英伟达Q2 FY2027营收962亿美元超预期4%，Rubin平台确认为史上最快产品爬坡，盘后涨超5%

Nvidia has confirmed its first shipments of H200 data-centre processors to China under a new US licensing scheme, ending a months-long lockout. However, the sales accounted for less than 1 per cent of the company’s US$89 billion data-centre revenue in its second quarter, which ended on July 26. The H200 disclosure marked a shift from the previous quarter, when Nvidia said it had made no shipments of Data Centre Hopper products to China. Although Washington approved H200 exports in January, Beijing initially restricted purchases as it promoted home-grown alternatives, before allowing selected Chinese AI companies to buy limited quantities from last month. The US chip giant’s US$108 billion third-quarter revenue forecast continued to assume no data-centre computing revenue from China. Despite the muted contribution from China, Nvidia struck an upbeat tone on global demand. The company said it expected revenue to grow by about 70 per cent in its 2028 financial year, describing the forecast as supply constrained. CEO Jensen Huang said actual demand was “much greater than 70 per cent”, adding that Nvidia’s available supply allowed it to confidently forecast growth at that level. “The unconstrained [growth] would be a lot higher,” Huang said in an earnings call webcast on Wednesday, adding that the company had secured significant capacity but still needed “a lot more”. He said hyperscalers represented only about half of the opportunity, with the remainder coming from enterprises, neo clouds and sovereign AI projects. As demand surges, Nvidia is also taking a more active role in helping artificial intelligence companies finance the infrastructure needed to buy and deploy its chips, prompting concerns over so-called “circular financing”. “We know some will call this circular financing. We see it differently,” chief financial officer Colette Kress said during the company’s earnings call on Thursday morning in Asia after the US market close, arguing that Nvidia’s investments were backing companies with rapidly expanding demand whose growth was increasingly limited by access to computing infrastructure rather than customers. Nvidia had invested nearly US$50 billion in frontier AI laboratories and recently partnered with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR on financing platforms designed to mobilise more than US$500 billion of third-party capital for AI infrastructure, Kress said. We get paid twice, once on the hardware sale and again through the share of rental revenue Colette Kress, Nvidia chief financial officer The company had introduced revenue-sharing arrangements with AI cloud operators that saw it provide a take-or-pay commitment on part of a facility’s capacity and a minimum revenue guarantee that could help the operator secure financing. In return, Nvidia received a share of revenue generated above that threshold, she said. “We’re not making loans in this model,” Kress said. “We get paid twice, once on the hardware sale and again through the share of rental revenue.” The third-quarter revenue forecast followed another record quarter. Nvidia reported second-quarter revenue of US$96.2 billion, up 106 per cent from a year earlier and ahead of analysts’ average estimate of US$92.5 billion, according to Bloomberg. Revenue from AI clouds, industrial and enterprise customers jumped 138 per cent year on year to US$40.3 billion. Nvidia shares rose 4.71 per cent in after-hours trading following the results. Nvidia also disclosed a major expansion of its partnership with Amazon Web Services (AWS), which will deploy an additional 2 million Nvidia GPUs from the current quarter through to the second quarter of the 2029 financial year. AWS will also deploy Vera central processing units, including some integrated with Rubin systems. The next major growth driver is expected to be Nvidia’s Vera Rubin platform. Production shipments began early this month, and the company said it had already received purchase orders from every major hyperscaler, AI cloud provider and system manufacturer. Nvidia said it expected Rubin to account for about 20 per cent of data-centre revenue in the third quarter and expected it to mark the fastest product ramp-up in the company’s history. Supply, however, was expected to remain a bottleneck through to at least the end of the 2028 financial year. Nvidia said sharply higher memory prices would also put pressure on profitability, with gross margin expected to fall from 74 per cent in the third quarter to between 71 and 72 per cent in the fourth quarter before recovering to 72 to 73 per cent in the 2028 financial year as price increases took effect.