When Dreams Wake Up: When the Dream Makers at OpenAI Start Settling the Accounts

OpenAI’s Journey from Interstellar Gate Blueprint to Financial Reality Check

By Ada

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

Lenders say: No.

The reason is straightforward: your business model hasn’t been validated yet, and analysts predict you might burn through cash by mid-2027. How will you repay?

This isn’t a funding mishap for a startup. It’s OpenAI’s real experience in 2025.

According to an exclusive report by The Information, OpenAI sent executives across the U.S. to scout locations, planning to build its own data centers and seeking hundreds of millions of dollars to start construction. But they were turned down by lenders. Tom’s Hardware cited analysts estimating that OpenAI could run out of cash as early as mid-2027.

A year ago, Sam Altman stood beside the White House podium and announced the Stargate plan: $500 billion over four years, building the world’s largest AI data center network with SoftBank and Oracle. Trump called it “the biggest AI infrastructure project in history.”

A year later, this joint venture has neither assembled a team nor developed any data centers. The three partners haven’t even agreed on responsibilities. OpenAI itself can’t even build what it wants.

So, OpenAI started doing the math.

The Dream of $500 Billion Crumbles Over “Who Will Manage”

The Information’s report reveals a story of a year-long collapse behind the spotlight.

Weeks after the White House briefing, the Stargate project fell into paralysis. No one took the lead, no coordination mechanism. OpenAI, Oracle, and SoftBank repeatedly argued over who would build, manage, and how to split the costs.

OpenAI’s initial obsession was to build its own data centers. The logic was sound: leasing computing power long-term is too expensive; only building in-house can control the destiny.

But lenders saw it differently.

A company that burned through $2.5 billion in cash in six months and expects to spend $8.5 billion in a year asking for billions to build data centers? Lenders look at your cash flow, not your PPT. And OpenAI estimates it won’t turn positive cash flow until 2029 at the earliest.

It’s like a person who hasn’t started earning money applying for a mortgage to build a villa, and the first question from the bank is: What will you use to pay it back? And he can’t answer.

The self-build route is blocked. OpenAI was forced back to the negotiation table, continuing talks with Stargate partners.

But negotiations are tough. SoftBank has several large data center projects in Texas. OpenAI wants to use one of them as its first facility. SoftBank refuses, wanting to retain control. In September and October, OpenAI teams flew multiple times to Japan to negotiate face-to-face with Masayoshi Son.

The final outcome: OpenAI signs a long-term lease, controls the design; SoftBank’s SB Energy will develop and hold the facility.

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

$800 Billion Vanished

If internal chaos at Stargate was a hidden wound, this number is a public self-correction.

According to CNBC, OpenAI lowered its total compute expenditure target before 2030 from about $600 billion to around $600 billion, with a clearer timeline and revenue forecast. By 2030, revenue is expected to exceed $280 billion, split evenly between consumer and enterprise segments.

From $1.4 trillion down to $600 billion—a 57% reduction.

The official explanation: “To better align spending with revenue growth.”

The real meaning: investors are no longer buying it.

That previous figure was more like a wish list; $600 billion is at least a number that can be modeled. But even so, to reach over $280 billion in revenue by 2030, OpenAI would need a compound annual growth rate of over 50% for five consecutive years. Who can guarantee that?

In 2025, OpenAI’s revenue was $13.1 billion, with $8 billion burned. Profitability is still far off. The company estimates it won’t turn positive cash flow until 2029. Before then, cumulative losses could reach $115 billion.

This is the wake-up call.

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

The Books Can’t Support the Dream

Why is OpenAI forced to switch from a dreamer to a bean counter? Not because of strategic mistakes, but because three harsh facts arrived simultaneously.

First, money is flowing out much faster than it’s coming in.

In the first half of 2025, OpenAI’s revenue was $4.3 billion, burning through $2.5 billion in cash. Full-year revenue was $13.1 billion, with $8 billion burned. According to investor documents cited by Fortune, the company expects losses to grow annually, with operational losses possibly reaching $74 billion by 2028, and only turning positive around 2029 or 2030. Total losses could reach $115 billion.

OpenAI is currently spending ten times faster than it earns. Mathematically, these lines will eventually cross—either in 2029 or never.

Second, can compute efficiency offset scale expansion? Although OpenAI’s “compute profit margin” (revenue minus model operation costs) improved from 52% in October 2024 to 70% in October 2025, thanks to algorithm optimization and hardware utilization, each time they release larger models or more compute-intensive features (like video generation), these efficiency gains are eaten up.

Third, the paid conversion rate has plateaued.

ChatGPT’s weekly active users surpassed 900 million. But 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 signals that the subscription model has reached a ceiling—if you start charging for user attention, there’s only so far it can go.

Meanwhile, competitors are stealing users with less money. According to Similarweb, ChatGPT’s global traffic share dropped from 87% to about 65% in a year. Google Gemini, integrated by default into Android and embedded in Workspace, surged from 5% to 21%, not because of a better model, but due to distribution dominance. Anthropic’s Claude, with only 2% traffic share, achieves the highest user engagement (average 34.7 minutes per day), targeting high-end enterprise clients, and burns money at a fraction of OpenAI’s rate.

“ChatGPT established this category, but when substitutes appear, users naturally disperse,” said Tom Grant, VP of research at Apptopia.

And competitors are doing the same with less money. DeepSeek uses open-source models and ultra-low costs to stir the market. Google leverages distribution to crush. Anthropic employs a focused strategy to attract high-value customers. If AI models tend toward feature convergence, ultimately the market will be decided not by whose model is strongest, but by whose ecosystem is deepest and costs lowest.

OpenAI is trying to win three battles simultaneously: model competition, infrastructure race, and commercialization. But historically, no company has succeeded on all three fronts at once.

Altman’s Plan B

The dream is shattered, but Altman hasn’t stopped.

He did something recommended by every business textbook but rarely undertaken by dreamers: giving up obsession and living pragmatically.

The self-built data center dream is abandoned. Instead, the strategy is to sign numerous agreements outside the Stargate framework—annual $30 billion compute purchase deals with Oracle, deepening cooperation with CoreWeave, and even supplementing gaps with AWS and Google Cloud. Chip supply is diversified beyond Nvidia, including AMD and startups like Cerebras.

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

A year ago, this was unimaginable. Back then, Altman talked about trillion-dollar infrastructure commitments, 10GW of compute capacity, and a mission to change humanity with general AI. Now, his CFO talks about “protecting the balance sheet.”

But OpenAI’s fundraising remains staggering, with the latest round possibly exceeding $100 billion. According to Bloomberg, OpenAI is close to completing the first tranche of a new funding round, with the entire valuation possibly surpassing $850 billion. Participating investors include Amazon (expected to invest $50 billion), SoftBank ($30 billion), Nvidia ($20 billion), and Microsoft.

Note the nature of these investors: chip suppliers, cloud platforms, and strategic investors requiring OpenAI to use their services. This isn’t venture capital betting on a dream; it’s supply chain upstream and downstream locking in a major customer.

Investing in OpenAI used to be like buying a lottery ticket; now it’s more like signing supply contracts—completely different.

Gravity

Let’s bring the lens back to Stargate.

A year ago, on that White House stage, Sam Altman announced the $500 billion “Stargate” plan.

A year later, the joint venture in that plan has become a mess. OpenAI bypassed its own framework and signed a separate deal with Oracle. Compute targets fell short—only 7.5GW of the 10GW goal. Spending estimates were cut from $1.4 trillion to $600 billion.

This isn’t a story of failure. OpenAI hasn’t fallen. It still raises money, still grows, and still has over 900 million users.

But it is a story of waking up from a dream.

From “building the world’s largest data center empire” to “first stay alive, then fight using others’ money and infrastructure.” From wanting to be the landlord to becoming a tenant. From a dreamer to a bean counter.

When Elon Musk commented on the stalled progress of Stargate, he coldly tweeted: “Hardware is hard.”

Though harsh, this points to a harsh reality all AI companies will face: the arms race for compute power has reached a stage where the real threshold isn’t who trained the strongest model, but who can physically deploy gigawatt-scale infrastructure without burning out.

Altman chose not to burn out. It may be his most unglamorous but most prudent decision.

As for the $500 billion Stargate dream, it hasn’t died, but it is no longer what it was a year ago. It has shifted from a narrative about changing human destiny to a balance sheet that must be checked line by line.