$120 billion. That is what Nvidia pocketed in net profit for its 2026 fiscal year, a figure so large it exceeds the annual economic output of most countries. For a company that once made its name selling graphics cards to PC gamers, the transformation is staggering. And the most remarkable part of this story is what is not in those numbers: not a single cent from China.
The GPU maker reported full-year revenues of $215.9bn, up 65% on the prior year, according to Nvidia's official results release. The fourth quarter alone delivered a record $68.1bn in revenue, beating analyst expectations by roughly $3bn. Its datacenter business, which sells the AI accelerator chips that power everything from large language models to corporate automation tools, contributed $62.3bn of that quarterly total. Gaming, professional visualisation, and automotive combined barely registered against those figures.
Looking ahead, Nvidia is guiding for $78bn in first-quarter fiscal 2027 revenues, implying close to 90% year-on-year growth. The Register reports that hyperscale customers — Meta, Google, Amazon, and Microsoft — plan to collectively invest $635bn in datacenters and AI infrastructure this year alone, giving Nvidia a deep and durable order book.
Here's the thing: that $78bn forecast contains zero assumed revenue from China's datacenters. Nvidia CFO Colette Kress was explicit on the analyst call. Small quantities of H200 chips for Chinese customers had received US government approval, she said, but Nvidia had yet to generate any revenue from them, and there was no certainty Beijing would grant import clearance.
The diplomatic back-story is complex. In early December 2025, the Trump administration reversed an export ban that had blocked sales of Nvidia's H200 accelerator to China, in exchange for a 25% revenue share flowing to the US Treasury. The H200, a 2023-vintage chip based on Nvidia's Hopper architecture, is not the company's most advanced product — its newer Blackwell and forthcoming Rubin platforms remain off the table for Chinese buyers — but it still significantly outperforms anything domestically produced in China. Despite Washington's green light, CNBC reported at the time that Beijing was widely expected to limit access, driven by both national security concerns and a strategic push toward chip self-sufficiency.
That prediction has held. Kress told analysts the danger of a technology decoupling was real. Chinese competitors, she argued, bolstered by recent IPOs and domestic investment, are making progress and could disrupt the global AI industry's structure over the long term. Her argument — that America must remain the platform of choice for developers everywhere, including in China — reflects a view shared by rivals including AMD and major partners such as Microsoft, CoreWeave, and OpenAI.
Those arguing for engagement with China have a legitimate point. Locking Nvidia out of the world's second-largest economy risks accelerating the very domestic chip development that US policymakers are trying to prevent. The Council on Foreign Relations has noted that years of export restrictions have prompted Chinese firms to intensify investment in homegrown alternatives, with Huawei's Ascend chip line making tangible progress. The harder Washington squeezes, the stronger the incentive for Beijing to build its own supply chain. That is a long-term strategic risk no balance sheet can fully hedge against.
At the same time, the scale of Nvidia's current success shows the China question is, for now, almost academic. The company returned $41.1bn to shareholders through buybacks and dividends during fiscal 2026 and still holds $58.5bn remaining under its repurchase authorisation. Its gross margin sits at 75%, a figure almost unheard of in hardware manufacturing. In real terms, that means Nvidia keeps 75 cents of every revenue dollar before operating costs. For a firm making physical chips dependent on global supply chains, that is an extraordinary position.
The quarter also produced an unusual sideshow. CEO Jensen Huang fielded analyst questions about orbital datacenters, a concept gaining traction among some in the technology sector. Huang acknowledged interesting AI applications in space — satellite imaging was his primary example — but was candid about the economics: poor for now, with heat dissipation through conduction rather than airflow presenting genuine engineering challenges. His caution was echoed by a Gartner analyst who, in a report this week, described orbital datacenters as "peak insanity" on economic grounds.
The real story, though, remains on terra firma. Nvidia's numbers reveal an AI infrastructure build-out with genuine momentum, driven by the world's largest technology companies committing capital at a scale that even sceptics find difficult to dismiss as a bubble. Whether Chinese revenues eventually materialise or not, the demand from American and allied hyperscalers appears sufficient to sustain extraordinary growth trajectories for the foreseeable future. The China stalemate is a risk to manage, not a crisis to weather. At $120bn in annual profit, Nvidia can afford to wait.