Nvidia shares set another record this week, lifting the AI chipmaker’s market capitalization to almost $6 trillion, as its graphics processors remain the most sought-after hardware in artificial intelligence and buyers work through a widening menu of ways to obtain them.
Management expects $108 billion in revenue for the October quarter, which would amount to an 89% jump from the same period a year earlier.
Cloud infrastructure providers have led Nvidia’s customer list for several years, but that base is broadening. A filing shows five clients each represented at least 10% of accounts receivable in the July quarter, up from three in January.
Industry research firm SemiAnalysis counted 323 providers of Nvidia GPUs as of September, compared with 209 fewer than 11 months before.
“You’re going to see a whole new crop of really, really exciting neoclouds with hundreds of billions of dollars backlog together,” Nvidia CEO Jensen Huang said last month at a Goldman Sachs technology conference in San Francisco.
Hyperscalers
Large enterprises spend tens of millions of dollars a year on a mix of cloud services from Amazon, Google and Microsoft, and since ChatGPT arrived in 2022 they have turned increasingly to those providers for GPUs to run generative AI workloads. Reputation is part of the draw — a software company sourcing chips and other capabilities from Amazon and Microsoft avoids awkward questions about its suppliers.
“When you’re talking to enterprises, your subprocessor had better be Azure,” said Bindu Reddy, CEO of AI assistant startup Abacus, referring to Microsoft’s cloud infrastructure.
Over the past year, leading AI labs Anthropic and OpenAI have committed more than $500 billion combined to Amazon and Microsoft, which Gartner estimates controlled 59% of the cloud infrastructure market in 2025.
“Hyperscalers are in a good position to show trust to the enterprises because of their 10-plus years of full-stack capabilities,” said Gartner analyst Hardeep Singh, though he added that hyperscalers do not always hold as many GPUs as enterprises need.
Amazon CEO Andy Jassy told analysts in July the company will not be able to serve all the demand it expects this year, and said: “I believe this dynamic will also be true in 2027.”
Neoclouds
If the hyperscalers were sufficient, neoclouds would not be multiplying.
Modal, a startup that runs virtual sandboxes where AI agents operate independently of main IT environments, began on the hyperscalers, then signed with the big neoclouds, and now uses 25 of them, according to CEO Erik Bernhardsson.
“You can get a few hundred GPUs or maybe a thousand, but at our scale, we needed way more GPUs,” he said.
The hyperscalers are chasing the neoclouds in turn. Google and Microsoft have started tapping CoreWeave even as they compete with one another.
“Some of the hyperscalers have approached us about taking care of customers they’re worried about because they don’t have the ability to service those customers when they need it,” said Marc Boroditsky, chief revenue officer of Nebius, a Netherlands-based neocloud with U.S. operations.
Video generation startup Reactor runs on Nebius and hyperscalers, CEO Alberto Taiuti said. Data center location matters, he said, because Reactor wants user-created videos to appear immediately; Nebius supplies the specific GPUs Reactor requires, solid customer service and adequate hardware and software at a good price.
The best-known neoclouds can ask for upfront payment, and chips may not come online for months, Bernhardsson said, because providers raise funding against contracts and then build out data center equipment.
Handing 10,000 GPUs to a new customer on a day’s notice would be difficult, said Chen Goldberg, an executive vice president at CoreWeave, whose CEO Mike Intrator said on the company’s August earnings call that near-term capacity remains essentially sold out.
Smaller and regional providers
Companies that need GPUs immediately may have to look past the marquee names. Some neoclouds stay out of the spotlight because they target particular countries, which can work in certain cases.
“Capacity right now is tight, and your relationships with your suppliers is actually one of the most closely guarded secrets for companies like ours,” said Zhen Lu, CEO of Runpod.
Specialist neoclouds can offer greater flexibility than larger GPU clouds, which often demand upfront payments and long-term commitments. Some sell bare-metal GPUs, which hand customers more control while leaving them more technical work to manage.
Buyers of these smaller services share the same questions: when will the GPUs arrive, and at what price? Sunny Smith, co-founder and technology chief at Massed Compute, said customers are often willing to commit to capacity when they expect prices to rise.
Oracle’s bring-your-own-hardware route
Oracle, one of the world’s largest cloud providers, lets clients supply their own GPUs. The software maker carries more debt than Amazon or Microsoft and has a lower credit rating, leaving it less room to spend heavily on chips, but it is willing to operate the technology.
“As we’re generally able to preserve and improve margins in the case of things like bring-your-own-hardware, the ROIC for those types of structures will be even higher,” Oracle CFO Hilary Maxson told analysts on a June earnings call, using the acronym for return on invested capital.
Oracle has not disclosed which companies take that path. Guggenheim Securities analyst John DiFucci, who recommends buying Oracle shares, said it would make sense for Advanced Micro Devices and Nvidia, the two largest GPU producers, to bring their own chips there.
Early-stage startups with limited capital can instead borrow GPUs for hours at a time through clouds at lower cost. For businesses with heavy computing needs, Oracle’s new route may beat building entire data centers. OpenAI committed to spending more than $300 billion with Oracle over five years, though it has not mentioned bringing in GPUs.
The approach may suit companies with the capital to buy AI chips but insufficient power, data center space or skilled labor — three things Oracle, like its hyperscaler peers, works hard to secure in healthy quantities.
Large deals with GPU owners
Another emerging option is striking sizeable agreements with firms holding truckloads of GPUs for rent.
SpaceX arranged to turn over excess capacity in separate deals with hyperscaler Google and open-source startup Reflection. In April, SpaceX agreed to provide GPUs to Cursor and then bought the AI coding startup outright for $60 billion. In May, SpaceX landed a deal to rent GPUs to Anthropic for $1.25 billion each month through mid-2029 — more than most startups can afford.
“The current economics have translated into a less than one-year payback on our new capital deployments for compute,” SpaceX finance chief Bret Johnsen told analysts in August.
SpaceX is not alone. CNBC reported in July that Meta was working to form a cloud unit that could sell AI computing power.
The on-premises option
Meanwhile, companies keep installing GPU-filled servers in their own data centers the old-fashioned way, as chief executives balance capability against cost control.
Revenue nearly doubled in the enterprise and small and medium business portions of hardware maker Lenovo’s Infrastructure Solutions Group during the June quarter. “We’re seeing more and more enterprises now starting to say, ‘How do I bring AI into my four walls?’” said Vlad Rozanovich, a senior vice president.
The hourly spot price for an Nvidia B200 GPU has more than doubled since March, according to data from Ornn, a startup that maintains indexes.
Dropbox, the collaboration software maker, relies on GPUs in its own data centers, CEO Ashraf Alkarmi said. “If we want to do a lot more, I think our supply chain connections will still be beneficial and a structural advantage,” he said.
Everpure, which sells data center storage hardware and software, has acquired its own GPUs to run open-weight AI models for its software engineers, said CEO Charlie Giancarlo. “In a very dynamic pricing environment, it’s always good to have multiple sources that you can go to,” he said.
COMMENTS