Nvidia CEO Jensen Huang told reporters at Taipei’s Songshan Airport on Saturday that the company’s recently disclosed forecast of a $200 billion market for central processing units includes China, a statement that carries far more weight than the casual delivery suggested. According to a Reuters report carried by KFGO, Huang was asked directly whether the CPU market opportunity assumes access to Chinese customers and answered, “I would think so.” For a company whose share price is the single largest driver of the S&P 500’s recent performance, the simple admission that Beijing remains a baseline assumption in Nvidia’s planning is one of the most consequential pieces of investor communication of the year.

The Vera CPU and its companion Rubin GPU sit at the center of this story. Together they form the Vera Rubin platform, which Nvidia disclosed earlier this week is now ramping into production at Taiwan Semiconductor Manufacturing Company. Huang told reporters the second half of 2026 will be “a very busy” period for Taiwan’s supply chain. That single sentence is the operational backbone behind a forecast that, if it materializes, would lift Nvidia’s addressable revenue base by an entire trillion-dollar arithmetic step over the next two years.

Nvidia’s previous expansion was almost entirely a GPU story. The Vera architecture marks the first time the company is making a serious claim on the broader CPU market, the chips that handle the orchestration logic, memory addressing, and general-purpose compute that pair with GPUs in modern data centers. By owning both halves of the bundle, Nvidia is no longer just selling AI accelerators. It is selling the rack. And the rack is what hyperscale customers actually buy.

The H200 Licensing Wrinkle

The China dimension is more delicate than Huang’s casual answer implied. Nvidia has received U.S. export licenses to ship its H200 chip, the second-most powerful AI chip in its current lineup, to roughly ten Chinese firms. Not a single delivery has yet occurred. The bottleneck is no longer on the American side. It is on the Chinese side, where the central government has been actively steering domestic buyers toward home-grown alternatives from Huawei, Cambricon, and Biren. The talks between President Donald Trump and President Xi Jinping in Beijing this month, which Huang attended as part of the U.S. delegation, produced no immediate breakthrough.

“H200 has been licensed to ship to China. It would be terrific to be able to serve that market. The Chinese market is very important. It’s very large, of course,” Huang said in Taipei.

The diplomatic subtext is worth unpacking. The Trump administration is using high-end AI chip exports as leverage in a broader trade negotiation that includes agriculture, rare earths, and pharmaceutical inputs. For Nvidia, this means the company’s most lucrative export channel is now part of a geopolitical instrument that the company itself does not control. Huang’s strategy has been to talk publicly about Chinese demand, line up American policy support, and structure new products so they are non-negotiable for any serious data-center buyer worldwide. The Vera Rubin platform is the technical execution of that strategy.

For context on the broader chip-sector dynamics that this announcement plugs into, see our earlier coverage of Nvidia’s $40 billion in 2026 AI supply chain investments and the Cerebras IPO at a $48 billion valuation as an Nvidia challenger.

Why CPUs Matter in the Agentic AI Era

The reason Nvidia is suddenly talking about CPUs the same way it has talked about GPUs is the rise of agentic AI. Large language models that simply generate text in response to a prompt rely heavily on GPUs for training and inference, with relatively modest CPU loads. Agentic AI systems, which take autonomous action across tools, memory, and long task horizons, are CPU-intensive in a way the previous generation of AI workloads was not. They require orchestration logic, tool routing, state management, error handling, and memory addressing that CPUs do better than GPUs. As enterprises move from chatbots to autonomous agents that book travel, run procurement, or operate code repositories, the CPU portion of every AI deployment grows.

Huang has read that shift correctly. The Vera CPU is purpose-built to pair with Rubin GPUs in agentic workloads, with shared memory architectures and high-bandwidth interconnects that reduce the latency penalties of mixing the two chip classes in older bundles. This is exactly what AMD has been trying to do with its EPYC plus Instinct combinations, but Nvidia is arriving with a more integrated software stack and a much larger installed customer base of GPU-anchored enterprises.

The $200 billion CPU figure also reframes the competitive landscape. Intel and AMD have spent the last decade competing in data-center CPUs, with Intel ceding share to AMD and AMD then giving ground in adjacent AI accelerators to Nvidia. The Vera launch puts Nvidia directly into the same revenue pool that Intel and AMD have been fighting over. If even a quarter of that $200 billion accrues to Nvidia, it would erode Intel’s already weakened data-center franchise and squeeze AMD’s CPU growth at the same time the company is trying to scale its AI accelerator business. For an early read on the AMD response, see our coverage of AMD’s massive forecast change as the stock roared 15 percent on earnings.

The Taiwan Supply Chain Premium

Huang’s Taipei visit is timed ahead of next month’s Computex trade show and is itself a signal about where the next year of capex will be aimed. AMD, in a parallel move announced Thursday, said it will invest more than $10 billion in Taiwan’s AI sector to deepen strategic partnerships and expand its capacity to build and assemble advanced AI chips. Asked whether Nvidia had a similar investment plan, Huang said the company had not made public announcements but emphasized that Nvidia has “invested in and supported our partners here far more than that.”

That posture matters because TSMC, which manufactures the most advanced semiconductors powering the AI trend, is the single point of leverage for the entire industry. Huang said he would meet with TSMC executives during the trip. Reading between the lines, those meetings are about advanced node allocation for the Vera Rubin platform, packaging capacity at TSMC’s CoWoS and SoIC facilities, and the schedule for ramping output to meet the second-half tempo Huang has now publicly committed to.

Taiwan’s supply chain has been the indispensable hub for AI hardware since 2023. Nvidia’s announcement effectively books a meaningful share of that hub’s capacity through the end of 2027. For competitors, the implication is uncomfortable. Capacity at TSMC’s leading-edge nodes is the binding constraint on the entire AI hardware industry, and Nvidia is securing it.

For investors thinking about how China’s domestic economic stress interacts with this story, our earlier reporting on China’s economy and the deflation crisis and the South Korea chip war provides essential context for the regional politics that will shape Nvidia’s export environment.

The Super Micro Smuggling Question

There is a darker thread in Huang’s Taipei comments that deserves attention. Taiwanese prosecutors said on Thursday they were investigating three people suspected of illegally exporting high-end AI servers made by Super Micro and containing Nvidia chips subject to U.S. export controls. In March, the U.S. Justice Department charged three Super Micro associates, including its co-founder, with helping smuggle at least $2.5 billion of U.S. AI technology into China in violation of export laws.

Asked what more Nvidia could do to prevent the diversion of chips, Huang told reporters his company is “very rigorous” in explaining laws and regulations to its partners and insists they comply with applicable rules. “Ultimately, Super Micro has to run their own company. I hope that they will enhance and improve their regulation compliance and avoid that from happening in the future,” he said.

The legal and reputational risk for Nvidia is contained but real. The Justice Department has so far focused on Super Micro and resellers rather than Nvidia itself. But the broader policy environment around chip exports is hardening, and any future incident that draws Nvidia’s name into the indictments would have material consequences for both the stock and the company’s export licensing position. Huang’s public emphasis on partner compliance is a signal that Nvidia is documenting its compliance posture aggressively in anticipation of further investigations.

What the $1 Trillion AI Chip Forecast Implies

The same Wednesday earnings call where Huang disclosed the Vera CPU’s $200 billion market also set Nvidia’s projected revenue from flagship AI chips at $1 trillion. That figure pencils out only if a meaningful share of the world’s data-center capex over the next three to five years routes through Nvidia hardware. Hyperscalers, sovereign AI initiatives in the Gulf, India, Japan, and Western Europe, and a fast-growing enterprise tier all need to keep buying for that forecast to mature.

The forecast also assumes that China remains accessible in some form. If U.S.-China relations deteriorate sharply, the H200 channel could be permanently lost. Nvidia’s ability to absorb that loss depends on whether non-China demand can grow fast enough to fill the gap, which is precisely the bet sovereign-AI deals in the Gulf and Japan are designed to support. Nvidia’s strategy is, in effect, to make itself indispensable to every government and hyperscaler outside China so completely that a China shutdown is unpleasant rather than catastrophic.

Three Indicators to Watch

Three signals will tell investors whether the Vera Rubin and CPU strategy is hitting the targets Huang has set. The first is TSMC capacity allocation, which will show up in quarterly capex commentary from TSMC and in any joint statements about node allocation. The second is hyperscaler purchase orders, which usually leak through hyperscaler capex guidance two to three months ahead of public Nvidia revenue disclosures. The third is H200 actual deliveries to China, which would mark the first proof that the China channel is convertible into revenue rather than just licensed potential.

If all three signals turn positive over the next two quarters, the $200 billion CPU figure will look conservative. If even one fails, the share price will need to absorb a revision lower in the second half of 2026.

FAQ

What is Nvidia's Vera Rubin platform?

The Vera Rubin platform combines Nvidia’s new Vera central processing unit with its Rubin graphics processing unit architecture into an integrated AI compute system. The combination is designed for high-performance data-center workloads, particularly emerging agentic AI applications that require both heavy GPU compute and substantial CPU orchestration capacity.

Why does Nvidia want access to the China market so badly?

China is the world’s second-largest semiconductor market and historically a major source of data-center demand. Even with U.S. export controls limiting which chips Nvidia can sell, the Chinese market represents tens of billions of dollars in annual revenue opportunity. Losing it permanently would force Nvidia to depend heavily on Western and Gulf customers to maintain projected growth.

What are H200 chips and why can't they be shipped freely to China?

The H200 is Nvidia’s second-most powerful AI chip, just below the flagship Blackwell-class accelerators. The U.S. government requires export licenses for H200 sales to China to limit the country’s access to advanced AI compute. Nvidia has received licenses for about ten Chinese firms but has not made deliveries because Chinese regulators are encouraging domestic alternatives.

How does the CPU market opportunity compare to Nvidia's GPU business?

The $200 billion CPU market is roughly one-fifth the size of the $1 trillion projected AI chip revenue Nvidia has flagged from its GPU-led flagship products. CPUs would not replace GPU revenue but would diversify Nvidia’s data-center footprint into a category currently dominated by Intel and AMD, giving Nvidia a fuller share of every rack it ships.

What is agentic AI and why does it favor CPUs more than older AI workloads?

Agentic AI describes systems that perform autonomous, multi-step tasks rather than just generating a single response to a prompt. These workloads require heavy orchestration, memory state management, and tool calls, all of which run better on CPUs than GPUs. As enterprises move from chatbots to autonomous agents, CPU demand inside AI deployments scales meaningfully higher.

How might the Super Micro smuggling case affect Nvidia?

The Department of Justice has so far targeted Super Micro and individual resellers, not Nvidia itself. The risk is that future investigations or congressional scrutiny could broaden, potentially affecting Nvidia’s export licensing posture. Huang’s public emphasis on partner compliance is designed to insulate Nvidia from being drawn into related cases.