Nvidia GPU providers more than double to 323 as access routes fragment

Nvidia’s dominance in AI chips is creating a fragmented market for GPU access, with companies now choosing between hyperscalers, specialized neoclouds, direct purchases, and creative arrangements like SpaceX’s rental deals. Understanding these distribution channels matters for investors tracking Nvidia’s revenue concentration and for enterprises navigating a capacity-constrained market.

• Nvidia projects $108 billion in October quarter revenue, an 89% year-over-year increase driven by unrelenting AI demand.
• Five clients accounted for at least 10% of Nvidia’s accounts receivable in July, up from three in January, showing growing customer concentration.
• SemiAnalysis counted 323 Nvidia GPU providers as of September, up from 209 less than 11 months earlier, fragmenting the access landscape.

• $108B Nvidia’s projected October quarter revenue versus 89% year-over-year growth rate
• 323 Nvidia GPU providers counted as of September versus 209 providers in prior year
• 59% Hyperscalers’ cloud infrastructure market share in 2025 versus all other providers
• $1.25B SpaceX’s monthly GPU rental rate to Anthropic through mid-2029

According to CNBC reporting, demand for Nvidia’s processors has become so acute that companies now face a paradox: while the chips are everywhere, accessing them requires navigating a rapidly multiplying set of suppliers and business models. The chipmaker’s stock climbed to yet another record this week, lifting its market cap close to $6 trillion. Customers can shop for access through Amazon, Microsoft and Google’s cloud services, specialized infrastructure providers known as neoclouds like CoreWeave, online marketplaces, or by purchasing hardware directly. This fragmentation masks a deeper constraint: despite the proliferation of suppliers, capacity remains tight across most tiers.

Hyperscalers dominate but lack capacity for surging demand

Amazon, Google and Microsoft control 59% of the cloud infrastructure market as of 2025, giving them significant pricing and supplier advantages. Leading AI labs including Anthropic and OpenAI have committed to spending over $500 billion between Amazon and Microsoft combined, cementing their position as the primary distribution channel for most enterprise customers. Gartner analyst Hardeep Singh noted that hyperscalers hold a structural advantage: “Hyperscalers are in a good position to show trust to the enterprises because of their 10-plus years of full-stack capabilities.” This reputation matters when enterprises evaluate technology partners for mission-critical AI workloads.

Yet even the largest cloud providers cannot meet current demand. Amazon CEO Andy Jassy told analysts in July that the retailer and cloud pioneer will not be able to serve all the demand it foresees in 2026, adding “I believe this dynamic will also be true in 2027.” The admission underscores why neoclouds have proliferated: hyperscaler capacity simply does not exist to serve every customer willing to pay for it.

Neoclouds filling hyperscaler gaps with different tradeoffs

Startups and mid-market companies are increasingly turning to specialized neoclouds to access GPUs at scale. Modal, a startup operating virtual sandboxes for AI agents, moved beyond hyperscalers to signing agreements with 25 major neoclouds, 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 themselves now contract with neoclouds, with Google and Microsoft both tapping CoreWeave even as they compete with the same provider.

The neocloud model comes with constraints that hyperscalers do not face. Most require upfront payments and substantial lead times, since providers raise funding based on signed contracts before deploying hardware. CoreWeave executive Chen Goldberg noted it would be difficult to turn over 10,000 GPUs to a new customer with one day’s notice. CoreWeave CEO Mike Intrator disclosed on the company’s August earnings call that near-term capacity remains essentially sold out. For companies needing GPUs immediately, the major neoclouds offer limited flexibility.

Smaller, less well-known neoclouds targeting specific geographies or use cases can offer faster provisioning and higher flexibility, but at the cost of brand recognition and stability. Zhen Lu, CEO of Runpod, emphasized that “capacity right now is tight, and your relationships with your suppliers is actually one of the most closely guarded secrets for companies like ours.”

Oracle’s bring-your-own-hardware model and alternative distribution channels

Oracle has opened a different avenue by permitting clients to bring their own GPUs into its infrastructure. The software maker operates this model partly due to financial constraints: it carries more debt than Amazon or Microsoft and holds a lower credit rating, limiting its flexibility to deploy massive capital for GPU purchases. But Oracle sees operational advantage in the arrangement. CFO Hilary Maxson told analysts in June that “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,” using the acronym for return on invested capital. Guggenheim Securities analyst John DiFucci noted it would make sense for chip makers Advanced Micro Devices and Nvidia to bring their own processors to Oracle, though neither has publicly disclosed such arrangements.

Beyond traditional cloud providers, other firms with capital and computing assets have begun renting excess GPU capacity. SpaceX arranged separate deals to provide GPUs to hyperscaler Google and open-source startup Reflection, and in May landed a deal to rent GPUs to Anthropic for $1.25 billion each month through mid-2029. Finance chief Bret Johnsen disclosed in August that “the current economics have translated into a less than one-year payback on our new capital deployments for compute.” OpenAI committed to spending over $300 billion with Oracle over five years but has not mentioned bringing in its own GPUs under Oracle’s bring-your-own-hardware program.

On-premises deployment and direct purchases as control strategies

A growing number of enterprises are returning to traditional on-premises deployment to balance capability with cost control and supply chain certainty. Revenue nearly doubled in Lenovo’s Infrastructure Solutions Group during the June quarter as enterprises sought to install GPU-filled servers in their own data centers. Senior Vice President Vlad Rozanovich observed: “We’re seeing more and more enterprises now starting to say, ‘How do I bring AI into my four walls?'”

The spot price for an Nvidia B200 GPU more than doubled since March according to data maintained by Ornn, a startup indexing GPU pricing. This escalation is driving some companies with existing capital to purchase rather than rent. Dropbox CEO Ashraf Alkarmi said the company relies on GPUs in its data centers, stating “if we want to do a lot more, I think our supply chain connections will still be beneficial and a structural advantage.” Everpure CEO Charlie Giancarlo noted his company has acquired its own GPUs to run open-weight AI models for software engineers, saying “in a very dynamic pricing environment, it’s always good to have multiple sources that you can go to.”

The BlockWest read. Nvidia’s revenue concentration risk actually intensified in the July quarter despite the proliferation of access channels. Five customers now control 10% or more of accounts receivable, up from three. While neoclouds and alternative models appear to democratize access, they represent a distribution layer built atop Nvidia’s chips, not competition for them. The real story for Nvidia shareholders is that scarcity remains the binding constraint, not choice.

Nvidia CEO Jensen Huang stated at a Goldman Sachs tech conference last month that “you’re going to see a whole new crop of really, really exciting neoclouds with hundreds of billions of dollars backlog together,” yet whether these new entrants can materially expand available capacity or simply repackage existing supply will become clearer as the market moves through 2027. Watch whether OpenAI pursues Oracle’s bring-your-own-hardware option, a move that would signal even the largest AI labs see greater value in direct ownership than in pure cloud consumption.

BlockWest is a news publication. Nothing here is investment advice. Read our disclaimer and editorial policy.