Nvidia is turning compute access into a revenue claim on the next wave of AI startups. If you can't afford the GPUs now, the company has found a way to get paid when your product starts earning.
Bloomberg reported on July 2 that Nvidia is pushing a revenue-sharing structure for researchers and early AI companies that need serious compute but don't have the cash to buy it outright. The company still sells the hardware to cloud partners, but the new piece is more interesting: those partners rent Nvidia-powered capacity to startups, and Nvidia takes a share of the cloud revenue that comes back.
That's the polite version. The blunt version is that Nvidia would rather own a slice of your usage and revenue than watch you wait, raise money, and shop around for a cheaper chip stack later.
The structure runs through cloud providers, not directly through a founder's bank account. Barron's reported in June that Sharon AI signed a six-year agreement with Nvidia to deploy 72 megawatts of data-center capacity in Australia, using up to 40,000 Nvidia AI processors for startups, enterprises and university researchers. Nvidia gets its standard product revenue and a cut of Sharon AI's cloud-service income. That is not a normal hardware sale. It is a hardware sale with a tail.
Firmus Technologies is the larger example. The Australian reported this week that its Nvidia partnership runs until 2034 and is anchored by a 360-megawatt AI facility in Batam, Indonesia. The agreement covers up to 170,000 Nvidia AI accelerators across Grace Blackwell, Vera Rubin and Vera platforms through 2027 and 2028, with Firmus expecting $25 billion to $30 billion of income from committed offtake deals over the first six years. Those figures tell you what Nvidia is really chasing. Not a one-off server order. A stream.
For startups, the offer is easy to understand. You skip the funding round earmarked for infrastructure, avoid signing a data-center contract you can barely explain to your board, and start training or running inference earlier than you otherwise could. If you're building an AI product and compute is the thing holding you back, this is a real offer.
It is also expensive in a quieter way.
Nvidia Is Selling Access, Not Just Chips
A normal GPU sale ends when the invoice clears. This model keeps Nvidia attached to the customer's growth. The more a startup uses, the more the cloud partner bills, and the more Nvidia earns from the capacity it helped put in place. You can see why the company likes it. Nvidia already has the hardware everyone wants. Now it is using that position to shape how young AI companies finance themselves.
This fits a bigger pattern. Nvidia has spent the past year tying itself more closely to the companies that consume its chips. Business Insider reported in February that OpenAI raised a huge financing round with backing from Amazon, SoftBank and Nvidia, including a reported $30 billion from Nvidia. Reuters and other outlets have also reported on Nvidia's late-2025 deal to license Groq's inference technology and bring senior Groq executives into the company. When your customers, suppliers and challengers all sit inside the same capital map, you are no longer just selling picks and shovels. You are shaping the mine.
Frankly, founders should read the deal twice. Revenue sharing sounds lighter than debt and cleaner than dilution, but it still gives someone else a claim on future dollars. If your company fails, the cost may not matter much. If it works, the bill follows you into every customer contract and every margin conversation. That's not a reason to reject it. It is a reason to stop pretending compute is just another cloud line item.
The defensive logic is obvious too. Google has its TPUs. Groq built a business around inference chips. Etched, the transformer-focused chip startup, reached a $5 billion valuation this week after announcing $1 billion in system orders, according to TechCrunch. A startup that can get Nvidia capacity today by sharing future revenue has less reason to spend a year testing alternatives. Lock-in does not always arrive as a contract clause. Sometimes it arrives as relief.
That is why this story is bigger than a financing tweak. Nvidia is giving early AI companies access to the one resource they can't fake, and taking a claim on the one thing they hope to create: revenue. If you're building in AI, you should be grateful that the compute door is opening wider. You should also notice who is standing at the door.
Also read: OpenAI Is Discussing Giving Washington a Stake Instead of Selling One • SoftBank pushes for a full $10 billion loan against its OpenAI stake • Robinhood Stops Renting Blockchain Rails and Builds Its Own
