"NVIDIA Q1 Financial Report: Surging Demand for AI, Tight GPU Supply"

2024-05-24

During the first quarter earnings conference call, NVIDIA CEO Jensen Huang emphasized the rapid growth of startups using NVIDIA's accelerated computing platform for generative artificial intelligence (GenAI). Huang stated, "The number of GenAI startups has grown to an astonishing level, with approximately 15,000 to 20,000 companies from various fields such as multimedia, digital characters, design, productivity applications, and digital biology. The autonomous driving industry has also turned to NVIDIA to train end-to-end models and expand the operational range of autonomous vehicles - this list is truly remarkable." He further emphasized that the demand for NVIDIA GPUs is "extremely strong" as major companies adopt NVIDIA's CUDA software and Tensor Core architecture to bring AI applications to the market. Consumer internet companies, enterprises, cloud service providers, automotive companies, and healthcare organizations are all investing heavily in building "AI factories" based on thousands of NVIDIA GPUs. The CEO of NVIDIA stated that the shift towards generative AI is driving a "fundamental, full-stack computing platform transformation" as computing moves from information retrieval to generating intelligent outputs. Huang explained, "Computers can now generate contextually relevant intelligent answers. This will completely change the global computing stack, even for personal computers." To meet the surging demand, NVIDIA started shipping its H100 "Hopper" architecture GPU in the first quarter and announced the next-generation "Blackwell" platform, which offers AI training and inference speeds 4 to 30 times faster than Hopper. Over 100 Blackwell systems from major computer manufacturers will be launched this year to achieve widespread adoption. The CEO of NVIDIA stated that their end-to-end AI platform capabilities give them a significant competitive advantage in the rapidly growing AI workload. He expects the demand for Hopper, Blackwell, and future architectures to far exceed supply and continue into next year with the rise of the GenAI revolution. AI chip demand difficult to meet Despite reporting record-breaking revenue of $26 billion in Q1, NVIDIA stated that customer demand for AI workload GPUs far exceeds their supply capacity. "We are working at full throttle every day," Huang said regarding their efforts to fulfill orders. "Customers are putting tremendous pressure on us to deliver systems quickly and ensure they run smoothly." Huang pointed out that even with increased production of the new Blackwell architecture, the demand for NVIDIA's flagship product, the H100 GPU, will continue to exceed supply for some time. "The demand for H100 has been growing this quarter... We expect the demand to exceed supply for some time as we transition to H200 and then to Blackwell," he said. The CEO attributed this urgency to the competitive advantage gained by companies that are the first to introduce groundbreaking AI models and applications. "The next company to reach the next major milestone will announce a breakthrough AI technology, while the second company to follow may only announce achievements that are 0.3% better than the first," explained Huang. "Training time is crucial. Three months of training time early means everything." Therefore, Huang stated that cloud providers, enterprises, and AI startups all feel immense pressure to secure as much GPU capacity as possible to reach milestones before their competitors. He predicts that the supply shortage of NVIDIA's AI platform will continue until 2024. "The supply of Blackwell is severely lagging, and we expect the demand to continue exceeding supply even into next year," Huang said. Significant economic returns brought by NVIDIA GPUs for cloud AI hosts Huang detailed how cloud providers and other companies have achieved substantial financial benefits by hosting AI models on NVIDIA's accelerated computing platform. "For every $1 invested by cloud providers in NVIDIA's AI infrastructure, there is an opportunity to earn $5 in revenue through GPU instance hosting over four years," Huang said. He cited the example of a language model with 70 billion parameters using NVIDIA's latest H200 GPU. He claimed that a single server can generate 24,000 tokens per second and support 2,400 concurrent users. "This means that for every $1 spent on the current price of NVIDIA H200 servers (calculated per token), API providers (providing tokens) can generate $7 in revenue over four years," Huang said. Huang further pointed out that NVIDIA's continuous software improvements have significantly enhanced the inference performance of their GPU platform. In the latest quarter, through optimization, the performance of H100 has tripled, saving customers three times the cost. Huang firmly believes that these high returns on investment are driving the strong demand for NVIDIA products from cloud giants such as Amazon, Google, Meta, Microsoft, and Oracle, as they compete to offer AI capacity to attract developers. He believes that combined with NVIDIA's excellent software tools and ecosystem support, these economic factors make NVIDIA the preferred platform for deploying GenAI. NVIDIA actively promotes the application of Ethernet network technology in the AI field While NVIDIA is renowned for its GPUs, the company is also an important player in the data center networking field with its Infiniband technology. In the first quarter, NVIDIA reported strong year-on-year growth driven by the adoption of Infiniband. However, Huang emphasized that Ethernet presents a significant new opportunity for NVIDIA to expand AI computing into a broader market. In the first quarter, the company began shipping its Spectrum-X platform, optimized for AI workloads on Ethernet. "Spectrum-X opens up a whole new market for NVIDIA's networking business, enabling data centers that solely use Ethernet to accommodate large-scale AI," Huang said. "We expect Spectrum-X to become a billion-dollar product line within a year." Huang stated that NVIDIA is "fully committed to Ethernet" and will release a series of important Spectrum switch roadmaps to complement its Infiniband and NVLink interconnect technologies. This three-pronged network strategy will enable NVIDIA to target all goals, from single-node AI systems to large clusters. NVIDIA also started sampling its Spectrum-4 Ethernet switch, which offers 51.2 terabits per second, in this quarter. Huang stated that leading server manufacturers like Dell are adopting Spectrum-X, pushing NVIDIA's accelerated AI networking technology into the market. "If you invest in our architecture today, it will seamlessly enter more and more clouds and data centers, and everything will run smoothly," Huang assured. Data centers and gaming drive record-breaking Q1 performance NVIDIA achieved record-breaking revenue of $26 billion in the first quarter, an 18% increase from the previous quarter and a remarkable 262% increase from the same period last year, surpassing their forecast of $24 billion. The data center business was the main driver of growth, with revenue soaring to $22.6 billion, a 23% increase from the previous quarter and an astonishing 427% increase from the same period last year. CFO Colette Kress emphasized the remarkable growth in the data center sector: "Compute revenue has grown over five times from last year, and networking revenue has grown over three times. Strong data center growth has been driven by all customer types, with enterprises and consumer internet companies leading the way. Large cloud providers continue to drive strong growth as they deploy and expand NVIDIA's AI infrastructure." Gaming revenue reached $2.65 billion, an 8% decrease from the previous quarter but an 18% increase from the same period last year, in line with NVIDIA's expectations for seasonal decline. Kress noted, "The market response to GeForce RTX SUPER GPUs has been strong, and demand and channel inventory for the product line remain healthy." Professional visualization revenue was $427 million, an 8% decrease from the previous quarter but a 45% increase from the same period last year. Automotive revenue reached $329 million, a 17% increase from the previous quarter and an 11% increase from the same period last year. For the second quarter, NVIDIA expects revenue of approximately $28 billion, with a fluctuation of 2%, and anticipates continuous growth in revenue across all market platforms.