Nvidia's $100B OpenAI 'deal' was never real — it was an IOU

By TechPulse USA Newsroom

The most consequential Nvidia–OpenAI story of October isn’t a story at all. It is a corpse. The Wall Street Journal reported on 30 January — nine months ago — that Nvidia had halted negotiations on its planned $100 billion investment in OpenAI, and the deal never came back to life. What is fresh, and what actually matters this week, is everything that replaced it: a $30 billion equity deal whose final wires landed this month, a brand-new $30 billion raise OpenAI is already courting at a $1.4 trillion valuation, and a chip giant suddenly shopping for allies everywhere else.

We checked the Reuters wire archive: the “pause” story is dated 30 January 2026. Zombie recirculations of it keep passing it off as this week’s news. Everything below is what actually happened next — and it is a far better story than the recycled headline.

Photo: Jensen Huang, Nvidia’s founder, president and CEO, at a 2024 meeting in Brussels. File photo. Credit: Peter Dasilva / European Commission, CC BY 4.0.

The $100 billion was never a deal

In September 2025, Nvidia and OpenAI announced what looked like the largest single commitment in the history of the AI boom: a letter of intent under which Nvidia said it “intends” to invest up to $100 billion in OpenAI, progressively, as each gigawatt of Nvidia systems was deployed into OpenAI’s data centres. The target was at least 10 gigawatts of Nvidia-powered compute — roughly the peak electricity demand of New York City — with the first phase due in the second half of 2026 on the Vera Rubin platform. Huang called it “the largest computing project in history”.

It was not a contract. It was a letter of intent — non-binding, milestone-linked, the corporate equivalent of a heavily caveated promise. Nvidia’s own securities filings had warned since November that the money might never materialise.

Then, on 30 January, the Journal reported that negotiations had broken down after some inside Nvidia expressed doubts about the deal, citing people familiar with the deliberations. The paper said Huang had privately emphasised that the agreement was non-binding, had privately criticised what he saw as a lack of discipline in OpenAI’s business approach, and was worried about the competition OpenAI faces from Google and Anthropic.

Nvidia’s official response was a study in corporate diplomacy. “We have been OpenAI’s preferred partner for the last 10 years,” a spokesperson said, carefully declining to address the state of the talks. OpenAI declined to comment.

Asked about the Journal’s reporting in Taipei a few days later, Huang was less diplomatic. It was a Sunday. A reporter wanted to know whether the story meant he wasn’t happy with OpenAI. “That’s nonsense,” he said — one word, and he moved on. Then he gave the line that should have killed the $100 billion mythology for good: “It was never a commitment… They invited us to invest up to $100 billion, and we will invest one step at a time.” And, for anyone still unsure where he stood: “We will invest a great deal of money. I believe in OpenAI.”

What replaced it: a $30 billion equity wire

By February, the shape of the replacement was public. The Financial Times reported, via a Reuters relay, that Nvidia was close to finalising a $30 billion equity investment in OpenAI that would replace the $100 billion commitment — a straight ownership stake rather than deployment-linked infrastructure financing, forming part of a funding round reported to value OpenAI at $730 billion before new money. Reuters noted it could not immediately verify the FT’s report.

It took months to close. An initial $10 billion tranche was meant to move once a definitive agreement was reached, but the negotiations dragged on far longer than expected. Then, in the first week of October, the wires finally moved: Nvidia and SoftBank each sent a final $10 billion tranche to OpenAI, completing their separate $30 billion commitments tied to the ChatGPT maker’s latest round.

SoftBank’s side of the ledger is the more startling story. Masayoshi Son has now put a staggering $64.6 billion into OpenAI, acquiring roughly 13% of the company and becoming one of its largest outside shareholders. To fund it, Son liquidated SoftBank’s entire Nvidia position — all 32.1 million shares, worth about $5.8 billion, gone by the end of 2025 — and in September issued $11.1 billion of high-yield bonds at rates as high as 9.75%, a global record for a junk-bond sale. Speaking at the Tokyo FII Priority Asia forum, Son said he had sold his Nvidia shares “with tears”: “I didn’t want to sell a single share.” He sold them anyway. OpenAI mattered more.

It is worth pausing on the circularity here, because it is the quiet scandal at the heart of the AI boom’s financing: Nvidia wires billions to OpenAI, and a large share of that capital flows straight back to Nvidia to pay for chips. The shovel merchant has joined the prospectors in the riverbed. For a broader look at how this debt-fuelled machine has been wobbling, see our coverage of AI’s $113B debt machine hitting a wall in September.

OpenAI is already back for more — $30 billion at $1.4 trillion

The $30 billion barely had time to land before OpenAI started asking for the next one. Last week, Bloomberg reported that OpenAI is in talks to raise $30 billion or more at a $1.4 trillion valuation, excluding new money, ahead of a possible 2027 IPO. The company told investors it expects annualised revenue to reach or exceed $70 billion by the end of 2026, up from about $50 billion at the end of September.

Photo: Sam Altman, OpenAI’s CEO, speaking at TED in April 2025. File photo. Credit: Steve Jurvetson, CC BY 2.0.

That $50 billion figure caused its own small earthquake. The Financial Times reported on 8 October that OpenAI’s annualised revenue was roughly $20 billion below the $68 billion figure that had been circulating — and AI stocks sold off hard. Nvidia dropped about 3%, erasing roughly $170 billion in market value in a single day; Broadcom fell more than 4%; AMD, Micron, Sandisk and Intel all slid. Analysts later noted the gap was mostly accounting: the higher number had included revenue from OpenAI’s cloud partners.

Still, the whiplash says something. OpenAI now burns cash at a pace that would make a sovereign wealth fund blink, and every funding round has to be bigger than the last to keep the story intact. The 2027 IPO, reportedly targeting an $852 billion valuation, is the escape hatch — but between here and there stands another raise, and quite possibly another after that.

Why Nvidia is suddenly shopping elsewhere

Here is the detail that should worry OpenAI most: Nvidia isn’t putting all its chips on one number anymore. Last week brought reports that Nvidia plans to invest in d-Matrix, a rival AI chip developer that previously raised at a $2 billion valuation and is now seeking $5 billion — in other words, Nvidia is funding a company that wants to eat its own lunch. Separately, Nvidia is said to be exploring a deeper stake in Reflection AI, potentially through a full acquisition.

Meanwhile, the competitive pressure is no longer theoretical. Microsoft’s new Maia 300 chip is engineered to deliver 40% more tokens per megawatt than leading GPU configurations, with a 96-accelerator rack promising up to seven times the throughput of the Maia 200 generation — and full production deployment across Azure slated for 2027. Every hyperscaler building its own silicon is a future Nvidia invoice that never arrives.

And the memory market just got a jolt from an unexpected direction. Over the weekend, China’s ChangXin Memory Technologies (CXMT) announced progress on a 4F² DRAM cell architecture — a genuine density breakthrough — and said it plans to ship DDR5 server memory on it before the end of 2026. On Monday, Micron fell nearly 4%, Samsung and SK Hynix dropped more than 4% each, and the KOSPI slid 3.65%. CXMT’s share of global DRAM revenue has doubled in a year, to 10%.

The market has noticed the wobble. Nvidia shares retreated toward $225 on Monday from a $243.37 record set on 6 October — a drawdown of nearly 8% in six days for a $5.5 trillion company. As Hock Tan warned this week, capital — not chips — is now AI’s binding constraint, and capital is getting choosier.

Our take: the IOU era of AI infrastructure is over

What was genuinely botched here was never the pause — pausing a $100 billion non-binding IOU was arguably the shrewdest decision Nvidia made this year. What was botched was the nine months of commentary, analyst decks and breathless headlines that kept treating the money as real, pricing it into valuations and data-centre plans as though a letter of intent were a signed cheque. It wasn’t. Jensen Huang told everyone that, on the record, in February. The market simply preferred the bigger number.

The new structure — $30 billion of real equity, wired and closed — is healthier for both sides. But the circularity hasn’t gone away: OpenAI’s largest investor after SoftBank is now its largest supplier, and its next $30 billion will, in large part, come straight back to Nvidia as purchase orders. That loop works beautifully until the day somebody asks who is actually paying for the shovels.

The honest question for the next twelve months isn’t whether Nvidia believes in OpenAI — Huang has said, on the record, that he does. It’s whether anyone still believes the AI buildout can be financed on promises. The era of the $100 billion IOU is over; the era of the $30 billion wire has begun. Anthropic’s scramble to lease 15GW of compute — including a secret deal with a bitcoin miner — suggests the rest of the industry has drawn the same conclusion.

Sources

Reuters, the Wall Street Journal and the Financial Times (via relays); The Hindu BusinessLine; Datacenter Dynamics; Livemint; the Economic Times; Morningstar/MarketWatch; Bloomberg (via relay reporting); Geo TV; TradingKey; AIStockWire; Finnhub market data.

TechPulse USA Newsroom is our US-based reporting team covering artificial intelligence and the technology reshaping the world.

Nvidia–OpenAI funding saga FAQ: the questions readers keep asking

Was the $100 billion deal ever real? No. It was a non-binding letter of intent from September 2025, milestone-linked to gigawatt deployments — never a signed contract. Nvidia's own filings warned it might never materialise, and Huang said in February: "It was never a commitment."

What replaced it? A $30 billion equity investment forming part of the round reported to value OpenAI at $730 billion pre-money. The final $10 billion wires from Nvidia and SoftBank each landed in the first week of October.

What did SoftBank give up to fund its stake? Son liquidated all 32.1 million Nvidia shares (~$5.8B) by end-2025 and issued $11.1 billion of high-yield bonds at up to 9.75% — a global junk-bond record. He now holds roughly 13% of OpenAI.

Why is the circularity a problem? Nvidia's billions to OpenAI largely flow back to Nvidia as chip purchase orders; OpenAI's largest investor after SoftBank is now also its largest supplier. The loop works beautifully until somebody asks who is actually paying for the shovels.

What is OpenAI raising next? At least $30 billion at a $1.4 trillion pre-money valuation (Bloomberg), with a possible 2027 IPO as the escape hatch.

Update, Oct. 15: the race's other runner is sprinting

While OpenAI courts $30 billion at a $1.4 trillion valuation, Anthropic is set to hold a pre-IPO investor day at its San Francisco headquarters on October 14, Bloomberg reported — a select group of institutional investors quizzing senior executives ahead of a potential November market debut, possibly before Thanksgiving. Investors reportedly believe a fair valuation for Anthropic sits in the $1.8 trillion to $2 trillion ballpark. So the scoreboard reads: one lab raising its fourth mega-round in private, the other preparing the largest AI listing yet — both chasing the same thing, money to rent the compute neither can afford to buy. As Hock Tan warned this week, capital — not chips — is now AI's binding constraint, and capital is getting choosier.