Nvidia wants to be the Wintel of robots. The $20 billion question is whether it can.

Jensen Huang has a second act, and it runs on CPUs.

The Nvidia CEO told analysts on May 20 that his company's new Vera processor opens a $200 billion total addressable market — a market he says Nvidia has never touched before. The number landed in headlines. What landed in checkbooks was something else.

Nvidia has already secured $20 billion in standalone Vera CPU bookings for fiscal year 2027, before the chip has reached broad commercial availability. Oracle Cloud Infrastructure has committed to deploying hundreds of thousands of Vera CPUs beginning this year — the first hyperscaler to commit to Vera at scale. Ian Buck, Nvidia's VP of Hyperscale and High-Performance Computing, hand-delivered the first units to Anthropic in San Francisco, OpenAI in Mission Bay, and SpaceXAI in Palo Alto in mid-May, followed by OCI in Santa Clara.

That deployment cadence is the real story. The $200 billion number is a CEO's framing.

"The world has a billion human users", Huang told analysts. "My sense is that the world is going to have billions of agents. We are going to need a lot more CPUs." Vera, he argued, is purpose-built for exactly that workload — not training large models, but running the orchestration layer, tool calls, agent sandboxes, and long-context retrieval operations that agentic AI actually requires at inference time.

The chip itself is a departure for Nvidia. Vera packs 88 custom Olympus cores — Nvidia's own design — with 1.2 terabytes per second of memory bandwidth and what the company claims is 50 percent faster per-core performance than traditional server CPUs under sustained load. Nvidia developed it in part using technology licensed from Groq, a specialized inference startup, in a deal reportedly worth around $17 billion. The full Vera Rubin platform, pairing Vera CPUs with Rubin GPUs in an NVLink-connected system, arrives later this year.

The competitive context matters here. In the last quarter of 2025, Intel held roughly 60 percent of server CPU market share; AMD held 24.3 percent. Nvidia held approximately 5 to 6 percent. Those are not the numbers of a company that has already won the market it's claiming. The $200 billion TAM is a leadership opportunity, not a current position.

What's driving urgency is the inference shift. The chip industry's narrative has moved from who can train the biggest model to who can serve answers cheapest and fastest. For training, Nvidia's GPUs remain dominant. For inference — generating outputs at scale, in real time, for millions of concurrent users — custom silicon from Google, Amazon, and Microsoft is making inroads. TPUs, Trainium, and Maia are no longer research projects; they're production infrastructure at the companies spending a combined $700 billion on AI capital expenditures this year, up from roughly $400 billion in 2025.

Vera is Nvidia's answer to that threat. By owning the CPU layer that orchestrates agentic workloads, the company argues it stays relevant in inference the same way it dominated training. CFO Colette Kress put it plainly: the results are setting Nvidia up to become the world's leading CPU supplier.

The financials back the ambition, up to a point. Q1 FY2027 revenue came in at $81.62 billion, up 85 percent year-over-year, beating a $78.86 billion consensus. Q2 guidance of $91 billion also exceeded forecasts. Nvidia announced an $80 billion share buyback and raised its quarterly dividend from one cent to 25 cents per share — financial confidence moves that signal the company believes the cash flow from its GPU business will fund the CPU expansion.

But the stock fell 1.6 percent in extended trading after the earnings print. Analysts noticed. "Nvidia delivered another beat, but at this point that's essentially priced in", said Jacob Bourne, an eMarketer analyst. "The lingering question is whether it can convince investors the AI buildout has durability into 2027 and 2028", especially as the narrative shifts toward inference workloads and competing silicon. The market's skepticism was visible elsewhere: AMD's stock has more than doubled in the past quarter while Nvidia gained roughly 25 percent over the same period.

Nvidia's supply commitments tell their own story. They reached $119 billion in Q1, up from $95.2 billion the prior quarter — a jump that reflects both confidence in demand and anxiety about a global memory crunch. Huang acknowledged the constraint directly: "My sense is that we'll be supply-constrained through the entire life of Vera Rubin."

The Groq licensing deal is the least examined piece of this. $17 billion is a large number to attach to a technology licensing agreement, and whether it's structured as upfront capex, amortized IP payments, or something else determines how much of a drag it is on Vera's margins. Nvidia did not break out the accounting in the materials available.

What's clear is that the race is real. Every major hyperscaler and system maker is "partnering with us to deploy" Vera, in Huang's words — but the same hyperscalers are also building the custom silicon that could eventually replace it. Oracle's commitment gives Nvidia a proof point and a reference customer. The question is whether hundreds of thousands of units from one cloud provider constitutes a platform, or whether the real validation requires Amazon and Google to follow.

The $200 billion market exists if agentic AI scales to billions of agents running on dedicated CPU infrastructure. Nvidia has the checkbooks and the customer commitments to suggest it believes that future. Whether it arrives before Intel, AMD, or the hyperscalers' own chips capture the same workload is the thing to watch.