats, and Microsoft Edge has now gained share for 10 consecutive quarters. This quarter, we introduced new personalized answers as well as support for DALL-E 3, helping people get more relevant answers and to create incredibly realistic images. More than 1.8 billion images have been created to date.
Bing is now incorporated into Meta’s AI chat experience
We’re also expanding to new end points, bringing Bing to Meta’s AI chat experience in order to provide more up-to-date answers as well as access to real-time search information.
Azure saw higher-than-expected AI consumption
In Azure, as expected, the optimization trends were similar to Q4. Higher-than-expected AI consumption contributed to revenue growth in Azure.
Micosoft’s management is seeing new AI project starts in Azure, and these bring other cloud projects
Given our leadership position, we are seeing complete new project starts, which are AI projects. And as you know, AI projects are not just about AI meters. They have lots of other cloud meters as well. So that sort of gives you one side of what’s happening in terms of enterprise.
Microsoft’s management believes the company has very high operating leverage with AI, since the company is using one model across its entire stack of products, and this operating leverage goes down to the silicon level
Yes, it is true that we have — the approach we have taken is a full-stack approach all the way from whether it’s ChatGPT or Bing chat or all our Copilots all share the same model. So in some sense, one of the things that we do have is very, very high leverage of the one model that we used, which we trained, and then the one model that we are doing inferencing at scale. And that advantage sort of trickles down all the way to both utilization internally, utilization of third parties. And also over time, you can see that sort of stack optimization all the way to the silicon because the abstraction layer to which the developers are riding is much higher up than no-level kernels, if you will. So therefore, I think there is a fundamental approach we took, which was a technical approach of saying we’ll have Copilots and Copilot stack all available. That doesn’t mean we don’t have people doing training for open-source models or proprietary models. We also have a bunch of open-source models. We have a bunch of fine-tuning happening, a bunch of RLHF happening. So there’s all kinds of ways people use it, but the thing is we have scale leverage of one large model that was trained and one large model that’s been used for inference across all our first-party SaaS apps as well as our API in our Azure AI service…
…In addition, what Satya mentioned earlier in a question, and I just want to take every chance to reiterate it, if you have a consistent infrastructure from the platform all the way up through its layers that every capital dollar we spend, if we optimize revenue against it, we will have great leverage because wherever demand shows up in the layers, whether it’s at the SaaS layer, whether it’s at the infrastructure layer, whether it’s for training workloads, we’re able to quickly put our infrastructure to work generating revenue on our BEAM workloads. I mean I should have mentioned all the consumer workloads use the same frame.
Microsoft’s management believes that having the discipline to concentrate Microsoft’s tech stack and capital spend is important because the costs of developing and using AI can run up really quickly
I think, is very important for us to be very disciplined on both I’ll call it our tech stack as well as our capital spend all to be concentrated. The lesson learned from the cloud side is this, we’re not running a conglomerate of different businesses. It’s all one tech stack up and down Microsoft’s portfolio. And that I think is going to be very important because that discipline, given what the spend like — it will look like for this AI transition, any business that’s not disciplined about their capital spend accruing across all their businesses could run into trouble.
Nvidia’s management believes that its chips, together with the Infiniband networking technology, are the reference architecture for AI
NVIDIA HDX with InfiniBand together are essentially the reference architecture for AI supercomputers and data center infrastructures.
Inferencing is now a major workload for Nvidia chips
Inferencing is now a major workload for NVIDIA AI compute.
Nvidia’s management is seeing major consumer internet companies ramping up generative AI deployment, and enterprise software companies starting to
Most major consumer Internet companies are racing to ramp up generative AI deployment. The enterprise wave of AI adoption is now beginning. Enterprise software companies such as Adobe, Databricks, Snowflake and ServiceNow are adding AI copilots and assistance with their pipelines.
Recent US export controls have affected Nvidia’s chip exports to China, Vietnam, and parts of the Middle East
Toward the end of the quarter, the U.S. government announced a new set of export control regulations for China and other markets, including Vietnam and certain countries in the Middle East. These regulations require licenses for the export of a number of our products, including our Hopper and MPIR 100 and 800 series and several others. Our sales to China and other affected destinations derived from products that are now subject to licensing requirements have consistently contributed approximately 20% to 25% of data center revenue over the past few quarters. We expect that our sales to these destinations will decline significantly in the fourth quarter, though we believe will be more than offset by strong growth in other regions.
Many countries are keen to invest in sovereign AI infrastructure, and Nvidia’s management is helping them do so as it is a multi-billion economic opportunity
Many countries are awaiting to the need to invest in sovereign AI infrastructure to support economic growth and industrial innovation. With investments in domestic compute capacity, nations can use their own data to train LLMs and support their local generative AI ecosystem. For example, we are working with India Government and largest tech companies, including Infosys, Reliance and Tata to boost their sovereign AI infrastructure. And French private cloud provider, Scaleway is building a regional AI cloud based on NVIDIA H100, InfiniBand and NVIDIA AI enterprise software to fuel advancement across France and Europe. National investment in compute capacity is a new economic imperative and serving the sovereign AI infrastructure market represents a multibillion-dollar opportunity over the next few years…
…The U.K. government announced it will build 1 of the world’s fastest AI supercomputer called Isambard-AI with almost 5,500 Grace Hopper Super chips. German Supercomputing Center, Elec, also announced that it will build its next-generation AI supercomputer with close to 24,000 Grace Hopper super chips and Quantum 2 InfiniBand, making it the world’s most powerful AI supercomputer with over 90 exaflops of AI performance…
…You’re seeing sovereign AI infrastructures. People countries that now recognize that they have to utilize their own data, keep their own data, keep their own culture, process that data and develop their own AI.
Nvidia has a new chip with inference speeds that are 2x faster than the company’s flagship H100 GPUs (graphics processing units)
We also announced the latest member of the Hopper family, BH 200, which will be the first GPU to offer HBM3E, faster, larger memory to further accelerate generative AI and LLMs. It moves inference speed up to 2x compared to H100 GPUs for running LLM like [indiscernible].
Major cloud computing services providers will soon begin to offer instances for Nvidia’s next-generation GPU, the H200
Compared to the H100, H200 delivers an 18x performance increase for infancy models like GPT-3, allowing customers to move to larger models and with no increase in latency. Amazon Web Services, Google Cloud, Microsoft Azure and Oracle Cloud will be among the first CSPs to offer H200 base instances starting next year.
Nvidia’s management is seeing very strong demand for Infiniband; management believes that Infiniband is critical in the deployment of LLMs (large language models); management believes that the vast majority of large-scale AI factories had standardised on Infiniband because of Infiniband’s vastly superior value proposition compared to Ethernet (data-traffic patterns are very different for AI and for typical hyperscale cloud environments)
Networking now exceeds a $10 billion annualized revenue run rate. Strong growth was driven by exceptional demand for InfiniBand, which grew fivefold year-on-year. InfiniBand is critical to gain the scale and performance needed for training LLMs. Microsoft made this very point last week highlighting that Azure uses over 29,000 miles of InfiniBand cabling, enough to circle the globe…
……The vast majority of the dedicated large-scale AI factories standardized on InfiniBand. And the reason for that is really because of its data rate and not only just the latency, but the way that it moves traffic around the network is really important. The way that you process AI and a multi-tenant hyperscale ethernet environment, the traffic pattern is just radically different. And with InfiniBand and with software-defined networks, we could do congestion control, adaptive routing, performance isolation and noise isolation, not to mention, of course, the data rate and the low latency that — and the very low overhead of InfiniBand that’s a natural part of InfiniBand. .
And so InfiniBand is not so much just a network. It’s also a computing fabric. We put a lot of software-defined capabilities into the fabric, including computation. We do floating point calculations and computation right on the switch and right in the fabric itself. And so that’s the reason why that difference in Ethernet versus InfiniBand where InfiniBand versus Ethernet for AI factories is so dramatic. And the difference is profound and the reason for that is because you’ve just invested in a $2 billion infrastructure for AI factories, a 20%, 25%, 30% difference in overall effectiveness, especially as you scale up is measured in hundreds of millions of dollars of value. And if you were renting that infrastructure over the course of 4 or 5 years, it really adds up. And so InfiniBand’s value proposition is undeniable for AI factories.
Nvidia’s management is expanding the company into Ethernet and Nvidia’s Ethernet technology performs better than traditional offerings; management’s go-to-market strategy for Nvidia’s new Ethernet technology is to collaborate with the company’s large enterprise partners
We are expanding NVIDIA networking into the Ethernet space. Our new Spectrum end-to-end Ethernet offering with technologies, purpose-built for AI will be available in Q1 next year. We support from leading OEMs, including Dell, HP and Lenovo. Spectrum X can achieve 1.6x higher networking performance for AI communication compared to traditional ethernet offerings…
…And our company is — for all of our employees, doesn’t have to be as high performance as the AI factories, we use to train the models. And so we would like the AI to be able to run an Ethernet environment. And so what we’ve done is we invented this new platform that extends Ethernet. It doesn’t replace the Ethernet, it’s 100% compliant with Ethernet, and it’s optimized for east-west traffic, which is where the computing fabric is. It adds to Ethernet with an end-to-end solution with Bluefield as well as our Spectrum switch that allows us to perform some of the capabilities that we have in InfiniBand, not all but some, and we achieved excellent results.
And the way we go to market is we go to market with our large enterprise partners who already offer our computing solution. And so HPL and Lenovo has the NVIDIA AI stack, the NVIDIA enterprise software stack. And now they integrate with BlueField as well as bundle take to market their Spectrum switch and they’ll be able to offer enterprise customers all over the world with their vast sales force and vast network of resellers the — in a fully integrated, if you will, fully optimized at least end-to-end AI solution. And so that’s basically bringing AI to Ethernet for the world’s enterprise.
Nvidia’s management believes that there’s a new class of data centres emerging, and they’ve named them as “AI factories”; these AI factories are being built all across the world
This is the traditional data centers that you were just talking about, where we represent about 1/3 of that. But there’s a new class of data centers. And this new class of data centers, unlike the data centers of the past where you have a lot of applications running used by a great many people that are different tenants that are using the same infrastructure and the data center stores a lot of files. These new data essentials are very few applications if not 1 application used by basically 1 tenant. And it processes data. It trains models and it generates tokens, it generates AI. And we call these new data center AI factories. We’re seeing AI factories being built out everywhere in just about every country.
Nvidia’s management is seeing the appearance of CSPs (cloud services providers) that specialise only in GPUs and processing AI
You’re seeing GTU specialized CSPs cropping up all over the world, and they’re dedicated to doing really 1 thing, which is processing AI.
Nvidia’s management is seeing an AI adoption-wave moving from startups and CSPs to consumer internet companies, and then to enterprise software companies, and then to industrial companies
And so we’re just — we’re seeing the waves of generative AI starting from the start-ups and CSPs moving to consumer Internet companies moving to enterprise software platforms, moving to enterprise companies. And then — and ultimately, 1 of the areas that you guys have seen us spend a lot of energy on has to do with industrial generative AI. This is where NVIDIA AI and NVIDIA Omniverse comes together. And that is a really, really exciting work. And so I think the — we’re at the beginning of a and basically across the board industrial transition to generative AI to accelerated computing. This is going to affect every company, every industry, every country.
Nvidia’s management believes that Nvidia’s AI Enterprise service – where the company helps its customers develop custom AI models that the customers are then free to monetise in whatever manner they deem fit – will become a very large business for Nvidia
Our monetization model is that with each 1 of our partners, they rent a sandbox on DGX Cloud where we work together. They bring their data. They bring their domain expertise. We’ve got our researchers and engineers. We help them build their custom AI. We help them make that custom AI incredible. Then that customer AI becomes theirs, and they deploy it on a run time that is enterprise grade enterprise optimized or outperformance optimized, runs across everything NVIDIA. We have a giant installed base in the cloud on-prem anywhere. And it’s secure, securely patched, constantly patched and optimized and supported. And we call that NVIDIA AI enterprise.
NVIDIA AI Enterprise is $4,500 per GP per year. That’s our business model. Our business model is basically a license. Our customers then would that basic license can build their monetization model on top of. In a lot of ways we’re wholesale, they become retail. They could have a per-subscription license base. They could per instance or they could do per usage. There’s a lot of different ways that they could take to create their own business model, but ours is basically like a software license like an operating system. And so our business model is help you create your custom models, you run those custom models on NVIDIA AI Enterprise. And it’s off to a great start. NVIDIA AI Enterprise is going to be a very large business for us.
PayPal’s management wants to use AI and the data collected from the company’s Rewards program to drive a shopping recommendation engine
For example, our PayPal Cashback Mastercard provides 3% cash back on PayPal purchases as well as cash back on all other purchases. Customers with this card make, on average, 56 more purchases with PayPal in the year after they adopt the product than they did the year before. Over 25 million consumers have used PayPal Rewards in the past 12 months, and we’ve put more than $200 million back in our customers’ pockets with cashback and savings during that time. But even more interesting, through our Rewards product, we have an active database of over 300 million SKUs of inventory from our merchant partners. These data points can help us use AI to power a robust shopping recommendation engine, to provide more relevant rewards and savings back to our customers.
PayPal’s management believes that machine learning and generative AI can be applied to the company’s data to improve fraud protection and better connect merchants and consumers
Our machine learning capabilities combine hundreds of risk and fraud models with dozens of real-time analytics engines and petabytes of payments data to generate insights by learning users’ behaviors, relationships, interests and spending habits. This scale gives us a very unique advantage in the market. Our ability to create meaningful profiles with the help of AI is exceptionally promising. You will see us using our data and the advances in generative AI in responsible ways to further connect our merchants and consumers together in a tight flywheel.
Shopify’s management has integrated Shopify Magic – the company’s suite of free AI features – across its products
At Shopify, we believe AI is for everyone, and its capabilities should be captured and embedded across the entirety of a business. We’ve integrated Shopify Magic, our suite of free AI-enabled features, across our products and workflows.
Shopify Magic can help merchants craft personalised pages and content, and is designed specifically for commerce
Shopify Magic can make the power of Shopify and a merchant’s own data to make it work better for them, whether it’s enabling unique personalized page and content generation like instantly crafting an About Us page in your brand voice and tone or building a custom page to showcase all the sizes available in your latest product collection…
…Now unlike other AI products, the difference with Shopify Magic is it’s designed specifically for commerce. And it’s not necessarily just 1 feature or 1 product. It’s really embedded across Shopify to make these workflows in our products just easier to use. It makes it easier for merchants to run and scale their businesses. And of course, we think it’s going to unlock a ton of possibilities for not just small merchants, but merchants of all sizes. And we’re going to continue to work on that over time. It’s just going to get better and better.
Shopify’s management is using AI internally so that the company can make better decisions and improve its customer support
We ourselves are using AI inside of Shopify to make better decisions, but also for things like — things like our support team using it so that questions like domain reconfiguration, or a new password, or I don’t know what my password is. Those things should not necessarily require high-touch communication. What that does is it means that our support team are able to have much higher-quality conversations and act as business coaches for the merchants on Shopify.
Shopify’s management believes that Shopify is uniquely positioned to harness the power of AI because commerce and the company represent the intersection of humans and technology, and that is the domain of AI
If you kind of think about commerce and Shopify, we kind of interact at the intersection of humans and technology, and that’s exactly what AI is really, really good at. So we think we’re uniquely positioned to harness the power of AI, and the ultimate result of it will be these capabilities for our merchants to grow their businesses.
Shopify has AI-powered language translations for merchants within its software products
This includes things like launching shipping guidance for merchants, navigating them through streamlined privacy guidance, initiating localization experiments across various marketing channels and bringing localization tools and AI-backed language translations to the Shopify App Store.
TSMC’s management sees strong AI-related demand for its chips, but it’s not enough to offset cyclicality in its business
Moving into fourth quarter 2023. While AI-related demand continues to be strong, it is not enough to offset the overall cyclicality of our business. We expect our business in the fourth quarter to be supported by the continued strong ramp of our 3-nanometer technology, partially offset by customers’ continued inventory adjustment.
TSMC;s management is seeing strong customer interest for its N2 technology node because the surge in AI-related demand leads to demand for energy-efficient computing, and TSMC’s technology platform goes beyond geometry-shrink (making transistors smaller), helping with power efficiency
The recent surge in AI-related demand supports our already strong conviction that demand for energy-efficient computing will accelerate in an intelligent and connected world. The value of our technology platform is expanding beyond the scope of geometry shrink alone and increasing toward greater power efficiency. In addition, as process technology complexity increases, the lead time and engagement with customers also start much earlier. As a result, we are observing a strong level of customer interest and engagement at our N2 similar to or higher than N3 at a similar stage from both HPC and smartphone applications.
TSMC’s management is seeing its customers add AI capabilities into smartphones and PCs and expects more of this phenomenon over time
We do see some activities from customers who add AI capability in end devices such as smartphone and PCs, [ so new growth ] engine and AI and PC, whatever. And we certainly hope that this one will add to the course, help TSMC more strengthen under our AI’s business…
…It started right now, and we will expect that the more and more customer will put that AI’s capability into the end devices, into their product.
TSMC’s management is seeing AI-related demand growing stronger and stronger and TSMC has to grow its manufacturing capacity to support this
The AI demand continues to grow stronger and stronger. So from TSMC’s point of view, now we have about — we have a capacity limitation to support them — to support the demand. We are working hard to increase the capacity to meet their demand, that’s for one thing.
TSMC’s management believes that any kind of AI-related chip will require leading edge chip technology and this is where TSMC excels
Whether customer developed the CPU, GPU, AI accelerator or ASIC for all the type for AI applications, the commonality is that they all require usage of leading-edge technology with stable yield delivery to support larger die size and a strong foundry design ecosystem. All of those are TSMC’s strengths. So we are able to address and capture a major portion of the market in terms of a semiconductor component in AI.
Tencent’s management is increasing the company’s investments in its AI models and management wants to use AI for the company’s own benefit as well as that of society and its customers
We are increasing investment in our AI models, providing new features to our products and enhancing our targeting capabilities for both content and advertising. We aspire to position our leading AI capability, not only as a growth multiplier for ourselves, but also as a value provider to our enterprise customers and the society at large.
Tencent’s management recently upgraded the size and capabilities of the company’s foundational model – Tencent Hunyuan – which is now available to customers on a limited basis and deployed in some of Tencent’s cloud services
For cloud, we upgrade the size and capabilities of our proprietary foundation model, Tencent Hunyuan. We are making Hunyuan available on a limited basis to the public and to customers and deploying QiDian in Tencent Meeting and Tencent Docs…
…We have upgraded our proprietary foundation model, Tencent Hunyuan. We have made Tencent Hunyuan bot initially available to a smaller but expanding number of users via a mini program. Hunyuan is also now powering meeting summarization in Tencent Meeting and content generation in Tencent Docs. And externally, we’re enabling enterprise customers to utilize our large language model via APIs or model as a Service solutions in our cloud in functions such as coding, data analysis and customer service automation.
Tencent’s management believes that Tencent is one of China’s AI leaders with the development of Hunyuan
In terms of Hunyuan and the overall AI strategy, I would say we have been pretty far along in terms of building up Hunyuan, and we feel that we are one of the leaders within China, and we are also continuously increasing the size of the model and preparing for the next generation of our Hunyuan model, which is going to be a mixture of experts architecture, which we believe will further improve the performance of our Hunyuan model. And by building up Hunyuan, we actually have really build up our capability in general AI across the board. Because Hunyuan, the transformer-based model include — involve the handling of a large amount of data, large amount of training data, large size of computing cluster and a very dedicated fine-tuning process in terms of improving the AI performance.
Tencent’s management is using AI to improve the company’s advertising offerings, in areas such as ad targeting, attribution accuracy, and the generation of advertising visuals – management sees this as evidence that Tencent’s AI investments are already generating tangible results
We have expanded our AI models with more parameters to increase their ad targeting and attribution accuracy contributing to our ad revenue growth. We’re also starting to provide generative AI tools to advertiser partners, which enables them to dynamically generate ad visuals based on text fronts and to optimize ad sizes for different inventories, which should help advertisers create more appealing advertisements with higher click-through rates boosting their transactions in our revenue…
…And the general AI capability is actually helping us quite a bit in terms of the targeting technology related to advertising and our content provisioning service. So in short video by improving our AI capability, we can actually ramp up our video accounts at the faster clip. And in terms of the advertising business by increasing the targeting capability, we are actually increasing our ad revenue and by delivering better results to the — to our customers. So they are generating — so our AI capabilities is generating tangible result at this point in time.
Tencent’s management wants to build an AI-powered consumer-facing smart agent down the road, but they are wary about the costs of inference
And we also feel that further in the future, when there’s actually a consumer-facing product that is more like a smart agent for people right now, that is further down the road, but it actually carries quite a bit of room for imagination…
…Now in terms of the Hunyuan and in the future, the potential of an AI assistant, I think it’s fair to say it’s still in a very, very early stage of concept design. So definitely not at the stage of product design yet and definitely not at the stage of thinking about monetization yet. But of course, right, if you look at any of these generative AI technology at this point in time, inference cost is a real variable cost, which needs to be considered in the entire equation. And that, to some extent, add to the challenge of the product design, too. So I would say, at this point in time, it’s actually very early stage. There is a promise and imaginary room for opportunity for the future.
Tencent’s management believes that the company has sufficient amount of chips for the company’s AI-related development work for a couple more generations; the US’s recent semiconductor bans will not affect the development of Tencent’s AI models, but it could affect Tencent’s ability to rent out these chips through Tencent Cloud
Now in terms of the chip situation, right now, we actually have 1 of the largest inventory of of AI chips in China among all the players. And one of the key things that we have done was actually we were the first to put in order for H800, and that allow us to have a pretty good inventory of H800 chips. So we have enough chips to continue our development of Hunyuan for at least a couple more generations. And the ban does not really affect the development of Hunyuan and our AI capability in the near future. Going forward, we feel that the shipment does actually affect our ability to resell these AI chips to — through our cloud services. So that’s one area that may be impacted.
Tencent’s management wants to explore the use of lower-performance chips for AI inference purposes and they are also exploring domestic suppliers of chips
Going forward, we feel that the shipment does actually affect our ability to resell these AI chips to — through our cloud services. So that’s one area that may be impacted. And going forward, we will have to figure out ways to make our — the usage of our AI chips more efficient. We’ll try to see whether we can offload a lot of the inference capability to lower-performance chips so that we can retain the majority of our high-performance AI chips for training purpose. And we also try to look for domestic stores for these training chips.
Tencent’s management believes that AI can bring significant improvement to a digital ad’s current average click-through rate of 1%
Today, a typical click-through rate might be around 1%. As you deploy large language models, then you can make more use of the thousands of discrete data points that we have potentially for targeting and bring them to bear and turn them into reality. And you can get pretty substantial uplifts in click-through rate and therefore, in revenue, which is what the big U.S. social networks are now starting to see.
Tesla vehicles have now driven over 0.5 billion miles with FSD (Full Self Driving) Beta and the mileage is growing
Regarding Autopilot and AI, our vehicle has now driven over 0.5 billion miles with FSD Beta, full self-driving beta, and that number is growing rapidly.
Tesla’s management sees significant promise with FSD v.12
We’re also seeing significant promise with FSD version 12. This is the end-to-end AI where it’s a photon count in, controls out or really you can think of it as there’s just a large bit stream coming in and a tiny bit stream going out, compressing reality into a very small set of outputs, which is actually kind of how humans work. The vast majority of human data input is optics, from our eyes. And so we are, like the car, photons in, controls out with neural nets, just neural nets, in the middle. It’s really interesting to think about that.
Tesla recently completed building a 10,000 GPU cluster of Nvidia’s H100s chips and has brought the cluster into operation faster than anyone has done (the H100s will help with the development of Tesla’s full self driving efforts)
We recently completed a 10,000th GPU cluster of H100s. We think probably bringing it into operation faster than anyone’s ever brought that much compute per unit time into production since training is the fundamental limiting factor on progress with full self-driving and vehicle autonomy.
Tesla’s management believes that AI is a game changer and wants the company to continue to invest in AI
We will continue to invest significantly in AI development as this is really the massive game changer, and I mean, success in this regard in the long term, I think has the potential to make Tesla the most valuable company in the world by far.
Tesla’s management believes that the company’s AI team is the best in the world
The Tesla AI team is, I think, one of the world’s best, and I think it is actually by far the world’s best when it comes to real-world AI. But I’ll say that again: Tesla has the best real-world AI team on earth, period, and it’s getting better.
Tesla’s management is very excited about the company’s progress with autonomous driving and it is already driving them around with no human-intervention
I guess, I am very excited about our progress with autonomy. The end-to-end, nothing but net, self-driving software is amazing. I — drives me around Austin with no interventions. So it’s clearly the right move. So it’s really pretty amazing.
Tesla’s management believes that the company’s work in developing autonomous driving can also be applied to Optimus (the company’s autonomous robots)
And obviously, that same software and approach will enable Optimus to do useful things and enable Optimus to learn how to do things simply by looking. So extremely exciting in the long term.
Tesla’s management believes that Optimus will have a huge positive economic impact on the world and that Tesla is at the forefront of developing autonomous robots; Tesla’s management is aware of the potential dangers to humankind that an autonomous robot such as Optimus can pose, so they are designing the robot carefully
As I’ve mentioned before, given that the economic output is the number of people times productivity, if you no longer have a constraint on people, effectively, you’ve got a humanoid robot that can do as much as you’d like, your economy is twice the infinite or infinite for all intents and purposes. So I don’t think anyone is going to do it better than Tesla, not by a long shot. Boston Dynamics is impressive, but their robot lacks the brain. They’re like the Wizard of Oz or whatever. Yes, lacks the brain. And then you also need to be able to design the humanoid robot in such a way that it can be mass manufactured. And then at some point, the robots will manufacture the robots.
And obviously, we need to make sure that it’s a good place for humans in that future. We do not create some variance of the Terminator outcome. So we’re going to put a lot of effort into localized control of the humanoid robot. So basically, anyone will be able to shut it off locally, and you can’t change that even if you put — like a software update, you can’t change that. It has to be hard-coded.
Tesla’s management believes that Mercedes can easily accept legal liability for any FSD-failures because Mercedes’ FSD is very limited whereas Tesla’s FSD has far less limitations
[Question] Mercedes is accepting legal liability for when it’s Level 3 autonomous driving system drive pilot is active. Is Tesla planning to accept legal liability for FSD? And if so, when?
[Answer] I mean I think it’s important to remember for everyone that Mercedes’ system is limited to roads in Nevada and some certain cities in California, doesn’t work in the snow or the fog. It must have a [indiscernible] car in plains, only 40 miles per hour. Our system is meant to be holistic and drive in any conditions, so we obviously have a much more capable approach. But with those kind of limitations, it’s really not very useful.
Tesla’s management believes that technological progress building on technological progress is what will eventually lead to full self driving
I would characterize our progress in real world AI as a series of stacked log curves. I think that’s also true in other parts of AI, like [ LOMs ] and whatnot, a series of stacked log curves. Each log curve gets higher than the last one. So if we keep stacking them, we keep stacking logs, eventually, we get to FSD.
The Trade Desk’s management believes that AI will change the world, but not everyone working on AI is delivering meaningful impact
AI has immense promise. It will change the world again. But not everyone talking about AI is delivering something real or impactful.
The Trade Desk’s management is not focusing the company’s AI-related investments on LLMs (large language models) – instead, they are investing in deep-learning models to improve bidding, pricing, value, and ad relevance for Trade Desk’s services
Large Language Models (the basis of ChatGPT) aren’t the highest priority places for us to make our investments in AI right now. Deep learning models pointed at bidding, pricing, value, and ad relevance are perfect places for us to concentrate our investments in AI—all four categories have private betas and some of the best engineers in the world pointed at these opportunities.
The Trade Desk’s management believes that they are many areas to infuse AI into the digital advertising dataset that the company holds
Second is the innovation coming from AI and the many, many opportunities we have ahead of us to find places to inject AI into what may be the most rich and underappreciated data asset on the Internet, which we have here at The Trade Desk.
The Trade Desk’s management believes that traders in the digital advertising industry will not lose their jobs to AI, but they might lose their jobs to traders who know how to work with AI
Traders know that their jobs are not going to be taken away by AI. But instead, they have to compete with each other. So their job could be taken away from a trader who knows how to use AI really well until all of them are looking at ways to use the tools that are fueled by AI that were provided, where AI is essentially doing 1 or 2 things. It’s either doing the math for them, if you will, of course, with very advanced learning models or, in other cases, it’s actually their copilot.
Old Navy achieved a 70% reduction in cost to reach each unique household using The Trade Desk’s AI, Koa
A great example of an advertiser pioneering new approaches to TV advertising with a focus on live sports is Old Navy… But as Old Navy quickly found out, programmatic guaranteed has limitations. Programmatic guaranteed, or PG, does not allow Old Navy to get the full value of programmatic such as frequency management, audience targeting and the ability to layer on their first-party data. So they took the next step in the form of decision biddable buying within the private marketplace and focused on live sports inventory. CTV live sports advertising was appealing because it offered an opportunity to expose their brand against very high premium content that might be more restrictive and expensive in a traditional linear environment. They were able to use Koa, The Trade Desk’s AI, to optimize pacing and frequency management across the highest-performing inventory. As a result, they saw a 70% reduction in the cost to reach each unique household versus their programmatic guaranteed performance.
Users of Wix’s Wix Studio product are enjoying its AI features
Users particularly [indiscernible] Studio responsive AI technology that simplify high-touch and time-sensitive tasks such as ensuring consistent design across web pages on different screen sizes. They are also enjoying the AI code assistant inside the new Wix IDE [integrated development environment], which allowed them to write clinic code and detect errors easily.
Wix recently released new AI products: (1) an SEO tool powered by AI called AI Meta Tags Creator, and (2) AI Chat Experience for Business, which allows new users to chat with an AI who will walk them through the Wix onboarding process; AI Chat Experience for Business is in its early days, but it has already driven a positive impact on Wix’s conversion and revenue
Earlier this week, we released our latest AI products. The first was AI Meta Tags Creator, a groundbreaking SEO tool powered by AI and our first AI-powered feature within our collection of SEO tools. Both self creators looking to generate SEO-friendly tags for each of their pages and professionals looking to enhance their efficiency and make real-time adjustments will benefit from this product. The second was our Conversational AI Chat Experience for Business. This feature, which is now live, paves the way to accelerate onboarding using AI in order to get businesses online more quickly and efficiently. These new tools continue to demonstrate our leadership in utilizing AI to help users of all types to succeed online…
…Avishai spoke about the AI chat experience for business and its early weeks — and in its early weeks, we have already seen its positive impact on conversion and revenue.
Wix’s management expects Wix’s AI products to drive higher conversion, monetisation, and retention in the company’s Self Creators business
Compounding Partners growth is complemented by re-accelerating growth in our stable and profitable Self Creators business, which we saw once again this quarter. We expect our market-leading product innovation as well as our powerful AI products and technology to drive higher conversion, monetization and retention as we maintain our leadership position in the website building space.
Wix’s management believes that Wix’s AI products are helping to improve conversion because the new AI tools help to generate content for users, which reduces the inertia to create a website
I believe your second question was in regards to what kind of effect we are seeing from different AI products that we are launching, and mostly in regards to improvement in conversion. And we do actually see an improvement in conversion, which is probably the most important KPI by which we measure our success in deploying new products. The reason for that is that with AI, we are able to ask the user better questions and to understand in a smarter way, why is that the user is trying to achieve. From that, we are able to generate a better starting point for their business on top of Wix. And that is not just the skeleton, we are also able to fill in a lot of information, a lot of the content that the user would normally have to fill in manually. The result is that the amount of effort and knowledge that you need to create a website and for your business on Wix is dramatically reduced. And from that, we are able to see very good results in terms of improvement of conversion.
The use of AI tools internally has helped to improve Wix’s margins
So we saw this year a tremendous improvement in margins — in gross margin. And it came mostly from 2 places. The first one is a lot of improvements and savings that we have with our infrastructure, most of you know the hosting activity. So we had a lot of savings over there, but also about our core organization, for example, benefiting from all kind of AI tools that enable us to be more efficient.
Wix’s management believes that the company’s AI features help users with website-creation when it would normally take specialists to do so
And then because of the power of the AI tools, you can create very strong, very professional websites because the AI will continue and finish for you the thing that would normally require to specialize in different variations of web designs.
Zoom AI Companion, which helps create call summaries, is included in Zoom’s paid plans at no additional costs to customers, and more than 220,000 accounts have enabled it, with 2.8 million meeting summaries created to-date
We also showcased newly-released innovations like Zoom AI Companion, as well as Zoom AI Expert Assist and a Quality Management for the Contact Center. Zoom AI Companion is especially noteworthy for being included at no additional cost to our paid plans, and has fared tremendously well with over 220,000 accounts enabling it and 2.8 million meeting summaries created as of today.
Zoom’s management believes that Zoom AI Companion’s meeting-summary feature is really accurate and really fast; management attributes the good performance to the company’s use of multiple AI models within Zoom AI Companion
I think we are very, very proud of our team’s progress since it launched the Zoom AI Companion, as I mentioned earlier, right, a lot of accounts enabled that. Remember, this is no additional cost to [ outpay ] the customer. A lot of features.One feature of that is like take a meeting summary, for example. Amazingly, it’s very accurate and it really save the meeting host a lot of time. And also, our federated AI approach really contributed to that success because we do not count on a single AI model, and in terms of latency, accuracy, and also the response, the speed and so on and so forth, I think, it really helped our AI Companion.
Free users of Zoom are unable to access Zoom AI Companion
For sure, for free users, they do not — they cannot enjoy this AI Companion, for sure, it’s a [ data health ] for those who free to approve for online upgrade. So anyway, so we keep innovating on AI Companion. We have high confidence. That’s a true differentiation compared to any other AI features, functionalities offered by some of our competitors.
Zoom’s management thinks that Zoom’s AI features for customers will be a key differentiator and a retention tool
But I think what Eric was just mentioning about AI is probably really going to be a key differentiator and a retention — retention tool in the future, because as a reminder, all of the AI Companion features come included for our free — sorry, for our paid users. So we’re seeing it not only help with conversion, but we really believe that for the long term, it will help with retention as well.
Zoom’s management believes that Zoom’s AI features will help to reaccelerate Zoom’s net dollar expansion rate for enterprise customers
[Question] You’re showing stabilization here on some of the major metrics, the Enterprise expansion metric took a step down to 105%. And so just wondering what it takes for that metric to similarly show stabilization as given like in Q1 renewal cohort and kind of walking through that. Anything on the product side for us to consider or just any other commentary there is helpful.
[Answer] Well, as a reminder, it’s a trailing 12-month metric. So as we’ve worsely seen our growth rates come down this year that’s following behind it. But absolutely, we believe that AI Companion in general as well as the success that we are seeing in Zoom Phone, in Zoom Contact Center, Zoom Virtual Agent, all of those will be key contributors to seeing that metric start to reaccelerate again as we see our growth rate starting to reaccelerate as well.
Zoom’s management thinks tjat Zoom’s gross margin could decline – but only slightly – due to the AI features in Zoom’s products being given away for free at the moment
[Question] As I look at gross margins, how sustainable is it keeping at these levels? I know AI Companion is being given away from as part of the package, I guess, prepaid users. But if you think about the cost to run these models, the margin profile of Contact Center and Phone. How durable is it to kind of sustain these levels?
[Answer] But we do expect there’s going to be some impact on gross margins. I mean we — I don’t think it’s going to be significant because the team will continue to operate in the very efficient manner that they do and run our co-los [co-locateds] that way, but we do expect there’s going to be some impact to our gross margin as we move forward.
Zoom’s management wants to leverage AI Companion across the entire Zoom platform
So again, it’s a lot of other features as well. And like for me, I also use our — the client, [indiscernible] client, connect and other services you can, right? You can have you compose e-mail as well, right? It’s a lot of features, right? And down the road awareness for the Whiteboard with AI Companion as well. Almost every service entire platform, we’re going to lever the AI Companion. So and a lot of features and the AI Companion.
Note: An earlier version of this article was published at The Good Investors, a personal blog run by our friends.
Are we really ready to live in a world with AI that could potentially take over our jobs? Check out our latest Special Free Report on this fascinating topic. We cover the latest developments in AI and how they could impact your life and investments. Click here to download a copy now.
Follow us on Facebook and Telegram for the latest investing news and analyses!
Disclosure: Ser Jing has an interest in the shares of Adobe, Alphabet, Amazon, Apple, Datadog, Etsy, Fiverr, Mastercard, MercadoLibre, Meta Platforms, Microsoft, PayPal, Shopify, TSMC, Tencent, Tesla, The Trade Desk, Wix, and Zoom.