In a moment that is more symbolic than material, Nvidia has resumed sending its H200 processors into China after a prolonged export embargo — yet the chips crossing that border represent less than one percent of the company's vast data-centre revenue. The larger story is one of constrained abundance: a company whose growth is limited not by what the world wants, but by how much it can build. As sovereign governments and enterprises join hyperscalers in the race for AI infrastructure, Nvidia finds itself at the center of a demand it cannot fully answer.
Nvidia ships H200s to China but forecasts zero China data-centre revenue
The unconstrained growth would be a lot higher
Mark
So Nvidia is shipping chips to China again, but it sounds like it barely matters to their business?
Mimi
Right—less than one per cent of their data-centre revenue in the last quarter. The licensing change is real, but it's more symbolic than substantial.
Mark
Why would they even bother announcing it if it's so small?
Mimi
Because the embargo was a big story. Ending it, even partially, signals something has shifted politically. But the company's own forecast tells you what they actually think: they're projecting zero China data-centre revenue going forward.
Mark
Zero? They're shipping there but forecasting nothing?
Mimi
It's their way of saying the licensing is too restrictive to build a real business around. They're not betting on China as a revenue driver.
Mark
So where is the real growth coming from?
Mimi
Everywhere else. Huang said actual demand is much higher than the 70 per cent growth they're forecasting—they're just limited by how many chips they can manufacture. And it's not just the big cloud companies anymore. Enterprises, smaller operators, governments building sovereign AI systems—that's where the untapped opportunity is.
Mark
So supply is the only thing holding them back?
Mimi
Completely. They have more orders than they can fill. China is a sideshow.
O Pulso
- Nvidia's return to the Chinese market, long anticipated, has arrived with a quiet anticlimax — sales so small the company excludes China entirely from its forward revenue projections.
- The real tension is not geopolitical but logistical: CEO Jensen Huang openly acknowledged that unconstrained demand would push growth far beyond the already striking 70% forecast for 2028.
- Supply chains, not sanctions or market saturation, are the binding constraint — Nvidia has secured significant manufacturing capacity and still says it needs considerably more.
- The customer base is fragmenting in ways that could reshape the AI industry: hyperscalers now account for only half of Nvidia's opportunity, with enterprises, smaller cloud operators, and government-backed sovereign AI projects filling the other half.
- The H200 shipments to China are technically real but strategically marginal — a door reopened under conditions so narrow it barely changes the room.
In a moment that is more symbolic than material, Nvidia has resumed sending its H200 processors into China after a prolonged export embargo — yet the chips crossing that border represent less than one percent of the company's vast data-centre revenue. The larger story is one of constrained abundance: a company whose growth is limited not by what the world wants, but by how much it can build. As sovereign governments and enterprises join hyperscalers in the race for AI infrastructure, Nvidia finds itself at the center of a demand it cannot fully answer.
Nvidia has resumed shipping H200 data-centre processors to China, ending a months-long embargo under US export controls. The shipments arrived under a newly revised licensing framework — but the moment carries a strange hollowness. Those China sales amounted to less than one percent of the $89 billion Nvidia earned from data-centre revenue in its second quarter, which closed on July 26.
The company's forward guidance makes the situation plain. Nvidia is projecting $108 billion in total revenue for the coming quarter, but that figure deliberately excludes any data-centre computing revenue from China. The licensing arrangement, in other words, reopens a door that remains mostly closed.
Elsewhere, the picture is one of strained abundance. Nvidia expects roughly 70 percent revenue growth in its 2028 financial year — a figure CEO Jensen Huang described not as ambitious but as supply-limited. On the earnings call, he was direct: actual customer demand sits well above what the company is publicly forecasting. "The unconstrained [growth] would be a lot higher," he said. Manufacturing capacity, not market appetite, is the ceiling.
Who is driving that demand has also shifted. Hyperscalers — Amazon, Google, Microsoft and their peers — now represent only about half of Nvidia's opportunity. The remainder is spreading across enterprises building private AI systems, smaller cloud operators, and sovereign AI projects funded by governments. This diversification signals that the AI infrastructure market is maturing well beyond the initial wave dominated by a handful of mega-companies.
The China situation sits awkwardly against this backdrop of global excess demand. Nvidia can ship certain chips there, but under conditions so constrained that the company sees no reason to project meaningful revenue from the region. The H200 shipments are real — and almost beside the point.
Nvidia has begun shipping its H200 data-centre processors into China again, marking the end of a months-long embargo under US export controls. The company confirmed the shipments arrived under a newly revised licensing framework that permits limited sales to the country. Yet the moment carries an odd hollowness: those China sales accounted for less than one per cent of the $89 billion the company pulled in from data-centre revenue during its second quarter, which closed on July 26.
The company's outlook for the coming quarter tells the fuller story. Nvidia is projecting $108 billion in total revenue, but that forecast contains a deliberate blank where China ought to be. The company is assuming zero data-centre computing revenue from the country going forward—a striking admission that whatever trickle of sales has resumed remains too marginal to factor into planning. The licensing arrangement, in other words, opens a door that remains mostly closed.
Yet Nvidia's leadership struck a notably confident tone about the broader picture. The company expects revenue to grow roughly 70 per cent during its 2028 financial year, a projection CEO Jensen Huang framed not as optimistic but as constrained by supply. In an earnings call on Wednesday, Huang made clear that the real ceiling on growth is not demand but manufacturing capacity. "The unconstrained [growth] would be a lot higher," he said, suggesting that actual customer appetite far outpaces what Nvidia can currently produce. He estimated true demand sits well above the 70 per cent figure the company is publicly forecasting.
The gap between what Nvidia can make and what the market wants has widened considerably. Huang noted that the company has secured substantial manufacturing capacity but still needs "a lot more" to meet orders. This supply crunch, he indicated, is the only thing preventing the company from promising even steeper growth. The constraint is not a problem to solve but a fact to manage—and manage profitably.
Where that demand is coming from has also shifted. Hyperscalers—the massive cloud providers like Amazon, Google, and Microsoft that have driven much of the recent AI boom—now represent only about half of Nvidia's opportunity, according to Huang. The other half is fragmenting across enterprises building their own AI systems, smaller cloud operators, and sovereign AI projects backed by governments. This diversification matters because it suggests the AI infrastructure market is maturing beyond the handful of mega-companies that dominated the initial wave of generative AI deployment.
The China situation sits awkwardly within this broader picture of abundance. The US government's export restrictions have effectively cordoned off what would otherwise be a massive market. Nvidia can now ship certain chips there, but under conditions so limiting that the company does not bother projecting meaningful revenue from the region. Meanwhile, everywhere else on the planet, the company faces the opposite problem: more demand than it can fulfill. The H200 shipments to China are real, but they are also almost beside the point.
Citações Notáveis
The unconstrained growth would be a lot higher— Jensen Huang, Nvidia CEO
Hyperscalers represented only about half of the opportunity, with the remainder coming from enterprises, neo clouds and sovereign AI projects— Jensen Huang, Nvidia CEO
Quer a matéria completa?
Leia o original em South China Morning Post ↗
Business