Nvidia delivered a very strong earnings report in every sense. Earnings per share grew by 111% year-over-year, while revenue increased by 106%. Pre-report expectations were exceeded by 6% for earnings per share and 4% for revenue. However, since the market is accustomed to such growth, the report was initially met with selling pressure.
The stock, which closed at $209, fell to $203 in the post-market trading before the earnings meeting began. However, during speeches by CFO Collette Kress and then CEO Jensen Huang, the stock surged by 10%, reaching $222, and mostly managed to maintain that level.
So what caused this 10% increase? What happened during the earnings meeting that caused such a sudden surge?
The critical shift in Nvidia's growth outlook is that demand is no longer solely coming from hyperscalers. Dominant AI projects, NeoClouds, AI startups, and enterprise customers currently account for roughly half of the business, and this segment is growing at approximately 100% annually. According to Jensen Huang, this market, excluding hyperscalers, could eventually grow larger than the current cloud economy.
This demand has already exceeded Nvidia's supply capacity. The company has projected approximately 70% growth next year, not because demand will increase by 70%, but because Nvidia's supply capacity can only grow by a maximum of 70%. In other words, we're talking about demand far exceeding supply.
Moreover, each new architecture increases Nvidia's economic share per data center: the revenue opportunity per gigawatt increases from approximately $18 billion at Hopper to $25 billion at Grace Blackwell and $40 billion at Vera Rubin. Therefore, growth isn't just about selling more GPUs, but about each new generation of AI data centers generating more revenue for Nvidia.
Huang stated that the next phase of demand is agentic AI, and that agentic AI already accounts for half of token consumption, emphasizing that this consumption will increase hundreds of times.
Regarding the question of whether NVDA is at risk given the large number of companies involved in chip manufacturing, Jensen clearly addressed these concerns. Customers building data centers can still assemble different components as they wish. However, most companies lack the expertise or willingness to do so. Therefore, we see a very large market for growth.
The development of custom chips by customers like OpenAI and Anthropic is not seen as a direct threat by Nvidia. According to Huang, while many of these chips are designed for specific service or inference workloads, Nvidia offers a platform that covers the entire AI lifecycle, from training to agentic inference.
