Nvidia has poured more than $40 billion into the companies buying its own chips this year, and analysts are openly comparing the pattern to the vendor financing that inflated the dot-com bubble.
Nvidia spent over $40 billion on equity stakes and supply-chain bets in the first several months of 2026 alone, according to CNBC. The single biggest check: $30 billion into OpenAI. That was part of a fundraising round that also pulled in $50 billion from Amazon and $30 billion from SoftBank, valuing the ChatGPT maker at roughly $830 billion. Nvidia also disclosed, in an August 14 regulatory filing, a stake in SpaceX worth $21 billion and a position in Intel worth roughly $30 billion. Add the smaller bets: $2 billion into CoreWeave, $2 billion into Nebius, $3.2 billion into Corning for optical fiber, $2.1 billion into IREN for data centers. Nearly every recipient is also a customer buying Nvidia's GPUs. That's the pattern.
That's the part making Wall Street nervous. Wedbush Securities analyst Matthew Bryson has said the investments fit "squarely into the circular investment theme" worrying investors about how durable AI demand really is, though he also credited Nvidia with building a possible "competitive moat" if it executes well. The concern is straightforward: when a supplier hands a customer billions of dollars, and that customer turns around and spends much of it buying the supplier's product, it becomes hard to tell how much demand is organic and how much is manufactured. Real, or bought? Nobody can quite say.
An Old Playbook #
The comparison to the dot-com era isn't loose colour. Between 1996 and 2001, equipment makers Lucent, Nortel and Cisco extended billions in loans and credit to the telecom carriers buying their gear. Lucent committed $8.1 billion. Nortel extended $3.1 billion, with $1.4 billion still outstanding when the music stopped. Cisco promised $2.4 billion in customer financing. The carriers used the money to buy routers and switches. The equipment makers booked the sales as revenue. Everyone's growth charts looked great, right up until the carriers couldn't pay the loans back and the whole chain unwound at once.
Nvidia's structure isn't identical. Most of these are equity stakes, not loans, and Nvidia's balance sheet, sitting on tens of billions in cash and generating enormous profit margins on Blackwell and Rubin chip sales, bears little resemblance to Lucent's in 2000. But the underlying mechanism is the same one that worried regulators after the telecom bust: a supplier's capital shows up as a customer's demand, and the two get recorded as if they were unrelated.
Nvidia Lines Up $500 Billion From Wall Street To Bankroll the AI Buildout Nvidia has struck deals with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to raise more than $500 billion for AI infrastructure, with Jensen Huang backstopping up to $125 billion of it. Investors wiped $130 billion off Nvidia's market value the same day, citing fears of circular financing tying the chipmaker's fortunes to its...
Nvidia's Defence #
Nvidia's own explanation for the OpenAI deal has shifted. That shift is worth noticing. The original arrangement, announced in September 2025, tied roughly $100 billion in Nvidia investment directly to OpenAI committing to 10 gigawatts of Nvidia compute, paid out in $10 billion installments as the chips were bought. Investors read that as vendor financing in its purest form: Nvidia's money, contractually required to come back as Nvidia orders. In February, the two companies scaled it down to a $30 billion pure equity stake with no purchase obligations attached, giving OpenAI full discretion over how it spends on hardware. Nvidia's argument now is that it's simply investing in a partner's growth, not buying guaranteed orders.
Frankly, that distinction matters less to skeptics than Nvidia would like. OpenAI is still, by a wide margin, Nvidia's largest customer, and a $30 billion cash infusion still makes it easier for OpenAI to keep signing enormous compute contracts, Nvidia's or otherwise. The same logic applies down the list. CoreWeave, now roughly 11% owned by Nvidia, rents out Nvidia GPUs by the rack. Nebius, in which Nvidia holds a stake after a $2 billion investment that sent its stock up 16% in a single day, builds AI infrastructure running on Nvidia silicon. Every dollar Nvidia sends out has a good chance of eventually landing back on its own revenue line.
Nvidia's counter-argument is that this is how a company locks in the next decade of its own market. Owning small pieces of OpenAI, SpaceX, CoreWeave and Nebius gives Nvidia a seat inside the buildout instead of just a purchase order, and a stake that appreciates alongside chip demand is a hedge, not a giveaway. Its Intel position alone has returned roughly 400% on paper since last year. That's not the balance sheet of a company propping up a fiction. It's the balance sheet of a company that, so far, keeps guessing right.
The honest answer is that nobody knows yet which read is correct, and that uncertainty is precisely why this has become the central fault line in AI markets. If AI compute demand keeps compounding the way Nvidia is betting, these stakes look like early, cheap positions in the winners. If it slows even modestly, a supplier that financed its own customer base will find that its revenue and its equity portfolio fall together, not separately. Nvidia shares wobbled and then stabilized in mid-August as these circular-financing fears resurfaced, according to Invezz. That's not a verdict. It's just the market still trying to price a question nobody has answered.
Also read: Anthropic's Dario Amodei Blames AI Backlash on a Crisis of Trust • Grok 4.6 matches GPT-5.6 Sol on benchmarks while undercutting its own $300 plan • OpenAI's New Ultrafast Mode Runs GPT-5.6 Sol on Cerebras Chips, Not Nvidia