Sam Altman Just Made Venture Capital History - But Read the Fine Print
In November 2025, OpenAI investor Brad Gerstner asked Sam Altman on a podcast what he called "the single biggest question hanging over the market." How can a company doing $13 billion in revenue commit to $1.4 trillion in infrastructure spending? Altman's response wasn't a financial argument. It was: "If you want to sell your shares, I'll find you a buyer. Enough."
Three months later, OpenAI quietly revised that $1.4 trillion down to $600 billion - a 57% reduction, no press release, no acknowledgment that the original number had ever been a problem. This week, the same company raised $122 billion at an $852 billion valuation. Almost nobody is connecting those three things together.
The Investors
The three anchors behind the $110 billion core:
- Amazon: up to $50 billion
- Nvidia: $30 billion
- SoftBank: $30 billion
Extended close co-leads: Andreessen Horowitz, D.E. Shaw Ventures, MGX (Abu Dhabi), TPG, T. Rowe Price Associates, plus $3 billion from individual investors via bank channels for the first time ever.
Three names are doing the heavy lifting, and what each of them actually gets out of this is worth understanding properly.
Amazon's $50 Billion Isn't $50 Billion
$15 billion landed at close. The remaining $35 billion is contingent on OpenAI achieving AGI or completing an IPO by end of 2026. The investment also came bundled with a commitment for OpenAI to spend $100 billion on AWS over eight years. Amazon puts in $50 billion, gets $100 billion in guaranteed cloud spend back, and if the commercial agreement between them terminates, the $35 billion equity commitment dies with it - the two are contractually inseparable. The AGI angle is genuine, but the structure tells you where the real priority sits.
Nvidia's $30 Billion is Mostly Chips
Nvidia didn't write a cheque. They committed $30 billion worth of GPU compute - 3 gigawatts of inference and 2 gigawatts of training on their Vera Rubin systems. Their entire market cap is built on the narrative that AI demand for GPUs is permanent and growing. OpenAI is the most visible proof point of that demand. This is Nvidia ensuring their most important customer stays solvent and keeps buying, because the alternative is bad for everyone sitting on Nvidia stock.
SoftBank is the Most Exposed
SoftBank borrowed the money to invest it - a $40 billion unsecured bridge loan via JPMorgan, Goldman Sachs and others, maturing March 2027. S&P immediately downgraded their credit outlook to negative. That 12-month term on an unsecured loan of this size is the banks pricing in an OpenAI IPO within the window. If that IPO doesn't materialise at this valuation, SoftBank's position unravels fast.
Microsoft is Already in the Walls
Microsoft participated at an undisclosed amount but already owns 27% of OpenAI, takes 20% of their total revenue through 2032, and routes most of OpenAI's API traffic through Azure. This round just deepens what was already an inseparable relationship.
The Revenue Question
OpenAI is reporting $2 billion a month in revenue, and it's being treated as proof of market demand. But look at who is actually spending. OpenAI committed to spending $100 billion on AWS as part of the Amazon deal. Nvidia's investment is chip supply OpenAI will consume. Microsoft captures revenue through Azure while holding 27% of the company. When your investors are also your largest contracted spenders, the revenue figure starts looking less like open market demand and more like money moving in a closed loop - what Wall Street calls circular financing.
There's also a detail most coverage skips: OpenAI shares 20% of all revenue with Microsoft through 2032, so the headline number isn't what they actually keep. And they have never disclosed their enterprise retention rate - the most important metric in any subscription business. As someone who moved from OpenAI to Anthropic myself, and who is watching the same migration happen across the industry, I don't think that omission is a coincidence. You don't hide the number when it helps your story.
What Happens When the Bill Comes Due
OpenAI is projecting losses of $14 billion this year, $35 billion in 2027, $47 billion in 2028. Eventually the path to profitability runs through raising prices, and that is where this becomes a bigger conversation than a balance sheet problem. AI is no longer a novelty — as someone running a startup myself, I see daily how deeply it is embedded in how people work and think. If the companies leading this technology can only survive by making it significantly more expensive, what was sold as democratising intelligence quietly becomes another premium service. The wealth gap widened by technology access has happened before. It could happen here, gradually and without anyone calling it a policy decision.
The Valuation and the Bigger Picture
$852 billion for a company losing $14 billion this year, valued at 14x projected 2027 revenue, while Google - one of the most profitable businesses ever built — trades at 6.7x. The fastest companies to ever scale from $10 billion to $100 billion were Tesla and Meta, both taking seven years at 1.3x annual growth. OpenAI is projecting 2.3x this year, 2x in 2027, 1.6x in 2028, while the CEO responded to questions about financial sustainability by telling his own investor to sell his shares .
The financing structure has the characteristics of a bubble. The technology does not. AI is too embedded in daily life to simply stop - even if the capital structure corrects, through a valuation reset, a disappointing IPO, or a SoftBank refinancing crisis, that isn't the end of AI. It's a slowdown, a reconfiguration, and a reckoning for the companies that built revenue from investors sitting across the table rather than from the market.
The $122 billion headline is genuinely historical - Sam Altman regardless of personal opinions on him is the best fundraiser ever. But the $25 billion that actually landed in cash at close is also real. Which number tells you more about what's actually going on here is a question worth sitting with before the S-1 drops.