Nvidia-OpenAI $100B Letter of Intent: Terms & Pullback

What was actually announced

On September 22, 2025, NVIDIA and OpenAI signed a letter of intent under which NVIDIA would invest up to $100 billion in OpenAI, paid out progressively as OpenAI deploys at least 10 gigawatts of NVIDIA systems. The first gigawatt is targeted for the second half of 2026, built on NVIDIA's Vera Rubin platform. Jensen Huang called it a partnership to "deploy 10 gigawatts to power the next era"; Sam Altman said OpenAI would "utilize what we're building with NVIDIA." Both companies said they intended to finalize binding terms in the coming weeks. [1]

A letter of intent is not a closed transaction. It states an intention, not a set of executed contracts, and NVIDIA's own language throughout was conditional: it "intends" to invest "up to" $100 billion, and the money moves only as OpenAI actually stands up each gigawatt of hardware. Reporting at the time described the underlying cash flow as OpenAI paying NVIDIA cash for chips, with NVIDIA in turn taking non-controlling equity in OpenAI rather than a seat at the table or operating control. [1][2]

The circular financing criticism

The mechanical objection is straightforward: NVIDIA is OpenAI's largest hardware supplier and, under this deal, also becomes an investor whose return depends on OpenAI buying more NVIDIA hardware with the money NVIDIA just supplied. Bernstein analyst Stacy Rasgon wrote that "the action will clearly fuel 'circular' concerns," noting NVIDIA had already made more than 50 venture investments into AI companies by 2024, many of which used that capital to buy NVIDIA chips, but that the OpenAI investment's scale "appears to dwarf all the others." Seaport Global's Jay Goldberg compared NVIDIA's equity stake to a parent co-signing a child's first mortgage, useful for securing cheaper debt but carrying, in his words, "a whiff of circular financing" and signs of "bubble-like behavior." [3]

Atlas interpretation: The mechanism critics were pointing to is not that the deal is fraudulent, but that it can make demand look larger than it is. If NVIDIA's own capital is what funds a chunk of OpenAI's chip purchases, then part of NVIDIA's reported revenue growth traces back to NVIDIA's own balance sheet rather than to a customer's independently sourced budget. The same dollars can show up as both an NVIDIA investment gain and an NVIDIA sales figure, and as both OpenAI compute capacity and an OpenAI liability, without any new outside money having entered the system. [3]

How this compares to the other 2025 mega-deals

Atlas interpretation: OpenAI reached three large infrastructure arrangements in 2025 with three different financing shapes. The Stargate project, announced in January with SoftBank and Oracle, is a separate joint venture funded by equity contributions from its founders plus project debt raised against signed compute contracts, structurally closer to conventional infrastructure finance than to a chip vendor investing in its own customer. The NVIDIA deal instead has the supplier itself as the funding source, which is the specific feature that drew the circular financing label; Stargate's investors are not the same firms selling OpenAI the hardware inside it. [4]

Atlas interpretation: The AMD deal announced the following October runs on a different mechanism again: instead of AMD investing cash into OpenAI, OpenAI receives a warrant for up to 160 million AMD shares, worth roughly 10 percent of AMD, vesting only as OpenAI deploys AMD chips and AMD's stock price climbs toward milestones as high as $600 a share. That ties AMD's upside to OpenAI's success as a customer without AMD writing OpenAI a check, which sidesteps the specific criticism leveled at the NVIDIA arrangement even though both deals link a chipmaker's fortunes to one buyer's spending. [5]

Before either deal, Microsoft had been OpenAI's exclusive cloud provider since a 2019 investment, meaning any GPT-4-class model run at enterprise scale ran on Azure. That exclusivity was not part of the NVIDIA announcement, but it is the backdrop against which OpenAI signing hardware-specific deals with NVIDIA and then AMD reads as a shift toward a multi-vendor compute strategy rather than a single locked-in relationship. [6]

Market reaction and what happened to the commitment

NVIDIA's stock rose roughly 3 to 4 percent on the day of the announcement, and the news lifted semiconductor stocks more broadly the following day. The market's initial reaction, in other words, treated the news as bullish for NVIDIA even as analysts were simultaneously publishing notes questioning the deal's structure. [7]

The $100 billion figure did not hold. By late January 2026, the Wall Street Journal reported the arrangement had shifted toward an equity investment of, in its words, "tens of billions of dollars" rather than the full amount, which Huang publicly called "nonsense" while declining to name a figure. By early February, Huang was describing the original number differently: "It was never a commitment," he said, adding that NVIDIA would "invest one step at a time" and that the current round "wouldn't approach $100 billion" even though he expected it to be NVIDIA's largest investment to date. Multiple outlets reported the figure under discussion had narrowed to around $30 billion, no longer structured around the original gigawatt-by-gigawatt deployment schedule. [8][9][10]

Atlas interpretation: That walk-back is consistent with what the circular financing critics were describing in September: a letter of intent is a statement of terms both sides can revise or abandon, not a liability either company was bound to honor, and the number that generated headlines at signing was never the number either party was contractually required to deliver. [8][9]