NVIDIA Corp (NVDA) — Q2 2024 Earnings Call Transcript

Thank you. Good afternoon, everyone and welcome to NVIDIA's conference call for the second quarter of fiscal 2024. With me today from NVIDIA are Jensen Huang, President and Chief Executive Officer; and Colette Kress, Executive Vice President and Chief Financial Officer. I'd like to remind you that our call is being webcast live on NVIDIA's Investor Relations website. The webcast will be available for replay until the conference call to discuss our financial results for the third quarter of fiscal 2024. The content of today's call is NVIDIA's property. It can't be reproduced or transcribed without our prior written consent. During this call, we may make forward-looking statements based on current expectations. These are subject to a number of significant risks and uncertainties, and our actual results may differ materially. For a discussion of factors that could affect our future financial results and business, please refer to the disclosure in today's earnings release, our most recent Forms 10-K and 10-Q and the reports that we may file on Form 8-K with the Securities and Exchange Commission. All our statements are made as of today, August 23, 2023, based on information currently available to us. Except as required by law, we assume no obligation to update any such statements. During this call, we will discuss non-GAAP financial measures. You can find a reconciliation of these non-GAAP financial measures to GAAP financial measures in our CFO commentary, which is posted on our website. And with that, let me turn the call over to Colette.

Thanks, Simona. We had an exceptional quarter. Record Q2 revenue of $13.51 billion was up 88% sequentially and up 101% year-on-year, and above our outlook of $11 billion. Let me first start with Data Center. Record revenue of $10.32 billion was up 141% sequentially and up 171% year-on-year. Data Center compute revenue nearly tripled year-on-year, driven primarily by accelerating demand from cloud service providers and large consumer Internet companies for HGX platform, the engine of generative AI and large language models. Major companies, including AWS, Google Cloud, Meta, Microsoft Azure and Oracle Cloud as well as a growing number of GPU cloud providers are deploying, in volume, HGX systems based on our Hopper and Ampere architecture Tensor Core GPUs. Networking revenue almost doubled year-on-year, driven by our end-to-end InfiniBand networking platform, the gold standard for AI. There is tremendous demand for NVIDIA accelerated computing and AI platforms. Our supply partners have been exceptional in ramping capacity to support our needs. Our data center supply chain, including HGX with 35,000 parts and highly complex networking has been built up over the past decade. We have also developed and qualified additional capacity and suppliers for key steps in the manufacturing process such as advanced packaging. We expect supply to increase each quarter through next year. By geography, data center growth was strongest in the U.S. as customers direct their capital investments to AI and accelerated computing. China demand was within the historical range of 20% to 25% of our Data Center revenue, including compute and networking solutions. At this time, let me take a moment to address recent reports on the potential for increased regulations on our exports to China. We believe the current regulation is achieving the intended results. Given the strength of demand for our products worldwide, we do not anticipate that additional export restrictions on our Data Center GPUs, if adopted, would have an immediate material impact to our financial results. However, over the long term, restrictions prohibiting the sale of our Data Center GPUs to China, if implemented, will result in a permanent loss of opportunity for the U.S. industry to compete and lead in one of the world's largest markets. Our cloud service providers drove exceptionally strong demand for HGX systems in the quarter, as they undertake a generational transition to upgrade their data center infrastructure for the new era of accelerated computing and AI. The NVIDIA HGX platform is the culmination of nearly two decades of full-stack innovation across silicon, systems, interconnects, networking, software and algorithms. Instances powered by the NVIDIA H100 Tensor Core GPUs are now generally available at AWS, Microsoft Azure and several GPU cloud providers, with others on the way shortly. Consumer Internet companies also drove very strong demand. Their investments in data center infrastructure purpose-built for AI are already generating significant returns. For example, Meta recently highlighted that since launching Reels, AI recommendations have driven a more than 24% increase in time spent on Instagram. Enterprises are also racing to deploy generative AI, driving strong consumption of NVIDIA powered instances in the cloud as well as demand for on-premise infrastructure. Whether we serve customers in the cloud or on-prem through partners or direct, their applications can run seamlessly on NVIDIA AI enterprise software with access to our acceleration libraries, pre-trained models and APIs. We announced a partnership with Snowflake to provide enterprises with an accelerated path to create customized generative AI applications using their own proprietary data, all securely within the Snowflake Data Cloud. With the NVIDIA NeMo platform for developing large language models, enterprises will be able to make custom LLMs for advanced AI services, including chatbots, search and summarization, right from the Snowflake Data Cloud. Virtually every industry can benefit from generative AI. For example, AI Copilot such as those just announced by Microsoft can boost the productivity of over 1 billion office workers and tens of millions of software engineers. Billions of professionals in legal services, sales, customer support and education will be able to leverage AI systems trained in their field. AI Copilot and assistants are set to create new multi-hundred billion dollar market opportunities for our customers. We are seeing some of the earliest applications of generative AI in marketing, media and entertainment. WPP, the world's largest marketing and communication services organization, is developing a content engine using NVIDIA Omniverse to enable artists and designers to integrate generative AI into 3D content creation. WPP designers can create images from text prompts while responsibly trained generative AI tools and content from NVIDIA partners such as Adobe and Getty Images using NVIDIA Picasso, a foundry for custom generative AI models for visual design. Visual content provider Shutterstock is also using NVIDIA Picasso to build tools and services that enable users to create 3D scene backgrounds with the help of generative AI. We've partnered with ServiceNow and Accenture to launch the AI Lighthouse program, fast-tracking the development of enterprise AI capabilities. AI Lighthouse unites the ServiceNow enterprise automation platform and engine with NVIDIA accelerated computing and with Accenture consulting and deployment services. We are also collaborating with Hugging Face to simplify the creation of new and custom AI models for enterprises. Hugging Face will offer a new service for enterprises to train and tune advanced AI models powered by NVIDIA HGX cloud. And just yesterday, VMware and NVIDIA announced a major new enterprise offering called VMware Private AI Foundation with NVIDIA, a fully integrated platform featuring AI software and accelerated computing from NVIDIA with multi-cloud software for enterprises running VMware. VMware's hundreds of thousands of enterprise customers will have access to the infrastructure and AI and cloud management software needed to customize models and run generative AI applications such as intelligent chatbots, assistants, search and summarization. We also announced new NVIDIA AI enterprise-ready servers featuring the new NVIDIA L40S GPU built for the industry standard data center server ecosystem and BlueField-3 DPU data center infrastructure processor. L40S is not limited by supply and is shipping to the world's leading server system makers. L40S is a universal data center processor designed for high volume data center delivery to accelerate the most compute-intensive applications, including AI training and inference through design, visualization, video processing and NVIDIA Omniverse industrial digitalization. NVIDIA AI enterprise-ready servers are fully optimized for VMware, Cloud Foundation and Private AI Foundation. Nearly 100 configurations of NVIDIA AI enterprise-ready servers will soon be available from the world's leading enterprise IT computing companies, including Dell, HP and Lenovo. The GH200 Grace Hopper Superchip, which combines our ARM-based Grace CPU with Hopper GPU, entered full production and will be available this quarter in OEM servers. It is also shipping to multiple supercomputing customers, including Atmos, National Labs and the Swiss National Computing Center. And NVIDIA and SoftBank are collaborating on a platform based on GH200 for generative AI and 5G/6G applications. The second generation version of our Grace Hopper Superchip with the latest HBM3e memory will be available in Q2 of calendar 2024. We announced the DGX GH200, a new class of large memory AI supercomputer for large AI language models, recommendation systems and data analytics. This is the first use of the new NVIDIA switch system, enabling all of its 256 Grace Hopper Superchips to work together as one, a huge jump compared to our prior generation connecting just eight GPUs over a switch. DGX GH200 systems are expected to be available by the end of the year, with Google Cloud, Meta and Microsoft among the first to gain access. Strong networking growth was driven primarily by InfiniBand infrastructure to connect HGX GPU systems. Thanks to its end-to-end optimization and in-network computing capabilities, InfiniBand delivers more than double the performance of traditional Ethernet for AI. For billions of dollar AI infrastructures, the value from the increased throughput of InfiniBand is worth hundreds of millions of dollars and pays for the network. In addition, only InfiniBand can scale to hundreds of thousands of GPUs. It is the network of choice for leading AI practitioners. For Ethernet-based cloud data centers that seek to optimize their AI performance, we announced NVIDIA Spectrum-X, an accelerated networking platform designed to optimize Ethernet for AI workloads. Spectrum-X couples the Spectrum or Ethernet switch with the BlueField-3 DPU, achieving 1.5x better overall AI performance and power efficiency compared to traditional Ethernet. BlueField-3 DPU is a major success. It is in qualification with major OEMs and ramping across multiple CSPs and consumer Internet companies. Now moving to gaming. Gaming revenue of $2.49 billion was up 11% sequentially and 22% year-on-year. Growth was fueled by GeForce RTX 40 Series GPUs for laptops and desktops. End customer demand was solid and consistent with seasonality. We believe global end demand has returned to growth after last year's slowdown. We have a large upgrade opportunity ahead of us. Just 47% of our installed base have upgraded to RTX and about 20% of the GPU with an RTX 3060 or higher performance. Laptop GPUs posted strong growth in the key back-to-school season, led by RTX 4060 GPUs. NVIDIA's GPU-powered laptops have gained in popularity, and their shipments are now outpacing desktop GPUs from several regions around the world. This is likely to shift the reality of our overall gaming revenue a bit, with Q2 and Q3 as the stronger quarters of the year, reflecting the back-to-school and holiday build schedules for laptops. In desktop, we launched the GeForce RTX 4060 and the GeForce RTX 4060 TI GPUs, bringing the Ada Lovelace architecture down to price points as low as $299. The ecosystem of RTX and DLSS games continues to expand. 35 new games were added to DLSS support, including blockbusters such as Diablo IV and Baldur’s Gate 3. There's now over 330 RTX accelerated games and apps. We are bringing generative AI to gaming. At COMPUTEX, we announced NVIDIA Avatar Cloud Engine or ACE for games, a custom AI model foundry service. Developers can use this service to bring intelligence to non-player characters. And it harnesses a number of NVIDIA Omniverse and AI technologies, including NeMo, Riva and Audio2Face. Now moving to Professional Visualization. Revenue of $375 million was up 28% sequentially and down 24% year-on-year. The Ada architecture ramp drove strong growth in Q2, rolling out initially in laptop workstations with a refresh of desktop workstations coming in Q3. These will include powerful new RTX systems with up to 4 NVIDIA RTX 6000 GPUs, providing more than 5,800 teraflops of AI performance and 192 gigabytes of GPU memory. They can be configured with NVIDIA AI enterprise or NVIDIA Omniverse inside. We also announced three new desktop workstation GPUs based on the Ada generation. The NVIDIA RTX 5000, 4500 and 4000, offering up to 2x the RT core throughput and up to 2x faster AI training performance compared to the previous generation. In addition to traditional workloads such as 3D design and content creation, new workloads in generative AI, large language model development and data science are expanding the opportunity in professional visualization for our RTX technology. One of the key themes in Jensen's keynote at the recent event was the conversion of graphics and AI. This is where NVIDIA Omniverse is positioned. Omniverse is OpenUSD's native platform. OpenUSD is a universal interchange that is quickly becoming the standard for the 3D world, much like HTML is the universal language for the 2D landscape. Together, Adobe, Apple, Autodesk, Pixar and NVIDIA form the Alliance for OpenUSD. Our mission is to accelerate OpenUSD's development and adoption. We announced new and upcoming Omniverse cloud APIs, including RunUSD and ChatUSD to bring generative AI to OpenUSD workload. Moving to automotive. Revenue was $253 million, down 15% sequentially and up 15% year-on-year. Solid year-on-year growth was driven by the ramp of self-driving platforms associated with a number of new energy vehicle makers. The sequential decline reflects lower overall automotive demand, particularly in China. We announced a partnership with MediaTek to bring drivers and passengers new experiences inside the car. MediaTek will develop automotive SoCs and integrate a new product line of NVIDIA's GPU chiplet. The partnership covers a wide range of vehicle segments from luxury to entry level. Moving to the rest of the P&L. GAAP gross margins expanded to 70.1% and non-GAAP gross margin to 71.2%, driven by higher data center sales. Our Data Center products include a significant amount of software and complexity, which is also helping drive our gross margin. Sequential GAAP operating expenses were up 6% and non-GAAP operating expenses were up 5%, primarily reflecting increased compensation and benefits. We returned approximately $3.4 billion to shareholders in the form of share repurchases and cash dividends. Our Board of Directors has just approved an additional $25 billion in stock repurchases to add to our remaining $4 billion of authorization as of the end of Q2. Let me turn to the outlook for the third quarter of fiscal 2024. Demand for our Data Center platform where AI is tremendous and broad-based across industries on customers. Our demand visibility extends into next year. Our supply over the next several quarters will continue to ramp as we lower cycle times and work with our supply partners to add capacity. Additionally, the new L40S GPU will help address the growing demand for many types of workloads from cloud to enterprise. For Q3, total revenue is expected to be $16 billion, plus or minus 2%. We expect sequential growth to be driven largely by Data Center with gaming and Professional Visualization also contributing. GAAP and non-GAAP gross margins are expected to be 71.5% and 72.5%, respectively, plus or minus 50 basis points. GAAP and non-GAAP operating expenses are expected to be approximately $2.95 billion and $2 billion, respectively. GAAP and non-GAAP other income and expenses are expected to be an income of approximately $100 million, excluding gains and losses from non-affiliated investments. GAAP and non-GAAP tax rates are expected to be 14.5%, plus or minus 1%, excluding any discrete items. Further financial details are included in the CFO commentary and other information available on our Investor Relations website. In closing, let me highlight some upcoming events for the financial community. We will attend the Jefferies Tech Summit on August 30 in Chicago, the Goldman Sachs Conference on September 5 in San Francisco, the Evercore Semiconductor Conference on September 6 as well as the Citi Tech Conference on September 7, both in New York. And the BofA Virtual AI conference on September 11. Our earnings call to discuss the results of our third quarter of fiscal 2024 is scheduled for Tuesday, November 21.

Thank you very much. Good afternoon. The results are impressive. Jensen, I have a question for you about the rapidly emerging application of large model inference. Most investors understand that you have secured a significant share of the training market. Previously, smaller model inference workloads were primarily handled by ASICs or CPUs. With the rise of large models like GPT, there's a new workload that is quickly gaining momentum in large model inference. I believe your Grace Hopper Superchip products and others are well-suited for this. Can you discuss how you perceive the differentiation between small model inference and large model inference, and how your product portfolio is set up for this? Thank you.

Yeah. Thanks a lot. So let's take a quick step back. These large language models are pretty phenomenal. They have the ability to understand unstructured language. But at their core, what they have learned is the structure of human language. And they have encoded, or compressed within them, a large amount of human knowledge that they have learned from the corpuses they studied. What happens is, you create these large language models and make them as large as you can, and then you derive from it smaller versions of the model, essentially teacher-student models. It's a process called distillation. And so when you see these smaller models, it's very likely they were derived from or distilled from or learned from larger models, just as you have professors and teachers and students. This process will continue going forward. Start from a very large model, which has a vast amount of generality and what's called zero-shot capability. For many applications or skills that you haven't specifically trained it on, these large language models can miraculously perform them. That's what makes them so magical. Conversely, you desire these capabilities in all kinds of computing devices, so you distill them down. These smaller models might have excellent capabilities on particular skills, but they don't generalize well or lack good zero-shot capabilities. They each hold unique strengths, but the foundation starts with large models.

Thank you. Just had a quick clarification and a question. Colette, if you could please clarify how much incremental supply you expect to come online in the next year? Do you think it's 20%, 30%, 40%, 50%? So just any sense of how much supply because you said it's growing every quarter. And then Jensen, my question for you is, when we look at overall hyperscaler spending, that buy is not really growing that much. So what gives you the confidence that they can continue to carve out more of that pie for generative AI? Just give us your sense of how sustainable is this demand as we look over the next one to two years? If I take your implied Q3 outlook of Data Center, $12 billion, $13 billion, what does that say about how many servers are already AI accelerated? Where is that going? So just give some confidence that the growth that you are seeing is sustainable into the next one to two years.

So thanks for that question regarding our supply. Yes, we do expect to continue increasing our supply over the next quarters as well as into next fiscal year. In terms of percent, it's not something that we have here. It is a work across so many different suppliers and so many different parts of building an HGX and many of our other new products that are coming to market. But we are very pleased with both the support that we have from our suppliers and the long time that we've spent with them improving their supply.

The world has approximately $1 trillion worth of data centers installed, in the cloud, in enterprises, and otherwise. That $1 trillion of data centers is in the process of transitioning into accelerated computing and generative AI. We're seeing two simultaneous platform shifts at the same time: one is accelerated computing, which is the most cost-effective, energy-efficient, and performant way of computing today. Then enabled by generative AI, which gives everyone two reasons to transition to a platform shift from general-purpose computing to this new way of doing computing. It's about $1 trillion worth of data centers; call it $0.25 trillion of capital spend each year. We're seeing data centers around the world diverting their capital spend on the two most important trends of computing today, accelerated computing and generative AI. This is not a near-term thing; it is a long-term industry transition.

Hi, guys. Thanks for taking my question. I was wondering, Colette, if you could tell me how much of Data Center in the quarter, maybe even the guide, is systems versus GPU, like DGX versus just the H100? I'm trying to understand how much pricing or content, however you want to define that, versus units is actually driving the growth going forward. Can you give us any color around that?

Sure, Stacy. Let me help. Within the quarter, our HGX systems were a very significant part of our Data Center, as well as our Data Center growth that we had seen. Those systems include our HGX of our Hopper architecture, but also our Ampere architecture. We are still selling both of these architectures in the market. When you think about that, it means both the systems as a unit are growing quite substantially, driving revenue increases. Both of these are the drivers of the revenue inside Data Center. Our DGXs are always a portion of additional systems that we will sell. They are significant opportunities for enterprise customers as well as many different types of customers we're seeing even in our consumer Internet companies. The importance there is also coming together with software that we sell with our DGXs, but that's a portion of our sales. The rest of the GPUs, we have new GPUs coming to market, and we talk about the L40S, which will add continued growth going forward. But, again, the largest driver of our revenue within this last quarter was definitely the HGX system.

And Stacy, if I could just add something. When you say H100, I know you have a mental image in your mind. But the H100 is composed of 35,000 parts, weighs 70 pounds, and contains nearly 1 trillion transistors combined. It takes a robot – actually many robots to build it because of its weight. It requires a supercomputer to test a supercomputer. So we often refer to the H100 as if it’s a chip that comes off of a fab, but H100s go out as HGXs sent to the world’s hyperscalers and are really, really quite large system components.

Hi. Thanks for taking my question and congrats on the success. Jensen, it seems like a key part of your success in the market is delivering the software ecosystem along with the chip and the hardware platform. I had a two-part question on this. I was wondering if you could help us understand the evolution of your software ecosystem, its critical elements, and if there’s a way to quantify your lead in this area in terms of person-years invested in building it. Part two, could you share your view on what percentage of the value of the NVIDIA platform is hardware differentiation versus software differentiation? Thank you.

Yeah, Mark, I appreciate the question. Let me use some metrics: we have a runtime called AI Enterprise; if you will, it’s the runtime that nearly every company uses for the end-to-end of machine learning from data processing, the training of any model on any framework, inference, and the deployment scaling out into a data center. It can scale out for deploying in hyperscale data centers or enterprise data centers, for instance, on VMware. We have hundreds of millions of GPUs in the field and millions of GPUs in the cloud across almost every single cloud. It runs in a single GPU configuration or multiple GPUs and multiple nodes throughout an entire data center scale. This runtime, called NVIDIA AI Enterprise, contains something like 4,500 software packages, software libraries, and something like 10,000 dependencies among each other. This runtime is continuously optimized for our installed base and our software stack. That’s just one example of the effort it takes to get accelerated computing to work. The combinations of code and applications are quite immense, and it has taken us two decades to reach this point. The elements of our company, if you will, are several. Number one is architecture: the flexibility, versatility, and performance of our architecture allows us to do all the things I just mentioned. The second characteristic of our company is the installed base and the third is our reach: we’re in the cloud today for public-facing services, which has attracted many developers and customers to use our platform, and we have a broad distribution from OEMs and ODMs. Then lastly, because of our scale and velocity, we were able to sustain this complex stack of software and hardware across all of these diverse environments. These capabilities allow our ecosystem to effectively build their business on top of us, making us special.

Hi. Thank you for taking my question. Great job on results and outlook. Colette, I have a question on the core L40S that you mentioned. Any idea how much of the supply tightness can L40S help with? And if you can talk about the incremental profitability or gross margin contribution from this product? Thank you.

Yeah, Atif. Let me take that for you. The L40S is really designed for a different set of applications. The H100 is designed for large-scale language models and processing very large models and a great deal of data. That’s not the focus of L40S. L40S is built to fine-tune models and it performs that remarkably well. It includes a transform engine and has a lot of performance. You can deploy multiple GPUs in a server. It's designed for hyperscale deployments, meaning it’s easy to install L40S servers in hyperscale data centers. It comes in a standard rack and standard server, making everything straightforward to set up. L40S also includes the software stack that accompanies it and all the partnerships we've established with VMware and other enterprise software producers. L40S is created for enterprises, and that is why HPE, Dell, and Lenovo, along with several other system makers, are constructing numerous configurations of enterprise servers with us to propel generative AI into the enterprise.

Great. Thank you. I guess the thing about these numbers that’s so remarkables is the amount of demand that remains unfulfilled. Talking to some of your customers, as impressive as these numbers are, you’ve more than tripled your revenue in several quarters. There’s demand in some cases for multiples of what people are currently receiving. Can you discuss how much unfulfilled demand you think exists? And you mentioned visibility extending into next year. Do you have a timeline for when you think we’ll reach supply-demand equilibrium?

Yeah. We have excellent visibility through this year and into next year. We're already planning the next-generation infrastructure with the leading CSPs and data center builders. The demand – the easiest way to think about it, the world is transitioning from general-purpose computing to accelerated computing. That’s the simplest way to conceptualize demand. The best method for companies to enhance their throughput, improve their energy efficiency, and increase cost efficiency is to allocate their capital budget to accelerated computing and generative AI. As firms recognize the tipping point and the beginning of this transition, they are shifting their capital investments to accelerated computing and generative AI. This is not driven by a singular application, but a broad-based new computing platform transition occurring worldwide. Data centers are responding to this shift.

Hi. Thank you for taking the question. I had one quick clarification question for Colette and another one for Jensen. Colette, I believe last quarter you mentioned that CSPs accounted for about 40% of your Data Center revenue, consumer Internet at 30%, and enterprise 30%. Based on your comments, it sounded like CSPs and consumer Internet may have been a larger portion of your business—if you could confirm or clarify that. And then Jensen, given your role as the main enabler of AI and your visibility into customer projects, how confident are you that sufficient applications or use cases will emerge for your customers to generate reasonable returns on their investments? I ask this because of concerns that demand might experience a slowdown in the coming years. Do you think there’s enough breadth and depth to sustain growth in your Data Center business?

Okay. Thank you, Toshiya, on the question regarding the types of customers we have in our Data Center business. We analyze this in terms of combining our compute and networking. Our large CSPs contributed a bit more than 50% of our revenue in Q2, with the next largest piece being from consumer Internet companies. Finally, the remaining segment is our enterprise and high-performance computing.

Toshi, I'm hesitant to project the future precisely, so I’ll base my response on fundamental principles of computer science. It recognizes that for a while now, general purpose computing isn't effective, and brute-forcing computation at scale is becoming inefficient. It is too energy costly, too expensive, and yields slower performance for applications. Accelerated computing is now recognized as the new standard. Generative AI turbocharges this transition. However, accelerated computing can also provide solutions for numerous existing applications in the data center. By leveraging this approach, companies can reduce CPU workloads and save significant capital and energy costs while increasing throughput. Many companies are converting their investments from general-purpose computing to generative AI and accelerated computing as they understand these truths. This is evident from emerging GPU-specific cloud providers like CoreWeave, which is flourishing, and regional GPU-specialized service models proliferating globally. Enterprises also recognize the need to transition. However, to do so, they must manage systems, security, and software-defined data centers, which require partnerships with established IT firms like VMware. Our recent collaboration with VMware, who effectively serve numerous enterprise customers, aims to facilitate their transition to this landscape.

Thanks a lot. Can you talk about the attach rate of your networking solutions to your compute shipments? For example, is half of your compute being shipped with your networking solutions, more than half, or less? Is this something you could potentially leverage to prioritize GPU allocation?

Well, starting with your last question, we don’t prioritize our GPU allocations based on that. We let customers choose their preferred networking options. For those building extensive infrastructures, InfiniBand is often the best choice due to its remarkable efficiency. An improvement of 10%, 15%, or 20% higher throughput for a billion-dollar infrastructure can lead to massive savings. If an infrastructure is chiefly dedicated to large language models or substantial AI systems, InfiniBand becomes an excellent choice. However, in diverse user environments where Ethernet is essential for data center management, we recently announced an excellent solution called Spectrum-X that helps optimize Ethernet for generative AI workloads, blending the efficacy of InfiniBand with Ethernet systems.

Hi. Good afternoon. Good evening. Thank you for the question. My inquiry pertains to DGX Cloud. Can you elaborate on its reception and current momentum? And Colette, could you provide insights into your software business? What is its current run rate and significance? It seems to be boosting margins as well.

DGX Cloud's strategy, let me start there. Its strategy is to achieve several things: Firstly, to foster a close partnership between us and leading CSPs. We acknowledge that many of our 30,000 clients worldwide are startups, with around 15,000 being AI startups; the most rapidly evolving segment is generative AI. They often prefer deployment in leading cloud environments, which is why we built DGX Cloud within several prominent clouds, facilitating seamless connection and integration for our AI partners. Secondly, it enhances performance by allowing our CSPs and ourselves to collaborate closely in optimizing hyperscale cloud performance, historically designed for multi-tenancy rather than high-performance distributed computing like generative AI. Additionally, our engineering relies significantly on DGX systems; self-driving projects, research, and generative AI refinements require extensive infrastructure. Thus, DGX Cloud serves different roles, and its success in the market has been remarkable. Our CSPs appreciate it, developers are enthusiastic, and our engineering teams are eager for more DGX Cloud.

Regarding our software revenue, remember that software is part of nearly all our products, from Data Center offerings to gaming and automotive. Yes, we’re also driving a standalone software business, which continues to grow as we provide services and upgrades across various fronts. Right now, we believe our software business runs in the hundreds of millions of dollars annually, and we anticipate NVIDIA AI enterprise will be included in many products, including our DGX systems and PCIe versions of H100. We expect more availability through our CSP marketplaces, indicating strong growth potential moving forward.