Same Old Nvidia: Beat, Raise, Trade Down
The stock shrugged, but the call showed AI moving from experimentation to industrialization
Two things remain reliably true in this unpredictable world: the sun rises in the east, and every quarter Nvidia beats earnings before somehow trading down.
The company delivered another absurd beat-and-raise. Revenue was $81.6B, up 85% year over year and 20% sequentially. Data Center revenue was $75.2B, up 92% year over year. Q2 guidance came in at $91B ±2%, despite assuming zero China Data Center compute revenue.
Jensen Huang’s summary was concise: “Demand has gone parabolic.” The market’s response: cute.
This is what happens when a company conditions the market to expect miracles. Eventually, walking on water is priced in. The selloff is about altitude, not weakness.
But the stock move is the least interesting part of the story. Here are the takeaways from the earnings call:
Nvidia’s new reporting structure is a map of the AI market. Nvidia is now splitting the business into Data Center and Edge Computing, with Data Center further divided into Hyperscale and ACIE: AI cloud, industrial, and enterprise. Hyperscale is the familiar world of public cloud and the largest internet platforms. ACIE is the messier, more interesting frontier: AI-native clouds, sovereign AI, enterprise AI factories, industrial deployments, and on-prem infrastructure.
Today the split is roughly 50/50: Hyperscale at $38B, ACIE at $37B. But ACIE grew faster sequentially, and Jensen suggested it could outgrow hyperscale over time. The AI infra market is shaping up to be more distributed, sovereign, verticalized, and operationally complex than the cloud market, where demand consolidated around a handful of hyperscalers.Nvidia’s sovereign AI business already grew more than 80% year over year, with deployments across nearly 40 countries. The number of partner data centers exceeding 10MW has nearly doubled in one year, now above 80 sites. The market has been trained to watch hyperscaler capex. Nvidia is telling us to widen the aperture.
AI demand is becoming economically self-reinforcing. Jensen’s core message was simple: compute capacity = revenue = profits. He emphasized: “Demand has gone parabolic. The reason is simple. Agentic AI has arrived. AI can now do productive and valuable work. Tokens are now profitable, so model makers are in a race to produce more.”
For two years, the AI compute buildout was a bet on future monetization. The capex was real; the revenue was speculative. That phase appears to be ending.
CPUs are back. One of the most interesting parts of the call was the Vera CPU discussion. Nvidia framed Vera, its purpose-built CPU for agentic AI, as a major new growth vector. Management said it has visibility into nearly $20B of standalone CPU revenue this year, opening what it characterizes as a $200B TAM that Nvidia historically has not addressed.
In the first phase of AI, the market obsessed over GPUs because the bottleneck was model training and inference. In the agentic phase, workloads become more system-like. An agent is a harness: orchestration, memory management, tool use, browser activity, code execution, I/O, security, and sub-agent coordination. Jensen’s framing: the thinking happens on GPUs, but much of the orchestration runs on CPUs.
That has big implications. If the world moves from a billion human users to billions of software agents, each agent effectively needs some compute environment in which to act. The CPU demand curve becomes tied not just to cloud workloads, but to the proliferation of agents themselves.Nvidia increases inference market share. For several years, the bear case has been that Nvidia would dominate training but lose share in inference as workloads shifted to cheaper, specialized silicon. The call pushed directly against that narrative.
Jensen said Nvidia is gaining share in inference, especially as it deepens partnerships with frontier model companies. The most notable example was Anthropic. Nvidia’s coverage of Anthropic had been “largely zero until recently” but it is now helping expand Anthropic compute capacity across AWS, Azure, CoreWeave, and others. That is incremental share gain in one of the most important frontier AI accounts.Physical AI age is coming. Nvidia’s current business is still overwhelmingly about data center AI, but management continues to point toward the next frontier: physical AI. Physical AI revenue surpassed $9B over the trailing twelve months. The category includes robotics, autonomous vehicles, industrial automation, medical devices, AI-RAN base stations, factories, and edge systems. This is Nvidia’s long-duration bet on AI moving from screens into the physical world.
The industry has spent the last few years automating knowledge work: text, code, images, research, support, analytics. The next phase is AI that acts: robots in warehouses, autonomous fleets, factory systems, telecom infrastructure, surgical devices, industrial inspection, and embodied agents.
The most revealing comment on the call may have been Jensen’s aside that Anthropic and OpenAI are “growing within one month what some SaaS companies took a decade to grow.” That sentence explains the whole AI trade.
The model companies are scaling at a speed software markets have never really seen. Their growth pulls in compute. Compute pulls in power, networking, memory, CPUs, data centers, and financing. And the wheel turns.
Most software companies scale with marginal cost approaching zero. AI companies scale with marginal cost approaching “please build another gigawatt-scale data center.” That is the magic and the problem.
The AI boom is a software revolution with an industrial cost structure and Nvidia is the arms dealer, the toll collector, and increasingly the factory architect.