Nvidia’s recent earnings report has highlighted both the company’s robust growth in the artificial intelligence (AI) sector and the challenges posed by international trade policies, particularly concerning China. The latter, famous for its strong graphics processing units (GPUs), announced record $39.1 billion in data center revenue this quarter alone. This increase was driven almost entirely by the AI hype and the massive demand for AI functionality.
CEO Jensen Huang emphasized the central importance of Nvidia’s GPUs for AI model inference. At the heart of this process is the need to reliably and efficiently deliver AI models in real-time to millions of customers. He noted that today’s AI models require an order of magnitude more computing power than their predecessors. What we’re seeing right now is a really steep increase in inference demand,” he added. This dramatic surge in demand only serves to highlight the increasing dependence on Nvidia’s technology to significantly improve the performance of AI across a multitude of applications.
In a bid to satisfy this new demand, Nvidia is preparing the release of its beleaguered replacement chip, the Blackwell Ultra. The main attraction this new release brings, though, is some big-memory and big-performance chops. That serves the growing needs of the big four cloud providers – Microsoft Azure, Google Cloud, Oracle Cloud Infrastructure and Amazon Web Services – that together account for nearly 50% of Nvidia’s data center revenue.
Huang focused on the rapidly evolving nature of AI computing. He said that today’s models need “a hundred, a thousand times more,” computing power compared to the “one-shot” method used by most models that came before, like ChatGPT in 2022. He articulated that these AI systems are not just answering your questions but are able to reason in more complex and powerful ways. It’s kind of like human-like reasoning, it’s really reasoning to itself, it’s decomposing a problem in ways that are human,” Huang added.
Even as Nvidia grapples with mounting competition, the U.S.’s own export controls are hurting its attempts to deny China access to cutting-edge semiconductor technology. According to Huang, the idea that China is physically unable to make AI chips is misguided at its core. The threat surrounding chips and AI. The U.S. has operated under a strategic assumption that China is unable to produce AI chips. That’s always been a dubious assumption, and now it’s clearly proven wrong, he said. This erroneous assumption threatens to push AI talent into the hands of Chinese competitors such as Huawei, who have their own chips under development and deployment.
Nvidia’s CEO admitted that the company does not have an alternative chip ready for the lucrative Chinese market. He underscored the fact that the company is looking everywhere for new product opportunities to make up this gap. Huang stressed the long-term ramifications of such export controls. This might force Chinese developers to develop their own solutions rather than relying on Nvidia’s technologies. On Tuesday, JPMorgan analyst Harlan Sur joined in, calling the export controls misguided.
Even with these difficulties, analysts are hopeful about Nvidia’s prospects moving forward. Bernstein’s Stacy Rason noted that the “general outlook and environment overall seems very encouraging” for Nvidia. The company is heads up on innovation and committed to product development. This strategic move ensures it continues to outpace competitors, like Google and Microsoft, in the rapidly evolving AI realm. Sur pointed out Nvidia’s unsurpassed technological lead. He put it, “Team has 1-2 step advantage over competitors with their silicon, hardware, and software platforms.”
At the same time, AI models are advancing quickly, requiring increasingly more advanced outputs. With every new model, Huang explained, OpenAI, Microsoft, and Google are increasing progress in their token generation infrastructure. OpenAI, Microsoft, and Google (though he refused to acknowledge the latter) are enjoying a step-function increase in token generation, said Zengler. Modern AI systems are now capable of utilizing various tools and resources to produce results, including reading documents and analyzing multimedia content.
Looking to the future, Huang reiterated the critical need for the United States to lead the world in technology. Intel’s former CEO Brian Krzanich expressed this sentiment in a single sentence: “The AI race isn’t about chips.” That’s really about the question of which stack the world runs on. As that stack expands quickly to include 6G and quantum, U.S. global infrastructure leadership is in the balance. As his remarks today demonstrate, technological advancement and economic competitiveness are inextricably linked.