NVIDIA has invented the "Computing Power Loan"

In 1848, gold was discovered in California. The following year, nearly 100,000 people flocked to the foot of the Sierra Nevada. The vast majority of them didn't strike gold, but the ship owners at the port of San Francisco made a fortune. The price of a ship ticket from the East Coast to California increased forty - fold. Levi Strauss never dug a single shovelful of sand, yet he became the biggest winner of the gold rush by selling canvas pants. He has a well - known quote that has never been verified but everyone is willing to believe: "The most profitable thing in the gold mine isn't gold, it's jeans."

In July 2026, NVIDIA announced a new business. It didn't come up with a fancy technical name, but its essence is clear: exchanging computing power for profit. AI startups no longer need to pay cash to buy GPUs. NVIDIA accepts their future revenues or profit shares in exchange for the currently scarce computing power.

Two partners are already in place. Sharon AI in Australia will deploy up to 40,000 GPUs. Firmus Technologies in Singapore is building a 360 - megawatt data center on Batam Island, Indonesia, which is expected to accommodate 170,000 GPUs. These 210,000 GPUs are the first batch of collateral for NVIDIA's "computing power loan".

GPUs have become a hard currency that can be mortgaged, paid in installments, and repaid with future earnings. This is the first time in the history of the semiconductor industry.

In the past, chips were just a commodity. It was a straightforward transaction of paying money and getting the goods, and once the deal was done, both parties had no further obligations. However, NVIDIA found that its customers couldn't afford what it produced. Global AI startups are queuing up to train models but can't buy GPUs. The spot price has been sky - high, and venture capital funds are drying up in the interest - rate hike cycle.

This is a classic consumer scenario dilemma: the store is full of customers at the door, but none of them can pull out their wallets.

There are two traditional solutions: either offer price promotions or wait for customers to get paid. NVIDIA chose a third way. It turned chips into financial products. It started accepting customers' future earnings to pay for today's bills.

This is similar in form to subprime mortgages but completely different in terms of risk. Subprime mortgages involve lending money to people without the ability to repay to buy houses. If house prices fall, banks will incur losses. The collateral for the computing power loan is the future profits of AI companies, which come from model training, inference services, and corporate subscriptions. NVIDIA is betting that some of these companies will become the next OpenAI or Anthropic. With each GPU it lends out, it locks in the profit shares of these future winners in advance. Venture capital institutions dream of getting on the shareholder list of AI unicorns, but NVIDIA doesn't need to dream; it just needs to deliver the goods.

Why is NVIDIA so generous with credit?

The first reason is that its customers are really short of money. In the past two years, global investment in AI infrastructure has expanded exponentially, but the revenue and profit at the end - user application level have not kept up. OpenAI and Anthropic are still making huge losses, and computing power leasing providers are also in the red. Cloud providers can only barely break even by amortizing huge capital expenditures over five or six years. All the cash in the entire industry has been sucked up by the upstream, leaving only a dry riverbed downstream.

NVIDIA itself announced in June that it plans to issue bonds worth at least $20 billion. It wants to sell chips to the downstream, but it has already drained the downstream of cash. It's like a winery that has bought all the grapes within a hundred miles, made them into sky - high - priced red wine, and then found that the villagers can no longer afford the wine. So it launches an installment - payment plan, accepting the villagers' next year's harvest as collateral. It's kind, but in a very calculated way.

The second reason is the intensifying competition. AMD is catching up, Intel is transforming, Google's TPU is eroding the inference market, and Amazon's Trainium is reducing the dependence on GPUs. Once there are alternatives, even if they are of slightly lower performance, price will become the decisive factor. NVIDIA needs to fill every data center in the world with its GPUs before the alternatives mature. The computing power loan is its tool to gain time and market share. Occupy the positions first and then collect rent slowly.

The most hidden motivation lies in the revenue structure.

NVIDIA's choice of profit sharing instead of simple installment payments shows that it has seen through the ceiling of the hardware business. Selling chips is a one - time deal; you make money for each GPU sold. But if you turn that GPU into a share in a money - printing factory, then every time that factory prints a bill, NVIDIA gets a cut. This is a financial engineering move to transform from a hardware company to an infrastructure platform, but it's disguised as a customer - care program.

NVIDIA is no longer just a shovel seller. It starts to use shovels to take stakes in gold mines and share the mining proceeds of every ounce of gold. The business model of this deal is no longer hardware but finance. It no longer just sells chips to customers; it uses chips to exchange for customers' future cash flows, turning its balance sheet from "inventory" to "equity". And the best part is that when those AI startups finally start to make a profit, they'll find that the first person they have to share their profits with isn't the angel investor who invested in them early on, but the hardware supplier who sold them the GPUs.

The most far - reaching impact of this is that it quietly changes the production relations in the AI industry.

In the past, computing power was a means of production that enterprises owned or leased. But the computing power loan turns computing power into a resource that can be exchanged for profit shares. It becomes an input similar to equity. In the future, AI startups no longer need to raise funds first, buy GPUs, and then train models. They can directly use their future profits to pay for today's computing power costs.

This sounds like a fairy tale of democratizing financing, but the ending may be more like a dystopian fable. NVIDIA is transforming from a simple chip supplier into a super hub. It controls the distribution of computing power and also holds the right to claim the profits of future AI winners. Those companies that accept the computing power loan have already given the first line of their profit statement to NVIDIA before their business models are proven. NVIDIA doesn't need to judge which company will win; it just needs to ensure that no matter who wins, the winner owes it money.

In every gold rush in history, the people who ultimately stay aren't the first ones to find gold, but those who exchange jeans for gold dust, shovels for shares, and ship tickets for land deeds. They never participate in gold - digging; they only participate in the fates of gold - diggers. When the last grain of gold dust is mined and the gold - diggers leave, they pack up their canvas and tools and open a bank on the deserted gold mine.

And NVIDIA is becoming the fate of all gold - diggers.

This article is written based on public information and is for information exchange only. It does not constitute any investment advice.

This article is from the WeChat official account "Silicon - based Starlight", author: Starlight. It is published by 36Kr with authorization.