Written by: Ada

A company valued at hundreds of billions of dollars wants to borrow billions to build houses.

The lender says: no loan.

The reason is simple: your business model is not yet validated, and analysts predict you may burn through cash by mid-2027. How will you repay?

This is not a financing accident of some startup. This is OpenAI's actual encounter in 2025.

According to an exclusive report by The Information, OpenAI had sent executives to survey locations across the U.S. with plans to build its own data centers, attempting to raise billions to kick-start construction. As a result, they were turned away by lending institutions. Tom's Hardware cited analysts predicting that OpenAI could run out of cash as early as mid-2027.

A year ago, Sam Altman stood next to the White House podium, announcing the Stargate project: $500 billion, four years, and building the world's largest AI data center network in collaboration with SoftBank and Oracle. Trump called it "the largest AI infrastructure project ever."

A year later, this joint venture had neither assembled a team nor developed any data centers; the three partners couldn't even agree on who would be responsible for what. OpenAI couldn’t build on its own either.

Thus, OpenAI began crunching the numbers.

The $500 billion dream shatters over "who will manage"

The report by The Information reveals a story that has decayed behind the spotlight for an entire year.

A few weeks after the White House press conference, Stargate fell into paralysis. No one was leading, and there was no coordination mechanism. OpenAI, Oracle, and SoftBank engaged in a tug-of-war over "who will build, who will manage, and how the money will be divided."

OpenAI wanted to build its own data centers; it was its original obsession. The logic made sense: long-term leasing of computing power is too expensive; only by building can it control its destiny.

But lenders saw it differently.

A company that burned through $2.5 billion in cash in six months and projected $8.5 billion in losses for the entire year comes and asks to borrow billions to build data centers? What lenders want to see is not your PPT; it's your cash flow. And OpenAI itself predicts that it won't achieve positive cash flow until at least 2029.

This is like someone who hasn’t even started making money applying for a loan to build a villa, and the bank's first question is: how will you repay? They can’t answer.

The self-building route was blocked. OpenAI was forced back to the negotiating table to continue discussions with its Stargate partners.

But negotiations were equally tough. SoftBank has several large data center projects in Texas, and OpenAI wanted to use one of them as its first facility. SoftBank disagreed, wanting to retain control. OpenAI's team flew to Japan multiple times in September and October to negotiate face-to-face with Masayoshi Son.

The final negotiation result was that OpenAI would sign long-term leases and control design; SB Energy, a subsidiary of SoftBank, would be responsible for development and ownership.

In other words, OpenAI went from wanting to be the landlord to becoming a tenant.

$800 billion evaporated

If the internal chaos of Stargate was a hidden wound, this next number represents a public self-correction.

According to CNBC, OpenAI has reduced its total spending target for computing power before 2030 to approximately $600 billion, along with a clearer timetable and revenue projections. Revenues are expected to exceed $280 billion by 2030, with consumer and enterprise markets each accounting for half.

From $1.4 trillion cut to $600 billion, a reduction of 57%.

The official statement is: "to better align spending with revenue growth."

The real meaning is: investors are no longer buying it.

The previous number was more like a dream wishlist; $600 billion is at least a modelable figure. But even so, to expect revenues to exceed $280 billion by 2030 requires a continuous compound annual growth rate of over 50% for five years. Who can guarantee this?

OpenAI's revenue for 2025 is projected at $13.1 billion, burning $8 billion. It is still a long way from profitability. The company itself expects to achieve positive cash flow by 2029. Before then, cumulative losses could reach $115 billion.

This is the sound of waking from the dream.

It's not that Altman does not want to spend $1.4 trillion. Reality tells him: you can’t afford it.

Books cannot support the dream

Why must OpenAI transform from dreamer to accountant? Not because it made a strategic error, but because three harsh realities hit simultaneously.

First, the speed of money going out far exceeds the speed of money coming in.

OpenAI's revenue in the first half of 2025 is $4.3 billion, burning through $2.5 billion in cash. Projected total revenue for the year is $13.1 billion, burning $8 billion. According to investor documents cited by Fortune, expected losses will expand annually, with operating losses possibly reaching as high as $74 billion by 2028, and only possibly turning positive by 2029 or 2030. Cumulative losses are expected to reach $115 billion.

OpenAI's current state is spending at ten times the speed while earning only at double the speed. Mathematically, this line will eventually cross; the only question is whether that crossing happens in 2029 or never.

Second, can computing efficiency offset scale expansion? Although OpenAI's "computing power profit margin" (revenues minus model operating costs) increased from 52% in October 2024 to 70% in October 2025. Algorithm optimization and hardware utilization have improved. However, each time a larger model or a feature that consumes more computing power (like video generation) is launched, these efficiency gains will be eaten away.

Third, the paid conversion rate is stuck.

ChatGPT's weekly active users have surpassed 900 million. However, according to Incremys data, the paid conversion rate is only about 5%, with over 95% of users on the free tier. OpenAI has begun testing ads in the free version. This act itself is a signal; when you start charging users for their attention, it indicates that you have hit the ceiling of the subscription model.

Meanwhile, competitors are stealing users with less money. According to Similarweb data, ChatGPT's global traffic share fell from 87% to about 65% within a year. Google Gemini surged from 5% to 21% thanks to its default integration on Android and embedding in Workspace, not by having a stronger model, but through overwhelming distribution. Anthropic's Claude achieved the highest user engagement with only 2% traffic share (averaging 34.7 minutes per day), focusing on high-value enterprise customers, burning cash at a fraction of OpenAI's speed.

"ChatGPT established this category, but as alternatives emerge, users will naturally disperse," said Tom Grant, Vice President of Research at Apptopia.

And competitors are doing the same with less money. DeepSeek is stirring the market with open-source models at extremely low costs. Google is crushing with distribution. Anthropic is acquiring high-value customers with a focused strategy. If AI models tend to converge in functionality, the market will ultimately be determined not by whose model is strongest, but by whose ecosystem is deeper and whose costs are lowest.

OpenAI is attempting to win three wars simultaneously: the model race, the infrastructure race, and the commercialization race. However, no company in history has won on all three fronts at the same time.

Altman's Plan B

The dream shattered, but Altman did not stop.

He did something that all business textbooks would recommend but few dreamers would dare to do: abandon obsession and pragmatically find a way to survive.

The dream of building its own data center was abandoned. The strategy now is to sign numerous contracts outside the Stargate framework. A $30 billion annual computing power procurement agreement with Oracle, deepening cooperation with CoreWeave, and even looking to AWS and Google Cloud to fill gaps. Chip supply is also being diversified, with AMD and startup Cerebras joining besides Nvidia.

OpenAI's CFO Sarah Friar publicly stated at the Davos Forum that the company is intentionally protecting its balance sheet through partnerships.

This statement would have been unimaginable a year ago. At that time, Altman was discussing trillion-dollar infrastructure commitments, 10GW of computing power capacity, and general artificial intelligence that would change the fate of humanity. Now, his CFO is talking about "protecting the balance sheet."

But OpenAI's financing scale remains astonishing, with the latest round potentially exceeding $100 billion. According to Bloomberg, OpenAI is nearing completion of the first phase of a new funding round, with the company's overall valuation possibly exceeding $850 billion when including the financing. Expected investors include Amazon (with a forecasted investment of $50 billion), SoftBank ($30 billion), Nvidia ($20 billion), and Microsoft.

But note the identity of these investors: chip suppliers, cloud computing platforms, and strategic investors demanding OpenAI use their services. This is not venture capital betting on a dream; it’s more like the upstream and downstream of the supply chain locking in a major customer.

Previously investing in OpenAI was like buying a lottery ticket; now investing in OpenAI is like signing a supply contract, fundamentally changing its nature.

Gravity

Let's zoom back to Stargate.

A year ago, on the stage of that White House press conference, Sam Altman announced the $500 billion Stargate plan from the center of the stage.

A year later, the joint venture within this plan has turned contentious. OpenAI bypassed its initiated joint venture framework to sign separate agreements with Oracle. Computing power targets were not met; out of the 10GW promised, only 7.5GW was achieved. Spending expectations dropped from $1.4 trillion to $600 billion.

This is not a story of failure. OpenAI has not collapsed; it is still raising money, still growing, with over 900 million users.

But this is a story of waking from a dream.

From "building the largest data center empire in the world" to "first ensure survival, then fight with other people's money and infrastructure." From wanting to be a landlord to becoming a tenant. From dreamer to accountant.

Faced with the stalled progress of the Stargate project, Elon Musk coldly tossed a remark on X: "Hardware is hard."

This statement may be harsh, but it points to a reality that all AI companies will eventually face. The computing arms race has reached this stage, and the real threshold is no longer about who trained the strongest model. It's about who can place a gigawatt of infrastructure in the physical world without burning themselves out.

Altman chose not to burn himself out. This may be the least sexy but the most correct decision he has ever made.

As for that $500 billion Stargate dream, it hasn't died, but it is no longer what it was a year ago. It has transformed from a narrative about changing humanity's destiny into a balance sheet that needs to be checked line by line.