Nvidia's Compute Leasing Model Gathers Pace as Google Commits $920M Monthly and Healthcare AI Beckons

Nvidia quietly pivots to leasing massive AI factories, with deals worth billions from Anthropic, Google, and SpaceX, while expanding into healthcare AI and authorizing an $80B buyback.

Nvidia is quietly morphing from a chip supplier into a provider of vast, leased computing infrastructure — a shift that is rewriting the economics of artificial intelligence. The company no longer just sells graphics processors; it is building and renting out entire AI factories. In Australia, Sharon AI Holdings recently locked in a long-term deal for a 72-megawatt cluster housing tens of thousands of the latest GPUs. South Korea’s SK Group is planning an even bigger installation with over 50,000 chips by the end of 2027. And the scale of these arrangements is drawing in some of the deepest pockets in tech.

The most striking evidence comes from the leasing cascade already underway. Anthropic is reportedly paying more than $1.2 billion each month for access to Nvidia’s Colossus cluster. Starting in autumn 2026, Google will add $920 million per month to Nvidia’s recurring revenue in exchange for roughly 110,000 processors. SpaceX, too, is funneling massive leasing capacity through its own networks. These are no longer one-off hardware sales — they are long-term service contracts that resemble utility bills more than purchase orders.

Alongside this infrastructure push, Nvidia is planting flags in an entirely different sector. The company has partnered with Abridge, a health-tech firm that has raised about $830 million, to build a clinical AI model based on the open-source Nemotron architecture. The model will analyze and document conversations between doctors and patients, aiming to ease the administrative burden in healthcare. Nvidia’s venture arm, NVentures, holds a stake in Abridge, and the broader Clara platform is designed to bind the medical industry to Nvidia’s CUDA ecosystem — a moat that pure hardware rivals will find hard to cross.

The financial engine behind these ambitions is running at full throttle. In the first quarter of fiscal 2027, Nvidia posted record revenue of $81.6 billion, an 85% jump from a year earlier. The data-center segment alone contributed $75.2 billion, fueled by the ramp of the Blackwell-300 family. Yet the stock has stalled. At Friday’s close of €177.28, the shares sit precisely on their 50-day moving average. Over the past month the stock has shed about 8.5%, and it remains well below the 52-week high of €202.50. Analysts see this as a healthy pause: the average price target stands at €258.25 (roughly $299), with 59 analysts in consensus.

Nvidia is also deploying its massive free cash flow to support the stock. An $80 billion share buyback program has been authorized, and the company paid its regular quarterly dividend in early June. The message to the market is that management sees the current valuation as compelling, especially given the product-cycle acceleration ahead. Production of the next-generation "Vera Rubin" platform is slated to start in the third quarter of 2026. The new Vera CPU is also designed to target the Chinese market, a move that, if it can navigate strict export controls, would cement Nvidia’s role as the backbone of an even broader industrial revolution.

Physical AI is another frontier. Since early June, Nvidia has been working with South Korea’s LG Group to bring its Isaac and GR00T platforms into household robots and autonomous vehicles. The bottleneck here is energy: the latest Blackwell systems can process twenty times more AI agents per megawatt than the previous generation, an efficiency gain that is becoming critical as global data-center spending is forecast to hit at least $630 billion in 2026. Whether power constraints, geopolitical headwinds, or the sheer complexity of scaling will slow Nvidia’s trajectory remains an open question — but the company is betting on every front that the answer is no.