Jensen Huang sees another massive opening for Nvidia. The CEO told investors this week that a new processor called Vera unlocks a fresh $200 billion market the company has never touched. Every major hyperscaler and system builder has signed on to deploy it. The claim comes as Nvidia reported another blowout quarter and as talk of agentic systems moves from lab experiments to production plans.
Huang made the statement during the company’s fiscal first-quarter earnings discussion. He described Vera as the world’s first CPU built from the ground up for agentic AI. Traditional cloud CPUs focus on running many instances of applications in parallel. Vera instead optimizes for rapid token processing. Agents rely on that speed to act, use tools, and make decisions in sequence.
“Vera opens a brand new $200 billion TAM for Nvidia, a market we have never addressed before, and every major hyperscaler and system maker is partnering with us to deploy it,” Huang said, according to TechCrunch. “The world is rebuilding computing for agentic AI and robotic physical AI. Nvidia sits at the center of these transitions.”
The numbers already show traction. Nvidia expects nearly $20 billion in standalone CPU revenue this year. That figure covers Vera sales alone and does not include systems that bundle the chip with upcoming Rubin GPUs. Shipments are set to ramp in the third quarter. Analysts heard clear signals that the product will add to growth rather than simply replace existing revenue streams.
Agentic AI requires a different balance of compute. Training and initial inference still lean on GPUs. Yet the follow-on work — the reasoning loops, tool calls, memory lookups, and multi-step planning — runs largely on CPUs. Huang expects the world to move from one billion human users to billions of autonomous agents. Each will behave like a digital worker with its own set of digital tools. “We’re going to need a lot more CPUs,” he added.
Vera builds on lessons from Nvidia’s earlier Grace CPU but goes further. It uses custom Arm cores co-designed with the Rubin GPU and NVLink fabric. The result, according to figures shared on the call, delivers up to 1.5 times faster performance per core, twice the performance per watt, and four times the rack density compared with x86 alternatives. Early tests at AI labs and enterprises back those gains. Agent sandboxes run 50 percent faster. Enterprise data queries finish up to three times quicker.
Wall Street has watched Nvidia’s every move for signs of saturation. Data-center revenue continues to climb. The company posted $81.6 billion in total revenue for the quarter and guided to $91 billion next. Yet competition in CPUs looks more crowded than in GPUs. Intel, AMD, and large cloud operators have their own silicon road maps. Huang’s assertion that this represents virgin territory attempts to reframe the discussion.
And the timing matters. Hyperscalers have committed hundreds of billions to AI infrastructure in 2026 alone. Recent reports put combined spending by Amazon, Microsoft, Google, and Meta above $700 billion. Much of that budget has gone toward training clusters. The shift toward agentic workloads could redirect a sizable share into inference-optimized CPUs. Nvidia wants to capture part of that reallocation.
Huang has made big market-size predictions before. At earlier events he projected $1 trillion in orders for Blackwell and Vera Rubin systems through 2027. Those forecasts have largely held as adoption accelerated. Investors give him latitude because past projections translated into record sales. Fiscal 2026 revenue reached $215.9 billion, up 65 percent from the prior year.
But skepticism lingers in some corners. CPU markets have long been dominated by specialists. Entering now means displacing incumbents in sockets already filled with x86 or other Arm designs. Nvidia counters that agentic demands differ enough to justify a purpose-built part. The chip’s tight integration with NVLink and the full Nvidia software stack creates a performance edge that general-purpose CPUs struggle to match.
Recent coverage reinforces the momentum. HPC Wire reported that Vera has landed at leading AI labs as demand for agentic systems grows. The article details early deployments focused on scaling autonomous workflows that combine language models with external tools and memory systems. Performance gains in those environments have encouraged broader testing.
Nvidia also highlighted enterprise use cases. At Dell Technologies World, company executives noted that data queries run markedly faster on Vera. The partnership with Dell, one of the system makers Huang referenced, includes plans for integrated racks aimed at corporate AI deployments. Similar collaborations exist across the hyperscaler ecosystem.
The broader context includes power and efficiency pressures. Agentic AI can require orders of magnitude more compute than simple generative tasks. Huang has said in other forums that compute needs for these systems have risen at least 1,000 percent in two years. That escalation makes efficiency per watt and per rack critical. Vera’s claimed advantages target exactly those constraints.
Still, execution risks remain. Production must scale without delays. Software tools for agent development need to mature. And customers must decide how quickly to shift budgets from GPU-heavy training to more balanced inference and agent platforms. Nvidia’s visibility into $20 billion of CPU revenue suggests many have already decided.
Huang’s confidence rests on a simple observation. The transition to agentic AI mirrors the earlier move to accelerated computing. Once organizations see the productivity lift from autonomous agents, they tend to expand usage rapidly. Each new agent becomes another consumer of tokens, memory bandwidth, and CPU cycles. “The world has a billion users, human users,” he said. “My sense is that the world is going to have billions of agents.”
That vision, if realized, would reshape data-center architecture. CPUs would no longer serve as mere support actors. They would handle the bulk of real-time decision loops that define agent behavior. Nvidia’s entry turns a peripheral market into a strategic one. The $200 billion figure represents the total addressable spend on CPUs suitable for these workloads over time.
Whether the company captures a dominant share will depend on delivery and competition. For now, the early orders and partner commitments give Huang’s prediction weight. Nvidia reported the quarter’s results hours before the market reacted. Shares traded mixed in after-hours action, reflecting both excitement over the new opportunity and the high expectations already baked into the valuation.
The coming quarters will test the thesis. If Vera ramps as projected and hyperscalers integrate it at scale, the CPU business could become a meaningful new leg of growth. If integration proves slower or alternatives close the performance gap, the $200 billion opportunity may prove more contested than advertised. Huang has bet before on tectonic shifts in computing. So far, the record shows those bets have paid off.
WebProNews is an iEntry Publication