Last month, I published The Latest Thoughts From American Technology Companies On AI (2024 Q2). In it, I shared commentary in earnings conference calls for the second quarter of 2024, 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. 

A few more technology companies I’m watching hosted earnings conference calls for 2024’s first quarter after I prepared the article. The leaders of these companies also had insights on AI that I think would be useful to share. This is an ongoing series. For the older commentary:

• 2023 Q1 – here and here
• 2023 Q2 – here and here
• 2023 Q3 – here and here
• 2023 Q4 – here and here
• 2024 Q1 – here and here

Here they are, in no particular order:

Adobe (NASDAQ: ADBE)

Adobe’s management believes that Adobe’s approach to AI is highly differentiated; the greatest differentiation is at the interface layer, as Adobe is able to rapidly integrate AI across its product portfolio and allow users to realise value

Adobe’s customer-centric approach to AI is highly differentiated across data, models and interfaces…

…Our greatest differentiation comes at the interface layer with our ability to rapidly integrate AI across our industry-leading product portfolio, making it easy for customers of all sizes to adopt and realize value from AI. 

Adobe’s Firefly models are trained on data that allow outputs to be commercially safe and management thinks this feature of being commercially safe is really important to enterprises; Adobe now has Firefly models for imaging, vector, design, and video; Firefly is in a wide range of Adobe products; Firefly has powered more than 12 billion generations since its launch in March 2023 (was 9 billion in 2024 Q1); management’s strategy is to build Firefly models into more streamlined and precise workflows within Adobe’s products; Adobe has Firefly Service APIs for organisations to generate content at scale, and the API calls tripled quarter-on-quarter; Firefly Service APIs are gaining real traction

We train our Firefly models on data that allows us to offer customers a solution designed to be commercially safe. We have now released Firefly models for imaging, vector and design and just previewed a new Firefly video model…

… Firefly-powered features in Adobe Photoshop, Illustrator, Lightroom, and Premier Pro help creators expand upon their natural creativity and accelerate productivity. Adobe Express is a quick and easy create anything application, unlocking creative expression for millions of users. Acrobat AI Assistant helps extract greater value from PDF documents. Adobe Experience Platform AI Assistant empowers brands to automate workflows and generate new audiences and journeys. Adobe GenStudio brings together content and data, integrating high-velocity creative expression with the enterprise activation needed to deliver personalization at scale…

…We have now surpassed 12 billion Firefly-powered generations across Adobe tools…

… Our strategy is to build technology that will create more streamlined and precise workflows within our tools through features like text-to-template in Express, Generative Fill in Photoshop, Generative Recolor in Illustrator, Generative Remove in Lightroom and the upcoming Generative Extend for Video and Premier Pro. We’re exposing the power of our creative tools and the magic of generative AI through Firefly Service APIs so organizations can generate and assemble content at scale…

…The introduction of the new Firefly video model earlier this week at IBC is another important milestone in our journey. Our video model, like the other models in the Firefly family, is built to be commercially safe with fine-grain control and application integration at its core. This will empower editors to realize their creative vision more productively in our video products, including Premier Pro…

…Strong demand for Firefly Services, which provide APIs, tools and services for content generation, editing and assembly, empowering organizations to automate content production while maintaining quality and control. Total API calls tripled quarter over quarter…

…Firefly Services, which is you can think of that also as a consumption model where we have that, it’s off to a really good start. Our ability to give enterprises the ability to automate content, create custom models within enterprises, we’re seeing real traction because it’s a differentiated solution and that it’s designed to be commercially safe…

…One other thing I’d just emphasize there is that the commercial safety is so important to businesses of all sizes, frankly, and that is something that we feel very, very differentiated.

Adobe released significant advancements in AI Assistant across Adobe Acrobat and Reader in 2024 Q2 (FY2024 Q3) and saw 70% sequential growth in AI interactions in AI Assistant; the advancements in AI Assistant include content creation capabilities; Tata Consultancy Services used AI Assistant in Adobe Acrobat to create event summaries of hours of conference videos in minutes; management intends to actively promote subscription plans for Adobe Acrobat and Reader that include generative AI capabilities

For decades, PDF has been the de facto standard for storing unstructured data, resulting in the creation and sharing of trillions of PDFs. The introduction of AI Assistant across Adobe Acrobat and Reader has transformed the way people interact with and extract value from these documents. In Q3, we released significant advancements, including the ability to have conversations across multiple documents and support for different document formats, saving users valuable time and providing important insights. We are thrilled to see this value translate into AI Assistant usage with over 70% quarter-over-quarter growth in AI interactions. 

In addition to consumption, we’re focused on leveraging generative AI to expand content creation in Adobe Acrobat. We’ve integrated Adobe Firefly Image Generation into our edit PDF workflows. We’ve optimized AI Assistant in Acrobat to generate content fit for presentations, e-mails and other forms of communication, and we’re laying the groundwork for richer content creation, including the generation of Adobe Express projects.

The application of this technology across verticals and industries is virtually limitless. Tata Consultancy Services recently used Adobe Premiere Pro to transcribe hours of conference videos and then used AI Assistant in Acrobat to create digestible event summaries in minutes. This allowed them to distribute newsletters on session content to attendees in real-time.

We’re excited to leverage generative AI to add value to content creation and consumption in Acrobat and Reader in the months ahead. Given the early adoption of AI Assistant, we intend to actively promote subscription plans that include generative AI capabilities over legacy perpetual plans that do not.

Adobe GenStudio is integrated across Experience Cloud and Creative Cloud and helps marketers quickly plan, create, store, deliver, and measure marketing content; Vanguard used Adobe GenStudio to increase quality engagement with investors by 176% through one-to-one personalisation, and to enjoy millions in savings

Customers are embracing the opportunity to address their content supply chain challenges with Adobe GenStudio. With native integrations across Experience Cloud and Creative Cloud, GenStudio empowers marketers to quickly plan, create, store, deliver, and measure marketing content and drive greater efficiency in their organizations. Financial services leader Vanguard is creating an integrated content supply chain to serve the strategic goal of deepening their relationships with a broad range of investors. Leveraging the GenStudio solution, Vanguard was able to increase quality engagement by 176% by focusing on one-to-one personalization and to realize millions in savings by improving content velocity and resource allocation with an end-to-end content creation workflow.

Adobe’s management has been very consistent over the past 1-1.5 years in how they have approached AI, and that is, Adobe would be developing a broad set of models for the creative community, and the models would be highly differentiated based on quality, commercial safety, and integrability into Adobe’s product portfolio

I think we’ve been incredibly consistent with what we’ve said, dating back 1 year, 1.5 years ago, where we talked about the fact that we were going to develop the broadest set of models for the creative community. And we were going to differentiate the models based on quality, commercial safety, integratability into our tools and controllability. And as you’ve seen very methodically over the last 18 months, we continue to bring more and more of that innovation to life. And that fundamentally is working as we’ve now started to integrate it much more actively into our base. If you look at it with photography, we now have in our tool, Generative Remove, we have AI-assisted edits in design, we have Generative Pattern, Generative Fill Shape. We have, in Photoshop, we have Gen Remove. We also have Gen Fill, and I can continue on with all the generations, but we’ve also now started to integrate it in Firefly Services for what we’re enabling enterprises to be able to access and use in terms of batch work and through APIs.

Adobe’s management is seeing the accelerated use and consumption of generative AI credits in Adobe’s products play out the way they expected it to; total consumption credits are going up with the introduction of each new generative AI capability 

If you look at sort of how that’s played out, as we talked about, we’re seeing accelerated use and generative credits being consumed because of that deeper integration into all of our tools, and that is playing out as we expected…

…And we do see with every subsequent capability we integrate into the tool, total credits consumed going up. 

Adobe’s management is seeing Adobe enjoying indirect monetisation from the AI features of its products, such as (1) the products having more value and pricing, (2) users being retained better when they use generative AI features, and (3) higher conversion of users when they try out Adobe products

When you look at then how that converts to monetization, first and foremost, we’ve integrated it a lot of that value into our core products with more value and more pricing. We’re also seeing that when people use these generative features, they retain better. We’re also seeing that when people come to Adobe to try our Creative Cloud applications or Express application, they’re able to convert better. And so there are all these ancillary implied benefits that we’re getting. 

For direct monetisation of the AI features in Adobe’s products, management is thinking of (1) instituting caps on generative AI credit consumption, (2) having AI plans with different AI capabilities; but direct monetisation is currently still not the key focus that management has, because they want to focus on proliferation and usage of generative AI across the user base

In terms of direct monetization, what we’ve said in the past is that the current model is around generative credits, which is I think where you’re going with this. And we do see with every subsequent capability we integrate into the tool, total credits consumed going up. Now what we are trying to do as we go forward, we haven’t started instituting the caps yet. And part of this is, as we’ve said all along, we want to really focus our attention on proliferation and usage across our base. We see a lot of users excited about it. It’s some of the most actively used features that we’ve ever released. And we want to avoid the generation anxiety that people feel. But we’re watching very closely as the economy of generative credits evolves, and we’re going to look at instituting those caps at some point when we feel the time is right and/or we’re also looking at other alternative models. What we did with Acrobat AI Assistant has proven to be very effective. And so we’re also considering other opportunities like having standard CC plans that have a core set of generative capabilities but also having premium API — sorry, premium AI plans that will include things more like video and other things.

Adobe’s management thinks Adobe’s generative AI video models are already pretty capable, but they are going to get better over time; management thinks that the real value of generative AI video models is not in their ability to create a video through a description the user gives, but in their ability to extend the video

I don’t know if you had a chance to see some of the videos we put out there integrated directly into premier, also text to video, images to video, more controllability. We have also the ability now to generate not just themes with humans and dogs and organic animals, but all these like overlays and things that creative professionals actually want to work with. And so we’re very excited about the set of things that they can get out of the box that get going. And human faces and things will just continue to get better…

…I spend a couple of hours with our video team. They have just absolutely hit it out of the park. I mean, the work that they have done, which is leveraging the image models with video, and again, I think to David’s point, the integration with Premier, that’s where we’ve always said, it’s the integration of the model and the application that differentiates it. I think when other models first came out, people were like, “Wow, you can describe it.” That’s just such a small part of where the value is. And the real value is, you have a video, you want to extend it. It’s a game changer in terms of what we can do. So really excited about the stuff that we’re doing in video. 

MongoDB (NASDAQ: MDB)

MongoDB’s management sees AI as a longer-term opportunity for MongoDB; management is seeing companies largely still experimenting with AI applications currently; management thinks inference workloads will come, but monetisation of AI apps will take time

AI continues to be an additional long-term opportunity for our business. At the start of the fiscal year, we told you that we didn’t expect AI to be a meaningful tailwind for our business in fiscal year 2025, which has proven accurate. Based on recent peer commentary, it seems that the industry now mostly agrees with this view. Companies are currently focusing their spending on the infrastructure layer of AI and are still largely experimenting with AI applications. Inference workloads will come and should benefit MongoDB greatly in the long run, but we are still very early when the monetization of AI apps will take time. AI demand is a question of when, not if.

MongoDB’s management has been talking to customers and they think MongoDB is the ideal database for AI apps for five reasons: (1) AI workloads involve a wide variety of data types and MongoDB’s document-model database is meant to handle this variety well, thus providing a well-rounded one-stop solution, (2) MongoDB’s database is high-performance and scalable, and allows AI workloads to utilise real-time operational data, (3) MongoDB’s database is integrated with leading app development frameworks and AI platforms, (4) MongoDB’s database has enterprise-grade security and compliance features, and (5) MongoDB’s database can be run anywhere on the customer’s choice; management feels very good about MongoDB’s positioning for AI

Our discussions with customers and partners give us increasing conviction that we are the ideal data layer for AI apps for a number of key reasons.

First, more than any other type of modern workload, AI-driven workloads require the underlying database to be capable of processing queries against rich and complex data structures quickly and efficiently. Our flexible document model is uniquely positioned to help customers build sophisticated AI applications because it is designed to handle different data types, your source data, vector data, metadata and generated data right alongside your live operational data, outdating the need for multiple database systems and complex back-end architectures.

Second, MongoDB offers a high performance and scalable architecture. As the latency of LLMs improve, the value of using real-time operational data for AI apps will become even more important.

Third, we are seamlessly integrated with leading app development frameworks and AI platforms, enabling developers to incorporate MongoDB into their existing workflows while having the flexibility to choose the LLM and other specific tools that best suit their needs.

Fourth, we meet or exceed the security and compliance requirements expected from an enterprise database, including enterprise-grade encryption, authorization and auditability.

Lastly, customers can run MongoDB anywhere, on-premise or as a fully managed service in 1 of the 118 global cloud regions across 3 hyperscalers giving them the flexibility to run workloads to meet — to best meet their application use cases and business needs…

… As the performance of these LLMs and latency of these LLMs increase, accessing real-time data becomes really important like, say, you’re calling and talking to a customer support chatbot, that you want that chatbot to have up-to-date information about that customer so that they can provide the most relevant and accurate information possible…

…I think it’s a quickly evolving space, but we feel very good about our positioning for AI, even though it’s still very early days.

MongoDB’s management sees 3 ways AI can accelerate MongoDB’s business over time: (1) AI will drive the cost of building applications, as all past platform shifts have done, thus leading to more apps and higher demand for databases, (2) MongoDB can be the database of choice for developers building AI applications (see Point 9 on MongoDB’s new AI Applications Program), and (3) MongoDB can help customers modernise their application estate (see Point 10 for more on this opportunity)

We see 3 main opportunities where we believe AI will accelerate our business over time. The first is that the cost of building applications in the world of AI will come down as we’ve seen with every previous platform shift, creating more applications and more data requiring more databases. The second opportunity is for us to be the database of choice for customers building greenfield AI applications…

…The third opportunity is to help customers modernize their legacy application estate. 

MongoDB’s management made the MongoDB AI Applications Program (MAAP) generally available in July 2024; MAAP brings the cloud computing hyperscalers and prominent AI model-building startups into one ecosystem to reduce the complexity and difficulty for MongoDB’s customers when they build new AI applications 

While we see that there’s tremendous amount of interest in and planning for new AI-powered applications, the complexity and fast-moving nature of the AI ecosystem slows customers down. That’s why we launched the MongoDB AI Applications Program, or MAAP, which became generally available to customers last month. MAAP brings together a unique ecosystem, including the 3 major cloud providers, AWS, Azure and GCP as well as Accenture and AI pioneers like Anthropic and Cohere. MAAP offers customers reference architectures and end-to-end technology stack that includes prebuilt integrations, professional services and a unified support system to help customers quickly build and deploy AI applications.

Modernising legacy application estates is a big opportunity, as most of the $80 billion database market is still in legacy relational databases; MongoDB has the Relational Migrator product to help customers migrate from legacy relational databases to the company’s document-model database; management thinks AI can significantly improve the process of modernising legacy applications by helping with understanding legacy code and rewriting them as modern versions; MongoDB launched a few pilots with customers earlier in 2024 to modernise their legacy applications with the help of AI and the results are exciting; the CIO (Chief Information Officer) of an insurance company in the pilots said the modernisation process was the first tangible return he had seen in his AI investments; management thinks it will take time for the modernisation program to contribute meaningful revenue to MongoDB, but they are excited 

Most of the existing $80-billion-plus database industry is built on dated relational architecture. Modernizing legacy applications has always been part of our business, and we have taken steps over the years to simplify and demystify this complex process through partnerships, education and most recently, our Relational Migrator product. AI offers a potential step function improvement, lowering the cost and reducing their time and risk to modernize legacy applications…

…Earlier this year, we launched several pilots with our customers where we work with them to modernize mission-critical applications, leveraging both AI tooling and services. The early results from these pilots are very exciting as our customers are experiencing significant reductions in time and cost of modernization. In particular, we have seen dramatic improvements in time and cost to rewrite application code and generate test suites. We see increasing interest from customers that want to modernize their legacy application estate, including large enterprise customers. As a CIO of one of the world’s largest insurance companies said about our pilot, this is the first tangible return he’s seen on his AI investments. While it’s still early days and generating meaningful revenue from this program will take time, we are excited about the results of our pilots and the growing pipeline of customers eager to modernize their legacy estate…

…Since day one, since our IPO, we’ve been getting customers to migrate off relational to MongoDB. But one of the biggest friction points has been that while it’s easy to move the data, you can map the schema from a relational schema to a document schema and you can automate that, the biggest stumbling block is that the customer has to or some third party has to rewrite the application, which, by definition, creates more costs, more time and in some cases, more risk especially for older apps, where the development teams who built those apps no longer exist. So what’s been compelling about AI is that AI has finally created a shortcut to overcome that big hurdle. And so essentially, you can start basically diagnosing the code, understand the code, recreate a modern version of that code and generate test suites to make sure the new code performs like the old code. So that definitely gets people’s interest because now, all of a sudden, what may take years or multiyears, you can do in a lot less time. And the pilots that we have done, the time and cost savings have been very, very compelling.

That being said, we’re in the very early days. There’s a lot of interest. We have a growing pipeline of customers across, frankly, all parts of the world from North America to EMEA and even the Pac Rim. And so we’re quite excited about the opportunity. But again, I would say it’s very early days.

Delivery Hero, a leading local delivery platform, is using MongoDB Atlas Vector Search to provide AI-powered hypersonalised results to users; Delivery Hero found that MongoDB Atlas Vector Search helped it build solutions for less cost than alternative technologies

Delivery Hero, a long-time MongoDB Atlas customer is the world’s leading local delivery platform, operating in 70-plus countries across 4 continents. Their quick commerce service enables customers to select fresh produce for delivery from local grocery stores. Approximately 10% of the inventory is fast-moving perishable produce that can go quickly out of stock. The company risks losing revenue and increasing customer churn if the customer didn’t have viable alternatives to their first choice. To address these risks, they are now using state-of-the-art AI models and MongoDB Atlas Vector Search to give hyperpersonalized alternatives to customers in real time if items they want to order are out of stock. With the introduction of MongoDB Atlas Vector Search, the data science team recognized that they could build a highly performant, real-time solution more quickly and for less cost than alternative technologies. 

MongoDB’s management believes that general-purpose LLMs (large language models) will win and will use RAG (retrieval augmented generation) as the primary way to combine generally available data to proprietary data; management is seeing advanced RAG use-cases in answering complex questions

There are some questions about LLMs, whether a general-purpose LLM or a fine-tune LLM, what the trade-offs are. Our belief is that given the performance of LLMs, you’re going to see the general purpose LLMs probably win and will use RAG as the predominant approach to marry generally available data with proprietary data. And then you are starting to see things like advanced RAG use cases where you get much more sophisticated ways to ask complex questions, provide more accurate and detailed answers and better adapt to different types of information and queries.

MongoDB’s management is seeing most AI workloads happen in the cloud, but they also see a lot of customers using open-source LLMs and running those workloads locally

We predominantly see most of the AI workloads in the cloud, but there are definitely lots of customers who are looking at using open source LLMs, in particular, things like Llama, and running those workloads locally.

MongoDB’s management believes MongoDB wins against Postgres for winning AI workloads because MongoDB can handle complex data types whereas Postgres, which is a relational – or SQL – database, struggles

MongoDB is designed to handle these different data structures. And I talked about we can help unify metadata, operational data, vector data and generate it all in one platform. Relational databases, and Postgres is one of them, have limitations in terms of what they can — how they can handle different types of data. In fact, when the data gets too large, these relational databases have to do what’s called off-row storage. And it becomes — it creates a performance overhead on these relational platforms. Postgres has this thing called TOAST, which is — stands for The Oversized-Attribute Storage Technique. And it’s basically a way to handle these different data types, but it creates a massive performance overhead. So we believe that we are architecturally far better for these more complex AI workloads than relational databases.

MongoDB’s management is seeing growing adoption of Vector, and Vector is helping attract new customers to MongoDB; an existing Atlas customer, a financial news organisation, migrated from Elastic Search to Atlas Search in order to use MongoDB’s Vector Search capabilities; an European energy company is using Vector Search for a geospatial search application

On Vector, we’re continuing to see growth in adoption, and we see Vector is effective in attracting new customers to the MongoDB platform. A world-renowned financial news organization, which is already running in Atlas, migrated from Elasticsearch to Atlas Search using Search Nodes to take advantage of our Vector Search capabilities to build a site search that combines lexical search with semantic search to find the most relevant articles for user query. And a European energy company built a geospatial search application using Atlas Search and Vector search and the app was built on-prem but — and to clouds to vectorize geospatial data and facilitate research and discovery.

MongoDB’s management is seeing MongoDB’s customers improve their software development productivity with the help of AI, but the rate of improvement is all over the place

[Question] We’ve talked before in the past that AI is just driving a lot of new code, making developers significantly more productive. Have you seen that behavior in any of your existing customers on Atlas where maybe their utilization rate goes up or the number of applications built per customer goes up?

[Answer] A common question I ask our customers when I meet with them in terms of what code generation tools that they’re using and what benefits they’re gaining. The answers tend to be a little bit all over the map. Some people see 10%, 15% productivity improvement. Some people say 20%, 25% productivity improvement. Some people say it helps my senior developers be more productive. Some people say it helps my junior developers become more like senior developers. So the answers tend to be all over the map.

Nvidia (NASDAQ: NVDA)

Nvidia’s Data Center revenue had incredibly strong growth in 2024 Q2, driven by demand for the Hopper GPU computing platform; compute revenue was up by 2.5x while networking revenue was up by 2x

Data Center revenue of $26.3 billion was a record, up 16% sequentially and up 154% year-on-year, driven by strong demand for NVIDIA Hopper, GPU computing and our networking platforms. Compute revenue grew more than 2.5x. Networking revenue grew more than 2x from the last year.

Even as Nvidia is getting ready to launch its Blackwell-architecture GPUs, customers are still buying the Hopper-architecture GPUs; the H200 platform, based on the Hopper architecture, started ramping in 2024 Q2 and offers 40% more memory bandwidth than the H100; management thinks that the reasons why the Hopper-architecture chips still enjoy strong demand despite the imminent arrival of the Blackwell-architecture chips are (1) AI companies need chips today to process data right now, and (2) AI companies are in a race to build the best model and they’re all racing to be the first

Customers continue to accelerate their Hopper architecture purchases while gearing up to adopt Blackwell…

…NVIDIA H200 platform began ramping in Q2, shipping to large CSPs, consumer Internet and enterprise companies. The NVIDIA H200 builds upon the strength of our Hopper architecture and offering over 40% more memory bandwidth compared to the H100…

…The demand for Hopper is really strong. And it’s true, the demand for Blackwell is incredible. There’s a couple of reasons for that. The first reason is, if you just look at the world’s cloud service providers and the amount of GPU capacity they have available, it’s basically none…

…A generative AI company spends the vast majority of their invested capital into infrastructure so that they could use an AI to help them create products. And so these companies need it now. They just simply can’t afford — you just raise money, they want you to put it to use now. You have processing that you have to do. You can’t do it next year. You got to do it today. And so that’s one reason. The second reason for Hopper demand right now is because of the race to the next plateau. The first person to the next plateau gets to introduce some revolutionary level of AI. The second person who gets there is incrementally better or about the same. And so the ability to systematically and consistently race to the next plateau and be the first one there is how you establish leadership…

…We believe our Hopper will continue to grow into the second half. We have many new products for Hopper or existing products for Hopper that we believe will start continuing to ramp in the next quarters, including our Q3 and those new products moving to Q4. So let’s say, Hopper, therefore, versus H1 is a growth opportunity for that. 

Nvidia’s management thinks that the next generation of AI models will need 10-20 times more compute to train

Next-generation models will require 10 to 20x more compute to train with significantly more data. The trend is expected to continue.

Nvidia’s management sees inferencing accounting for 40% of Data Center revenue over the last 4 quarters (was 40% as of 2024 Q1)

Over the trailing 4 quarters, we estimate that inference drove more than 40% of our Data Center revenue.

Nvidia’s management is seeing demand coming from builders of frontier AI models, consumer Internet companies, and companies building generative AI applications for a wide range of use cases

Demand for NVIDIA is coming from frontier model makers, consumer Internet services, and tens of thousands of companies and start-ups building generative AI applications for consumers, advertising, education, enterprise and health care, and robotics. 

Nvidia’s Data Center revenue in China grew sequentially in 2024 Q2, but still remains below the level seen prior to export controls; management expects tough competition in China

Our Data Center revenue in China grew sequentially in Q2 and a significant contributor to our Data Center revenue. As a percentage of total Data Center revenue, it remains below levels seen prior to the imposition of export controls. We continue to expect the China market to be very competitive going forward.

Nvidia has leadership in inference

The latest round of MLPerf inference benchmarks highlighted NVIDIA’s inference leadership, with both NVIDIA Hopper and Blackwell platforms combining to win gold medals on all tests.

Nvidia’s Blackwell family of chips combines GPUs, CPUs, DPUs (data processing units), NVLink, and networking; the GB200 NVL72 system in the Blackwell family links up 72 GPUs to act as 1 GPU and is up to 30 times faster for LLM (large language model) inference workloads; Nvidia has made a change to the Blackwell architecture to improve production yields; Blackwell’s production is expected to ramp in the fourth quarter of 2024; management sees demand for Blackwell exceeding supply by a wide margin up to 2025; there are more than 100 different Blackwell architecture systems; Nvidia’s Blackwell systems come in both air-cooled and liquid-cooled flavours; management expects Nvidia’s Data Center business to grow significantly in 2025 and 2026, powered by the Blackwell system; management sees Blackwell as a step-function improvement over Hopper that delivers 3-5 times more AI throughput than Hopper; Blackwell required 7 one-of-a-kind chips to build; Nvidia designed and optimised the Blackwell system end-to-end

The NVIDIA GB200 NVL72 system with the fifth-generation NVLink enables all 72 GPUs to act as a single GPU and deliver up to 30x faster inference for LLM’s workloads and unlocking the ability to run trillion-parameter models in real time…

…We executed a change to the Blackwell GPU mass to improve production yields. Blackwell production ramp is scheduled to begin in the fourth quarter and continue into fiscal year ’26. In Q4, we expect to get several billion dollars in Blackwell revenue…

…Demand for Blackwell platforms is well above supply, and we expect this to continue into next year…

…There are something like 100 different types of Blackwell-based systems that are built that were shown at Computex, and we’re enabling our ecosystem to start sampling those…

…We offer multiple configurations of Blackwell. Blackwell comes in either a Blackwell classic, if you will, that uses the HGX form factor that we pioneered with Volta. I think it was Volta. And so we’ve been shipping the HGX form factor for some time. It is air cooled. The Grace Blackwell is liquid cooled…

…We expect to grow our Data Center business quite significantly next year. Blackwell is going to be a complete game changer for the industry. And Blackwell is going to carry into the following year…

…Blackwall is a step-function leap over Hopper. Blackwell is an AI infrastructure platform, not just the GPU. It also happens to be the name of our GPU, but it’s an AI infrastructure platform. As we reveal more of Blackwell and sample systems to our partners and customers, the extent of Blackwell’s lead becomes clear. The Blackwell vision took nearly 5 years and 7 one-of-a-kind chips to realize: the Gray CPU, the Blackwell dual GPU and a colos package, ConnectX DPU for East-West traffic, BlueField DPU for North-South and storage traffic, NVLink switch for all-to-all GPU communications, and Quantum and Spectrum-X for both InfiniBand and Ethernet can support the massive burst traffic of AI. Blackwell AI factories are building size computers. NVIDIA designed and optimized the Blackwell platform full stack, end-to-end, from chips, systems, networking, even structured cables, power and cooling and mounts of software to make it fast for customers to build AI factories. These are very capital-intensive infrastructures. Customers want to deploy it as soon as they get their hands on the equipment and deliver the best performance and TCO. Blackwell provides 3 to 5x more AI throughput in a power-limited data center than Hopper…

…The Blackwell system lets us connect 144 GPUs in 72 GB200 packages into 1 NVLink domain, with an aggregate NVLink bandwidth of 259 terabytes per second in 1 rack. Just to put that in perspective, that’s about 10x higher than Hopper.  

Nvidia’s Ethernet for AI revenue doubled sequentially; management sees Nvidia’s ethernet product, Spectrum-X, enjoying wide support from the AI ecosystem; Spectrum-X performs 1.6 times better than traditional Ethernet; management plans to launch new Spectrum-X products every year and thinks that Spectrum-X will soon become a multi-billion dollar product

Our Ethernet for AI revenue, which includes our Spectrum-X end-to-end Ethernet platform, doubled sequentially with hundreds of customers adopting our Ethernet offerings. Spectrum-X has broad market support from OEM and ODM partners and is being adopted by CSPs, GPU cloud providers and enterprises, including xAI to connect the largest GPU compute cluster in the world. Spectrum-X supercharges Ethernet for AI processing and delivers 1.6x the performance of traditional Ethernet. We plan to launch new Spectrum-X products every year to support demand for scaling compute clusters from tens of thousands of GPUs today to millions of DPUs in the near future. Spectrum-X is well on track to begin a multibillion-dollar product line within a year.

Japan’s government is working with Nvidia to build an AI supercomputer; Nvidia’s management thinks sovereign AI revenue will be in the low-teens billion-range this year; management is seeing countries want to build their own generative AI that incorporates their own language, culture, and data

Japan’s National Institute of Advanced Industrial Science and Technology is building its AI Bridging Cloud Infrastructure 3.0 supercomputer with NVIDIA. We believe sovereign AI revenue will reach low double-digit billions this year…

…It certainly is a unique and growing opportunity, something that surfaced with generative AI and the desires of countries around the world to have their own generative AI that would be able to incorporate their own language, incorporate their own culture, incorporate their own data in that country.

Most of the Fortune 100 companies are working with Nvidia on AI projects

We are working with most of the Fortune 100 companies on AI initiatives across industries and geographies. 

Nvidia’s management is seeing a range of applications driving the company’s growth; these applications include (1) Amdocs’ smart agent which is reducing customer service costs by 30%, and (2) Wistron’s usage of Nvidia AI Ominiverse to reduce cycle times in its factories by 50%

A range of applications are fueling our growth, including AI-powered chatbots, generative AI copilots and agents to build new, monetizable business applications and enhance employee productivity. Amdocs is using NVIDIA generative AI for their smart agent, transforming the customer experience and reducing customer service costs by 30%. ServiceNow is using NVIDIA for its Now Assist offering, the fastest-growing new product in the company’s history. SAP is using NVIDIA to build Joule copilot. Cohesity is using NVIDIA to build their generative AI agent and lower generative AI development costs. Snowflake, who serves over 3 billion queries a day for over 10,000 enterprise customers, is working with NVIDIA to build copilots. And lastly, Wistron is using NVIDIA AI Omniverse to reduce end-to-end cycle times for their factories by 50%.

Every automobile company that is developing autonomous vehicle technology is working with Nvidia; management thinks that automotive will account for multi-billions in revenue for Nvidia; Nvidia won the Autonomous Brand Challenge at the recent Computer Vision and Pattern Recognition Conference

Every automaker developing autonomous vehicle technology is using NVIDIA in their data centers. Automotive will drive multibillion dollars in revenue across on-prem and cloud consumption and will grow as next-generation AV models require significantly more compute…

…At the Computer Vision and Pattern Recognition Conference, NVIDIA won the Autonomous Brand Challenge in the end-to-end driving at scale category, outperforming more than 400 entries worldwide. 

Nvidia’s management announced Nvidia AI Foundry – a platform for building custom AI models – in 2024 Q2; users of Nvidia AI Foundry are able to customise Meta’s Llama 3.1 foundation AI model; Nvidia AI Foundry is the first platform where users are able to customise an open-source, frontier-level foundation AI model; Accenture is already using Nvidia AI Foundry 

During the quarter, we announced a new NVIDIA AI foundry service to supercharge generative AI for the world’s enterprises with Meta’s Llama 3.1 collection of models… 

…Companies for the first time can leverage the capabilities of an open source, frontier-level model to develop customized AI applications to encode their institutional knowledge into an AI flywheel to automate and accelerate their business. Accenture is the first to adopt the new service to build custom Llama 3.1 models for both its own use and to assist clients seeking to deploy generative AI applications.

Companies from many industries are using NIMs (Nvidia inference microservices) for deployment of generative AI; AT&T saw 70% cost savings and 8 times latency reduction with NIM; there are 150 organisations using NIMs; Nvidia recently announced NIM Agent Blueprints, a catalog of reference AI applications; Nvidia is using NIMs to open the Nvidia Omniverse to new industries

NVIDIA NIMs accelerate and simplify model deployment. Companies across health care, energy, financial services, retail, transportation, and telecommunications are adopting NIMs, including Aramco, Lowes, and Uber. AT&T realized 70% cost savings and 8x latency reduction affter moving into NIMs for generative AI, call transcription and classification. Over 150 partners are embedding NIMs across every layer of the AI ecosystem. 

We announced NIM Agent Blueprints, a catalog of customizable reference applications that include a full suite of software for building and deploying enterprise generative AI applications. With NIM Agent Blueprints, enterprises can refine their AI applications over time, creating a data-driven AI flywheel. The first NIM Agent Blueprints include workloads for customer service, computer-aided drug discovery, and enterprise retrieval augmented generation. Our system integrators, technology solution providers, and system builders are bringing NVIDIA NIM Agent Blueprints to enterprises…

…We announced new NVIDIA USD NIMs and connectors to open Omniverse to new industries and enable developers to incorporate generative AI copilots and agents into USD workloads, accelerating our ability to build highly accurate virtual worlds.

Nvidia’s AI Enterprise software platform is powering Nvidia’s software-related business to approach a $2 billion annual revenue run-rate by the end of this year; management thinks Nvidia AI Enterprise represents great value for customers by providing GPUs at a price of $4,500 per GPU per year; management thinks the TAM (total addressable market) for Nvidia’s AI software business can be significant

NVIDIA NIM and NIM Agent Blueprints are available through the NVIDIA AI Enterprise software platform, which has great momentum. We expect our software, SaaS and support revenue to approach a $2 billion annual run rate exiting this year, with NVIDIA AI Enterprise notably contributing to growth…

…At $4,500 per GPU per year, NVIDIA AI Enterprise is an exceptional value for deploying AI anywhere. And for NVIDIA’s software TAM, it can be significant as the CUDA-compatible GPU installed base grows from millions to tens of millions. 

Computers that contain Nvidia’s RTX chip can deliver up to 1,300 AI TOPS (tera operations per second); there are more than 200 RTX AI computer models from computer manufacturers; there is an installed base of 100 million RTX AI computers; a game called Mecha BREAK is the first game to use Nvidia ACE, a generative AI service for creating digital humans

Every PC with RTX is an AI PC. RTX PCs can deliver up to 1,300 AI tops and there are now over 200 RTX AI laptops designed from leading PC manufacturers. With 600 AI-powered applications and games and an installed base of 100 million devices, RTX is set to revolutionize consumer experiences with generative AI. NVIDIA ACE, a suite of generative AI technologies is available for RTX AI PCs. Mecha BREAK is the first game to use NVIDIA ACE, including our small language model, Nemotron-4 4B, optimized on device inference. 

Foxconn, the largest electronics manufacturer in the world, and Mercedes-Benz, the well-known auto manufacturer, are using Nvidia Omniverse to produce digital twins of their manufacturing plants

The world’s largest electronics manufacturer, Foxconn, is using NVIDIA Omniverse to power digital twins of the physical plants that produce NVIDIA Blackwell systems. And several large global enterprises, including Mercedes-Benz, signed multiyear contracts for NVIDIA Omniverse Cloud to build industrial digital twins of factories.

Many robotics companies are using Nvidia’s AI robot software

Boston Dynamics, BYD Electronics, Figure, Intrinsyc, Siemens, Skilled AI and Teradyne Robotics are using the NVIDIA Isaac robotics platform for autonomous robot arms, humanoids and mobile robots.

Nvidia’s management is seeing some customers save up to 90% in computing costs by transitioning from genera-purpose computing (CPUs) to accelerated computing (GPUs)

We know that accelerated computing, of course, speeds up applications. It also enables you to do computing at a much larger scale, for example, scientific simulations or database processing. But what that translates directly to is lower cost and lower energy consumed. And in fact, this week, there’s a blog that came out that talked about a whole bunch of new libraries that we offer. And that’s really the core of the first platform transition, going from general-purpose computing to accelerated computing. And it’s not unusual to see someone save 90% of their computing cost. And the reason for that is, of course, you just sped up an application 50x. You would expect the computing cost to decline quite significantly.

Nvidia’s management believes that generative AI is a new way to write software and is changing how every layer of computing is done

Generative AI, taking a step back about why it is that we went so deeply into it, is because it’s not just a feature, it’s not just a capability, it’s a fundamental new way of doing software. Instead of human-engineered algorithms, we now have data. We tell the AI, we tell the model, we tell the computer what are the expected answers, what are our previous observations, and then for it to figure out what the algorithm is, what’s the function. It learns a universal — AI is a bit of a universal function approximator and it learns the function. And so you could learn the function of almost anything, and anything that you have that’s predictable, anything that has structure, anything that you have previous examples of. And so now here we are with generative AI. It’s a fundamental new form of computer science. It’s affecting how every layer of computing is done from CPU to GPU, from human-engineered algorithms to machine-learned algorithms, and the type of applications you could now develop and produce is fundamentally remarkable.

Nvidia’s management thinks AI models are still seeing the benefits of scaling

There are several things that are happening in generative AI. So the first thing that’s happening is the frontier models are growing in quite substantial scale. And we’re still all seeing the benefits of scaling.

The amount of compute needed to train an AI model goes up much faster than the size of the model; management thinks the next generation of AI models could require 10-40 times more compute 

Whenever you double the size of a model, you also have to more than double the size of the data set to go train it. And so the amount of flops necessary in order to create that model goes up quadratically. And so it’s not unexpected to see that the next-generation models could take 10x, 20x, 40x more compute than last generation.

Nvidia’s management is seeing more frontier model makers in 2024 than in 2023

Surprisingly, there are more frontier model makers than last year.

Nvidia’s management is seeing advertising-related computing needs shifting from being powered by CPUs to being powered by GPUs and generative AI

The largest systems, largest computing systems in the world today, and you’ve heard me talk about this in the past, which are recommender systems moving from CPUs. It’s now moving from CPUs to generative AI. So recommender systems, ad generation, custom ad generation targeting ads at very large scale and quite hyper-targeting, search and user-generated content, these are all very large-scale applications that have now evolved to generative AI.

Nvidia’s management is seeing generative AI startups generating tens of billions of revenue-opportunities for cloud computing providers

The number of generative AI start-ups is generating tens of billions of dollars of cloud renting opportunities for our cloud partners

Nvidia’s management is seeing that cloud computing providers have zero GPU capacity available because they are using it for internal workloads (such as accelerating data processing) and renting it out to model makers and other AI startups

If you just look at the world’s cloud service providers and the amount of GPU capacity they have available, it’s basically none. And the reason for that is because they’re either being deployed internally for accelerating their own workloads, data processing, for example…

…The second is, of course, the rentals. They’re renting capacity to model makers. They’re renting it to start-up companies. 

Nvidia’s management thinks Nvidia’s GPUs are the only AI GPUs that process and accelerate data; before the advent of generative AI, the number one use case of Nvidia’s GPUs was to accelerate data processing

NVIDIA’s GPUs are the only accelerators on the planet that process and accelerate data. SQL data, Panda’s data, data science toolkits like Panda’s, and the new one, Polar’s, these are the ones that are the most popular data processing platforms in the world, and aside from CPUs which, as I’ve mentioned before, are really running out of steam, NVIDIA’s accelerated computing is really the only way to get boosting performance out of that. And so the #1 use case long before generative AI came along is the migration of applications one after another to accelerated computing.

Nvidia’s management thinks that those who purchase Nvidia AI chips are getting immediate ROI (return on investment) for a few reasons: (1) GPUs are a better way to build data centers compared to CPUs because GPUs save money on data processing compared to CPUs, (2) cloud computing providers who rent out GPUs are able to rent out their GPUs the moment they are built up in the data center because there are many generative AI companies clamouring for the chips, and (3) generative AI improves a company’s own services, which delivers a fast ROI

The people who are investing in NVIDIA infrastructure are getting returns on it right away. It’s the best ROI infrastructure, computing infrastructure investment you can make today. And so one way to think through it, probably the easiest way to think through it is just to go back to first principles. You have $1 trillion worth of general-purpose computing infrastructure. And the question is, do you want to build more of that or not?

And for every $1 billion worth of Juniper CPU-based infrastructure that you stand up, you probably rent it for less than $1 billion. And so because it’s commoditized, there’s already $1 trillion on the ground. What’s the point of getting more? And so the people who are clamoring to get this infrastructure, one, when they build out Hopper-based infrastructure and soon, Blackwell-based infrastructure, they start saving money. That’s tremendous return on investment. And the reason why they start saving money is because data processing saves money, and data processing is probably just a giant part of it already. And so recommender systems save money, so on and so forth, okay? And so you start saving money.

The second thing is everything you stand up are going to get rented because so many companies are being founded to create generative AI. And so your capacity gets rented right away and the return on investment of that is really good.

And then the third reason is your own business. Do you want to either create the next frontier yourself or your own Internet services, benefit from a next-generation ad system or a next-generation recommender system or a next-generation search system? So for your own services, for your own stores, for your own user-generated content, social media platforms, for your own services, generative AI is also a fast ROI.

Nvidia’s management is seeing a significant number of data centers wanting liquid-cooled GPU systems because the use of liquid cooling enables 3-5 times more AI throughput compared to the past, resulting in cheaper TCO (total cost of ownership)

The number of data centers that want to go to liquid cooled is quite significant. And the reason for that is because we can, in a liquid-cooled data center, in any power-limited data center, whatever size of data center you choose, you could install and deploy anywhere from 3 to 5x the AI throughput compared to the past. And so liquid cooling is cheaper. Our TCO is better, and liquid cooling allows you to have the benefit of this capability we call NVLink, which allows us to expand it to 72 Grace Blackwell packages, which has essentially 144 GPUs.

Nvidia does not do the full integration of its GPU systems into a data center because it is not the company’s area of expertise

Our customers hate that we do integration. The supply chain hates us doing integration. They want to do the integration. That’s their value-add. There’s a final design-in, if you will. It’s not quite as simple as shimmying into a data center, but the design fit-in is really complicated. And so the design fit-in, the installation, the bring-up, the repair-and-replace, that entire cycle is done all over the world. And we have a sprawling network of ODM and OEM partners that does this incredibly well.

Nvidia has released many new libraries for CUDA, across a wide variety of use cases, for AI software developers to work with

Accelerated computing starts with CUDA-X libraries. New libraries open new markets for NVIDIA. We released many new libraries, including CUDA-X Accelerated Polars, Pandas and Spark, the leading data science and data processing libraries; CUVI-S for vector databases, this is incredibly hot right now; Ariel and Ciona for 5G wireless base station, a whole world of data centers that we can go into now; Parabricks for gene sequencing and AlphaFold2 for protein structure prediction is now CUDA accelerated.

Nvidia now has 3 networking platforms for GPUs

We now have 3 networking platforms, NVLink for GPU scale-up, Quantum InfiniBand for supercomputing and dedicated AI factories, and Spectrum-X for AI on Ethernet. NVIDIA’s networking footprint is much bigger than before. 

Salesforce (NYSE: CRM)

Agentforce is a new architecture and product that management believes will be fundamental to Salesforce’s AI leadership in the next decade; Salseforce will be introducing Agentforce in its upcoming Dreamforce customer-event; Agentforce is an autonomous AI and management will be getting every attendee at Dreamforce to turn on their own AI agents; Salesforce is already building agents for the company, Workday, and Workday will be Salseforce’s first Agentforce partner; Agentforce allows companies to build custom agents for sales, service, marketing, and commerce; management believes that within a year, most companies will be deploying autonomous AI agents at scale, and these agents will have a big positive impact on companies’ operations; Agentforce is currently management’s singular focus; many companies are already using Agentforce, including one of the world’s largest healthcare companies, which is resolving more than 90% of patient inquiries with Agentforce, and thinks Agentforce is much better than any other competing AI platform; a very large media company is using Agentforce to resolve 90% of employee and consumer issues; management thinks Salesforce is the first company to deploy high-quality enterprise AI agents at scale; Agentforce is not a co-pilot, it is an autonomous agent that is accurate and can be deployed right out of the box; users of Agentforce can do advanced planning and reasoning with minimal input; management sees Agentforce as being a trusted colleague that will complement human users; management sees thousands of companies using Agentforce by January 2025; early trials of Agentforce has showed remarkable success

We’re going to talk about a whole different kind of sales force today, a different kind of architecture and a product that we didn’t even talk about on the last earnings call that is going to be fundamental to our future and a manifestation of our decade of AI leadership, which is Agentforce. Now in just a few weeks, we’re going to kick off Dreamforce, and I hope all of you are planning to be there, the largest AI event in the world with more than 45,000 trailblazers in San Francisco. And this year, Dreamforce is really becoming Agentforce…

…We’re going to show our new Agentforce agents and how we’ve reimagined enterprise software for this new world of autonomous AI. And every customer, I’m going to try to get every customer who comes to Dreamforce to turn agents on while they’re there…

…This idea that you’re not just going to have sales agents and service agents who probably read, heard maybe you saw in CBC, we’re building the agents for Workday and we’re going to be building custom agents for so many of you as well with Agentforce, because it is a development platform as well as this incredible capability to radically extend your sales and service organizations.

So when you arrive at the Dreamforce campus, you’re going to see a big sign outside that says, humans with agents drive customer success together. And that’s because we now so strongly believe the future isn’t about having a sales force or a service force or a marketing force or a commerce force or an analytics force. The future is about also having an Agentforce. And while many customers today don’t yet have agent forces, but they do have sales forces or service forces, I assure you that within a year, we’re all going to have agent forces, and we’re going to have them at scale. And it’s going to radically extend our companies and it’s going to augment our employees, make us more productive. It’s going to turn us into these incredible margin and revenue machines. It’s going to be pretty awesome…

…with this Agentforce platform, we’re making it easy to build these powerful autonomous agents for sales, for service, for marketing, for commerce, automating the entire workflow on their own, embedding agents in the flow of work and getting our customers to the agent future first. And this is our primary goal of our company right now. This is my singular focus…

…We’re going to talk about the customers who have it, customers like OpenTable and Wiley and — and ADP and RBC and so many others who are deploying these agents and running them on top of our Data Cloud and our apps…

…At Dreamforce, you’re going to hear one of the very largest health care companies in the world. It’s got 20 million consumers here in the United States who is resolving more than 90% of all patient inquiries with Agentforce and they’re benchmarking us significantly higher than any other competing AI platform, and that’s based on some incredible AI breakthroughs that we have had at Salesforce…

…One of these very large media companies that we work with, a lot of probably know who have everything, every possible media asset, while they’re just resolving 90% of all of their employee and consumer issues with Agentforce, pretty awesome. So there’s nothing more transformational than agents on the technology horizon that I can see and Salesforce is going to be the first company at scale to deploy enterprise agents and not just any enterprise agents, the highest quality, most accurate agents in the world…

…We’re seeing that breakthrough occur because with our new Agentforce platform, we’re going to make a quantum leap for in AI, and that’s why it wants you all at Dreamforce, because I want you to have your hands on this technology to really understand this. This is not co-pilots…

…These agents are autonomous. They’re able to act with accuracy. They’re able to come right out of the box. They’re able to go right out of the platform…

…These agents don’t require a conversational prompt to take action. You can do advanced planning, reasoning with minimal human input. And the example of this incredible health care company, you’re going to be able to say to the agent, “Hey, I want to look at my labs, I want to understand this. It looks like I need repeat labs. Can you reschedule those for me? It looks like I need to see my doctor, can you schedule that for me? I also want to get an MRI, I want to get this.” And the level of automation that we’re going to be able to provide and unleash the productivity back into these organizations is awesome…

…This is going to be like having these trusted colleagues can handle these time-consuming tasks engaging with these — whether it’s inbound lead or resolving this customer, patient inquiry or whatever it is, this is humans with agents driving customer success together and Agentforce agents can be set up in minutes, easily scalable, work around with the block, any language. And by the beginning of next fiscal year, we will have thousands of customers using this platform. And we will have hand help them to make it successful for them, deploy it. The early trials have been remarkable to see these customers have the success, it has been just awesome…

…We’re just at the beginning of building an Agentforce ecosystem with companies able to build agents on our platform for their workforce and use cases, and we’re excited to have Workday as our first agent force partner.

Salesforce has been able to significantly reduce hallucinations with its AI products, and thus deliver highly accurate results, through the use of new retrieval augmented generation (RAG) techniques

The accuracy of our results, the reduction of hallucinations and the level of capability of AI is unlike anything I think that any of us have ever seen, and we’ve got some incredible new techniques, especially incredible new augmented RAG techniques that are delivering us the capability to deliver this accuracy with our — for our customers.

Salesforce’s management still sees the company as the No.1 AI CRM in the world

Of course, Salesforce is the #1 AI CRM.

 In 2024 Q2, Einstein delivered 25 trillion transactions and 1 trillion workflows; Wyndham is using Einstein to reduce average call times to free up time for service agents for higher-value work

We’re just operating at this incredible scale, delivering 25 trillion Einstein transactions across all of the clouds during the quarter, that’s 25 trillion and more than 1 trillion workflows…

…MuleSoft allows Wyndham to unlock business-critical data from various platforms and onboard future franchisees faster. And with Einstein generated recommended service replies, average call times have been reduced and service agents can focus on higher priority work

Salesforce’s management thinks many of the company’s customers have a misconception about AI in that they need to build and train their own AI models; management is able to use Salesforce’s AI models and resolve issues much better than its customers’ own models; management thinks Salesforce’s AI models have the highest efficacy

I think that there’s a lot of misconceptions about AI with my customers. I have been out there very disappointed with the huge amount of money that so many of these customers have wasted on AI. They are trying to DIY their AI…

…This idea that our customers are going to have to build their own models, train their own models, retrain their own models, retrain them again and I’m meeting with these customers, and they’re so excited when they and they say, “Oh, I built this model, and we’re resolving 10%, 20%, 30%, 40% and — of this or that and whatever. ” And I’m like, really, we’ll take a look at our models and our capability where you don’t have to train or retrain anything and you’re going to get more than 90%. And then they say, wait a minute, how do you do that? And this is a moment where every customer needs to realize you don’t need the DIY your AI. You can use a platform like Salesforce to get the highest efficacy of artificial intelligence, the best capability to fully automate your company, achieve all of your goals and you can do it with professional enterprise software…

…We’re in met with one of the largest CIOs in the world is telling me how excited he was for the B2C part of this business, he built this model and accuracy rates, than I was like, really, let me show you what we’re doing here. And then he said to me, why am I doing this? Why am I not just using your platform? And I said good question. So these customers are spending millions of dollars, but are they really getting the results that they want? It feels like this early days of cloud. It’s just early days of social mobile. Customers feel like they have to DIY it, they don’t need to, they can make it all happen themselves. And you can just see that to deliver this very high-quality capability they can use a deeply integrated platform like Salesforce.

Salesforce’s management is seeing the company’s customers get immediate ROI (return on investment) from deploying AI automation solutions because the solutions can be easily and quickly customised and configured

We’ve created out-of-the-box platform to deliver all of this for them. So this could be service reply recommendations, account summaries, report generation, you’ve seen in Slack, this kind of auto summarization, recaps, all of these amazing things, the level of automation, the amount of code that our team has written, the transformation of our platform in the last 18 months, it’s remarkable. And customers love it because they can take the platform and then all of this generative AI use case customize it for their own needs or configure it using our capability because they’re doing that without writing a line of code. It’s clicks, not code, deploy them in days, not weeks. They’re doing this in months, not years, and you’re getting immediate ROI. 

Salesforce’s management thinks many of the company’s customers are really disappointed with Microsoft Co-pilot because of a lack of accuracy

So many customers are so disappointed in what they bought from Microsoft CoPilots because they’re not getting the accuracy and the response that they want. Microsoft is disappointed so many customers with AI. 

Wiley is achieving double-digit percentage increase in customer satisfaction and deflection rate, and 50% case resolution with the first generation of Agentforce; Royal Bank of Canada and AP are seeing 90% case resolution rates with the second generation of Agentforce; OpenTable is using Agentforce to support its interactions with 60,000 restaurants and 160 million diners

Wiley is a long-standing Salesforce customer. It’s one of our first deployments in the first agent force trial. It’s pretty awesome. And you all know they make textbooks and it’s back-to-school. But maybe you don’t know that Wiley has to like surge their sales and service organization at back-to-school time when everyone’s buying these textbooks. Well, now they can use agents to do that surge. They don’t have to go buy a bunch of gig workers and bring them in. and that age and capacity is so exciting for them. What we saw with Wiley was, this is a quote from them, “we’re seeing double-digit percentage increase in customer satisfaction and deflection rate compared to older technologies and in these early weeks of our busiest season. ” So that was very reassuring to us. that we have the right thing that’s happening. And Wiley has already seen a 50% increase in case resolution. That’s with our first generation of Agentforce.

As I mentioned, the second generation of Agentforce, which we have with customers already, including some of these amazing organizations like Royal Bank of Canada, ADP and others is 90% case resolution. It is an awesome moment in this tech business.

OpenTable is another super great story. You all know they are managing 60,000 restaurants, 160 million diners to support. They’re on Agentforce now. They require that incredible scale to deliver top-notch customer service. That’s why they’re using the product. It’s been awesome to get their results and it can be all kinds of questions resolving basic issues, account activations, reservation management, loyalty point expiration. Agentforce for service can easily answer all of these questions like when do my points expire for a diner asset, a follow-up question like, what about in Mexico? What about — can I make this change? That’s where we’re delivering those incredible moments for OpenTable, giving them this kind of productivity enhancement. 

Agentforce is driving growth in cloud products’ sales for Salesforce

Agentforce for sales, you can imagine extending your sales force with SDRs, BDRs who are agents that are going out and building pipeline for you and generating all kind of demand and even really closing deals. So, this is going to drive sales cloud growth. It already is, service cloud growth. It already is because customers are going to extend their sales and service organizations and become a lot more productive with these agents.

Salesforce will be releasing industry-specific AI agents in the coming months

In the coming months, we’re going to release Agentforce agents for other roles, including industry-specific agents, health agents, as I mentioned. 

Data Cloud provides the foundation for Agentforce because it holds a huge amount of data and metadata; management continues to believe that data is the foundation of AI; Data Cloud federates and connects to all other data clouds of a user to deliver super accurate AI; Data Cloud is Salesforce’s fastest-growing organic product and will be the fastest to hit $1 billion, $5 billion, and $10 billion in revenue; Data Cloud customers were up 130% year-on-year in 2024 Q2; number of Data Cloud customers spending more than $1 million annually have doubled; Data Cloud processed 2.3 quadrillion records in 2024 Q2 (was 2 quadrillion in 2024 Q1); Data Cloud consumption was up 110% year-on-year in 2024 Q2; American Family Insurance is using Data Cloud to create a 360-degree view of customers; Adecco Group is using Data Cloud to create seamless access to information for 27,000 of its employees; Windhma is using Data Cloud to unify profiles of 165 million guest records, many of which are duplicates across multiple sources

This type of performance from our Agentforce platform wouldn’t be possible without Data Cloud. One of the reasons that our agents are so accurate is because of the huge amount of data and metadata that we had. And data is the foundation for every AI transformation. And with Data Cloud, we’re providing a high-performance data lake that brings together all our customer and business data, federating data from external repositories through this credible zero-copy alliance. So customers can use our Data Cloud and then federate and connect to all their other data clouds and then we can bring it all together to deliver the super accurate AI. 

That’s why Data Cloud is absolutely our fastest-growing organic product in history. It will be the fastest product to $1 billion — it’s going to probably be the fastest product of $5 billion, $10 billion. In Q2, the number of paid Data Cloud customers grew 130% year-over-year and the number of customers spending more than $1 million annually have already doubled. In the second quarter alone, and this is amazing, data Cloud processed 2.3 quadrillion records with 110% platform consumption growth year-over-year…

…American Family Insurance with millions of policyholders nationwide is using Data Cloud to consolidate data from multiple sources through our zero-copy partner network, creating a 360-view of the customers, enabling quick segmentation and activating lead data, including their real-time web interactions. The Adecco Group expanded their data cloud in the quarter, a great example of a company leveraging its gold mine of data to gain a unified view of its customers. Connecting all this data means that 27,000 Adecco employees using Salesforce will have seamless access to key information, including financial metrics and job fulfillment status, to help Adecco improve their job fill rate ratio and reduce their cost to serve…

…Wyndham utilizes Data Cloud to unify profiles of 165 million guest records, many of which were duplicates across many sources like Amazon Redshift and the Sabre Reservation System as well as Sales Cloud, Marketing Cloud and Service Cloud. 

Salesforce has rewritten all of its software to be under one unified platform; management thinks building AI agents without a unified platform is risky; the decision to unite all of Salesforce’s software was made 18 months ago with the shift to AI

We’ve automated every customer touch point and now we’re bringing these apps, data and agents together. It’s these 3 levels, and this is in 3 separate pieces of code or 3 different platforms or 3 different systems. This is 1 platform. We’ve rewritten all of our acquisitions, all of our core modules, our Data Cloud and our agents as 1 unified platform, which is how we are delivering not only this incredible functionality but this high level of accuracy and capability. And from this first-hand experience in meeting with these customers around the globe, I can unequivocably tell you that building these agents without a complete integrated platform is like trying to assemble a plane mid-flight, it’s risky chaotic and it’s not likely to succeed…

…With the shift to AI, it just became clear 18 months ago, we need to hit the accelerator pedal and rewrite all these things onto the core platform because customers are going to get this incredible value by having 1 integrated system, and it scales from small companies to extremely large companies. 

Bookings for Salesforce’s AI products was up more than 100% year-on-year in 2024 Q2; Salesforce signed 1,500 AI deals in 2024 Q2; aircraft maker Bombardier is using Salesforce’s AI products to arm sales reps with better information on, and recommendations for, prospects

We’re already accelerating this move from AI hype to AI reality for thousands of customers with amazing capabilities across our entire AI product portfolio. New bookings for these products more than doubled quarter-over-quarter. We signed 1,500 AI deals in Q2 alone. Some of the world’s largest brands are using AI solutions, including Alliant, Bombardier and CMA CGM. Bombardier, the maker of some of the world’s top performing aircraft, is enabling sales reps to sell smarter by consolidating need to know information on prospects in advance of meetings and providing recommendations on how to best engage with them through the Einstein copilot and prompt builder. 

Salesforce has a new team called Salesforce CTOs that will work alongside customers in deploying AI agents

To help our customers navigate this new world, we just launched a new team called Salesforce CTOs. These are deeply technical individuals who work alongside our customers to help them create and execute a plan for every stage of their AI journey to become agent first companies. 

Salesforce sees itself as customer zero for all its AI products, including Agentforce, and it is deploying its own AI products internally with success; 35,000 Salesforce employees are using Einstein as an AI assistant; Salesforce has already used Slack AI to create 500,000 channel summaries since February 2024, saving 3 million hours of work

We’re continuing our own AI journey internally as a Customer Zero of all of our products with great results. We now have 35,000 employees using Einstein as a trusted AI assistant, helping them work smarter and close deals faster. And since we launched Slack AI in February, our employees have created more than 500,000 channels — channel summaries, saving nearly 3 million hours of work. We’ll, of course, deploy Agentforce agents soon in a variety of different roles and tasks to augment, automate and deliver productivity and unmatched experiences for all employees and customers at scale.

Salesforce will be introducing Industry Toolkit at Dreamforce; Industry Toolkit contains more than 100 ready-to-use AI-powered actions; Industry Toolkit can be used with Agentforce 

At Dreamforce, we’re excited to share our new AI tool kit — industry toolkit, which features more than 100 ready-to-use customizable AI-powered actions. All of these actions can be applied to build industry-specific agents with Agentforce.

Salesforce’s management wants to see 1 billion AI agents by FY2026; there are already 200 million agents identified in trials

I’ll just give you my own personal goals. So I’m not giving any guidance here. My goal is that by the end of fiscal year ’26 that we will have 1 billion agents. Already in just looking at the number of consumers identified just in the trials that we have going on, we have like 100 million identified or more. Okay. call it, 200 million. But the funny thing is, of course, it’s only 1 agent. But let’s just think it’s like a manifestation of all these agents talking to all these consumers.

Salesforce already has a long history of selling non-human consumption-based products; with AI agents, management sees pricing on a consumption basis or on a per conversation basis (at $2 per conversation); management thinks AI agents is a very high-margin opportunity

On pricing. When you think about — when you think about apps and you think about humans, because humans use apps, not in all cases. So for example, the Data Cloud is a consumption product. The Commerce Cloud is a consumption product. Of course, the e-mail product, Marketing Cloud is a consumption product. Heroku is a consumption product. So of course, we’ve had non-human consumption-based products for quite a long time at Salesforce…

…When we look at pricing, it will be on a consumption basis. And when we think about that, we think about saying to our customers, and we have, it’s about $2 per conversation. So, that is kind of how we think about it, that we’re going to have a lot of agents out there, even though it’s only 1 agent. It’s a very high margin opportunity, as you can imagine, and we’re going to be reaching — look, you have to think about these agents are like, this is the new website. This is your new phone number. This is how your customers are going to be connecting with you in this new way, and we’re going to be helping our customers to manage these conversations. And it’s probably a per conversational charge as a good way to look at it or we’re selling additional consumption credits like we do with our data cloud. 

Veeva Systems (NYSE: VEEV)

Veeva’s management is seeing the company’s customers appreciate the patience they have displayed in adopting AI; customers started using Veeva’s Vault Direct Data API for AI use cases in 2024 Q2; Vault Direct Data API provides data access 100 times faster than traditional APIs; management thinks that the advantage of providing API access for AI use cases is the possibility of partners developing use cases that management could not even forsee; customers have to pay a fee to turn on Vault Direct Data API and the fee is for covering Veeva’s compute costs; there’s no heavy implementation tasks needed for Vault Direct Data API

When groundbreaking technology like GenAI is first released, it takes time for things to settle and become clearer. That’s starting to happen now. Customers have appreciated our taking the long view on AI and our orientation to tangible value rather than hype. In Q2, our first early customers started using the Vault Direct Data API to power AI and other use cases. The ability to retrieve data 100 times faster than traditional APIs is a major software platform innovation and will be a big enabler of AI that uses data from Vault applications…

… When you make an API like the Direct Data API, you don’t know the innovation you’re unleashing. And that’s the whole point because the data can be consumed so fast and transactionally accurately, use cases that weren’t practical before can become practical. I mean if I step back way back when to designing the first salesforce.com API, I knew it was going to unleash a lot of innovation, and you just don’t know. It’s not predictable, and that’s the good thing…

…[Question] Looking at Vault Direct Data API, how seamless is it for customers to turn it on and start using it? Is it something that needs an implementation? 

[Answer] That is something that’s purchased by the customer, so that is something that is not free for the customers to use. They purchase it. The fee is not that large. It covers our compute cost, that type of thing… 

…After that, no, there’s no implementation. You turn it on, and it’s on. And that’s that.

Veeva’s AI Partner Program is progressing well and has seen 30 AI use cases being developed by 10 partners across Veeva Development Cloud and Veeva Commercial Cloud; the AI use cases in Veeva Commercial Cloud are mostly related to data science while the use cases in Veeva Development Cloud are mostly related to generation of documents and reports; management does not want to compete with the partners that are in the AI Partner Program 

Our AI Partner Program is also progressing well. We now have more than 10 AI partners supporting roughly 30 use cases across R&D and Commercial. We also continue to explore additional AI application opportunities beyond our current AI solutions…

… [Question] You talked about some of the early traction you’re seeing with the AI Partner Program. Can you maybe talk about what are some of the use cases you’ve seen so far?

[Answer] The types of use cases in commercial often have to do with data science. So things like next best action, dynamic targeting, pre-call planning, things like that. And then in R&D, they can be more things like document generation, generate a clinical study report or doing specific medical coding, things like that. So those are the type of use cases…

…In terms of us monitoring that and informing our own road map, I guess there may be some of that. But mostly, that type of innovation really comes from internally our own thinking with our customers. We don’t want to really disrupt our partners, especially when the partners are having customer success. If there’s a major use case that we’re very clear that customers need and for some reason, the ecosystem is not delivering customer success, yes, maybe we might step in there. But I would guess that what we would do would be more holistic, I guess, in some sense and not specifically something a partner would tackle because we’re generally going to have more resources and more ability to sway our own road map than a partner would, and we want to be respectful to the ecosystem.

Zoom Video Communications (NASDAQ: ZM)

Zoom’s management is seeing customers seeking out the AI capabilities of Zoom’s Contact Center packages; Zoom’s management saw the ASP (average selling price) for its Contact Center product double sequentially because of the product’s AI-tier, which comes with higher pricing

We are seeing increased adoption of our advanced Contact Center packages, as customers seek to utilize our AI capabilities to enhance agent performance…

…If you remember, we started with one pricing tier. We eventually added two more and the AI agent is like that Eric was speaking about earlier, is in the highest tier. We actually saw our ASPs for Contact Center almost double quarter-over-quarter because it’s such a premium feature. And when I look at the Q2 deals, the majority of them were purchasing in one of the top 2 tiers, so all of that is contributing to what I would say is not only expansion in terms of seat count but expansion in terms of value being derived from the product.

Zoom’s AI companion uses generative AI to produce meeting summaries, live translations, image generation and more; Zoom AI Companion is now enabled on over 1.2 million accounts; management wIll be upgrading AI companion as Zoom transitions into the 2.0 phase of AI-enabled work; customers really like ZoomAI Companion; Zoom AI Companion is provided at no additional cost; in Zoom meetings, the No.1 use case of Zoom AI Companion is to create meeting summaries; management is constantly improving the quality of Zoom AI Companion’s meeting summaries; customers are giving positive feedback on Zoom AI companion

Today, AI Companion enhances an employee’s capabilities using generative AI to boost productivity through features like meeting summary, chat compose, image generation, live translation and enhanced features in Contact Center. As these features have grown in popularity, we are happy to share that Zoom AI Companion is now enabled on over 1.2 million accounts…

…Our progress broadening Zoom Workplace, building out enhanced AI tools for Contact Center and amassing a large base of AI users sets us up well to transition into the 2.0 phase of AI-enabled work. In this phase, Zoom AI Companion will move beyond enhancing skills to simplifying your workday, providing contextual insights, and performing tasks on your behalf. It will do this by operating across our collaboration platform to ensure your day is interconnected and productive…

…Our customers really like Zoom AI Companion. First of all, it works so well. Secondly, at no additional cost, not like some of other vendors who got to charge the customer a lot. And in our case, this is a part of our package…

… You take a Meeting, for example, right? For sure, the #1 use case like a meeting summary, right? And we keep improving that quality like in the [indiscernible] and or meeting summary are getting better and better. Like in July, we had another upgrade quarter-wise, even better than previous deliveries, right?..

… [Question] One question I had is when you’re looking at Zoom AI Companion, we’ve heard a lot of great things in the field if customers kind of comparing that to other products that are offered out there. Can you kind of remind us about how you guys think about tracking success with the product internally, given that you don’t kind of charge for it directly beyond having millions of people using it?

[Answer] The metrics that we’ve been talking about on here is account activation. So looking at how many — it’s not individual users, it’s actual customer accounts that have activated it… And also they share the stories like how Zoom AI Companion like is very accurate summary, action items are helping their employees’ productivity as well. And yes, a lot of very positive feedback about adopting Zoom AI Companion.

Zoom’s management has intention to monetise AI services for the Contact Center product, but not for Zoom Workplace

[Question] Now that you’re seeing more adoption, Kelly, of Zoom Companion, how do you think about the cost of providing these generative AI features and capabilities? And do you think Zoom could eventually charge on a usage basis for power users of the generally just trying to weigh cost versus revenue opportunities here?

[Answer] I mean when we launched AI Companion, right? So we do not want to charge the customer. However, that’s for the workplace for the business services like a Contact Center, all those new offerings. And I think for sure, we are going to monetize. As I mentioned in the previous earnings calls, new — new solutions or the billing services, AI, I think we are going to charge. They are AI Companion, right? But the workplace and our core you see offering and collaboration offering we do not want to charge. I want to see — I really appreciate our AI team’s great effort, right? And focus on the quality, focus on the cost reduction and so on and forth.

AI services are a drag on Zoom’s margins at the moment (as the company is providing a lot of AI services for free now) but management sees them as important investments for growth

[Question] Just on gross margins, like the impact of generative AI and maybe what you can do to alleviate some of that off there.

[Answer] I mean we’re guiding to 79% for this year, which we will, reflects the prioritization of AI, but also the very strong discipline that we continue to apply. And we are holding to our long-term target for gross margins of 80%. But of course, we think at this point in time, it’s very important to prioritize these investments as they really set us up for future growth.

Zoom’s dev ops team is saving costs for the company to make room for more AI investments

I also want to give a credit to our dev ops team. On the right hand, for sure, we are going to buy more and more GPUs, right? And also leverage that. Our team tried to save the money from other areas, fully automated, and so on and so forth, right? So that’s another way for us to save the cost, right, to make some room for AI.

The regulatory environment for AI in the US and Europe has so far had very little impact on Zoom’s business because Zoom’s management has been adamant and clear that it is not going to use customer’s data to train its AI models
[Question] Are you seeing anything in the broad sweep of AI regulation in the U.S. or Europe that you think can dampen innovation?

[Answer] That’s the reason why we launch AI Companion, we already mentioned, we are not going to use any of our customer data to train our AI models, right? And we take customers data very, very seriously, right? And as a customer, they know that they trust our brand and trust of what we’re doing. And so far, I do not see any impact in terms of like regulation. And again, this AI is moving rapidly, right? So almost the EMEA here and we all look at the potential regulation. But so far, impact actually to us, to our business, I think it’s extremely limited. So like meeting summary, and it’s a very important feature, customer like that. I think we do not use our customer data to train our AI model. And so why not keep using the feature? I think there’s no impact so far.

---

Disclaimer: The Good Investors is the personal investing blog of two simple guys who are passionate about educating Singaporeans about stock market investing. By using this Site, you specifically agree that none of the information provided constitutes financial, investment, or other professional advice. It is only intended to provide education. Speak with a professional before making important decisions about your money, your professional life, or even your personal life. I have a vested interest in Adobe, MongoDB, Salesforce, Veeva Systems, and Zoom Video Communications. Holdings are subject to change at any time.