OpenAI's Government-Stake Proposal Puts Its Burn Rate Back Under Scrutiny

OpenAI has floated giving the U.S. government a 5% stake, worth roughly $42 billion to $43 billion, and pushed its planned public offering into 2027, two moves that put an old question back at the center of the AI debate, whether the ChatGPT maker can pay for everything it has promised to build. The stake proposal was first reported by the Financial Times, which said Sam Altman and other executives had suggested each leading American AI developer allot 5% of its equity to a public vehicle modeled on the Alaska Permanent Fund, the sovereign fund that pays residents a dividend from the state's oil wealth.

The proposal landed while a confidential draft prospectus sat unread. OpenAI filed its S-1 with the SEC on June 8 and has not made it public, so both its critics and its defenders are arguing from leaks, projections and partial figures rather than audited accounts.

Mallaby, Zitron, and the loss column

Sebastian Mallaby, a senior fellow at the Council on Foreign Relations, gave the sharpest version of the skeptical read on the July 10 episode of Prof G Markets. He stood by a January prediction that OpenAI would run out of money within 18 months and put the odds at one in two that, by next summer, the company cannot go public, cannot raise enough privately, and has to sell itself "at some sort of discount." He said its internally projected burn rate reached $660 billion over five years, and he read the $122 billion funding round OpenAI announced this spring as theater. Two-thirds of the headline number, he argued, was future promises or payment in kind, and the point of announcing it was to "head fake investors into putting more money in."

The independent writer Ed Zitron went further in a July 7 essay, Let AI Burn. He called the AI business "inessential to the economy" and argued the government stake "would only exist to prolong the inevitable." It is "not a bailout," he wrote, because a bailout has an endpoint at which the company no longer needs the money. Zitron also flagged a legal wrinkle other critics skipped. Under the Takings Clause of the Fifth Amendment, the government cannot simply seize equity, so it would likely have to buy the stake at a valuation it deemed "just," which would require congressional approval rather than a stroke of the pen.

Those arguments rest on numbers Zitron obtained and published from OpenAI's accounts. They show the company generating about $13 billion in revenue in 2025 against roughly $34 billion in costs, an operating loss near $21 billion, with research and development alone accounting for more than half of spending. The Wall Street Journal, which has reviewed internal projections, reported that OpenAI expects to burn $85 billion in 2028 even after doubling sales, and does not expect to generate more cash than it spends for at least four more years. Inference, the cost of actually running the models, exceeds half of revenue at both OpenAI and Anthropic, according to the same reporting.

OpenAI's own shares can't find buyers

In the same week OpenAI announced its record round, about half a dozen institutional investors tried to sell roughly $600 million of their OpenAI shares and could not find takers, Bloomberg reported. "We literally couldn't find anyone in our pool of hundreds of institutional investors to take these shares," said Ken Smythe of Next Round Capital, which handles such trades. Bids that did come in valued OpenAI around $765 billion, a 10% discount to its $852 billion primary mark, while buyers told the same brokers they had billions in cash ready for Anthropic at a premium. "It's just better risk-reward right now," said Adam Crawley of the marketplace Augment.

Inside the company, the caution was documented earlier. Chief financial officer Sarah Friar told colleagues in early 2026 that she did not believe OpenAI would be ready for a public offering that year, citing the organizational work involved and the risk from its spending commitments, The Information reported; Implicator earlier covered her warning about a late-2026 listing. By late June, The New York Times reported that OpenAI was leaning toward waiting until 2027, and that Altman treated any cut to the company's $1 trillion target as a nonstarter. Mallaby compared the bind to WeWork, whose 2019 offering collapsed once investors read the prospectus.

Peter Boockvar, chief investment officer at One Point BFG Wealth Partners, told Bloomberg after an April report that OpenAI had missed internal targets that he had come to see the company as too big to fail, "not from the perspective of a government backstop but because their tentacles have reached so wide in the data center ecosystem buildout." Jeff Park, an investor at ProCap, put the wager in scale. "OpenAI is now worth more than every S&P 500 company except 12," he said. "It has no profits. Those 12 companies have a combined age of nearly 1,000 years. OpenAI is 10."

A $1 billion raise that closed at $3 billion

OpenAI rejects the framing. After the April report, Steve Sharpe, its head of business and financial communications, called the account "clickbait" and said the business was "firing on all cylinders." Friar, in an on-record CNBC interview the same month, described demand that overshot the company's own plans. A retail placement that targeted $1 billion closed at $3 billion, the largest such placement the banks involved had ever run, with one bank's system breaking under the traffic. "Everybody wants to own part of a rocket company," she said. "I hope everyone wants to own part of ChatGPT." She gave the strategic reason for going public at all, which is to stop raising equity "forever" and tap cheaper debt against a compute bill she has put at $600 billion over five years.

The demand behind that bill keeps showing up in places where buyers can measure it. Uber capped each engineer's spending on AI coding assistants at $1,500 per tool a month after burning through its entire 2026 AI budget by April, yet roughly 10% of the company's committed code now comes from autonomous agents, chief executive Dara Khosrowshahi told investors. Asked on a podcast whether the surging usage connected to consumer-facing innovation, Uber's operating chief, Andrew Macdonald, was blunt about the gap. "That link is not there yet," he said. Uber's finance team could measure the cost to the cent but not the value per token, so it set a cap rather than pull the tools.

OpenAI argues its unit economics get better as it scales. In an April memo to shareholders, the company said each new generation of infrastructure "lets us train more capable models," while "algorithmic gains and hardware improvements reduce the cost to serve each token, lowering the cost per unit of intelligence." Its engineers have since cut inference costs by more than half on some workloads, and it is building a custom inference chip with Broadcom that Bloomberg reported could cut those costs by roughly another half.

Intel's government stake is up on paper

Mallaby reached for Intel, and the comparison is not the clean warning the critics want. When Washington took a roughly 10% stake in Intel last August for $8.9 billion, critics called it a bailout. By late April the holding was worth about $40 billion on paper. "This is the reason that the White House bought 10% of Intel," said Patrick Moorhead of Moor Insights, pointing to Intel as the only U.S. maker of leading-edge chips. Intel's chief executive, Lip-Bu Tan, told analysts that "the CPU is reinserting itself as the indispensable foundation of the AI era."

Mallaby drew the line at the rationale. Washington backed Intel, he said, because it does not want the most advanced chips made only in Taiwan, an island that "could be invaded by China." OpenAI, he said, is "just one of multiple American foundation model builders," and "we don't need OpenAI for any strategic reason." A colleague of his at the Council on Foreign Relations, Jonathan Hillman, has counted 30 American companies in which the U.S. government has taken or announced an equity stake since the Trump administration took office in January 2025.

What the confidential S-1 would settle

OpenAI filed its confidential paperwork on June 8 and has not said when, or at what price, it will let the public see it. Until it does, the argument runs on projections no one outside the company can audit. Mallaby cites a $660 billion burn, the Journal an $85 billion loss in 2028, and Zitron a $21 billion operating loss for last year, while OpenAI counters with $2 billion in monthly revenue and 900 million weekly users. The next hard data points are already scheduled. Nvidia's and SoftBank's next funding tranches are due this fall, and Anthropic, which filed its own confidential prospectus a week before OpenAI, may put the question to public investors first.

DeepSeek Still Wants AGI. Beijing Is Writing Part of the Check.

In at least one investor meeting this month, Liang Wenfeng told prospective backers that DeepSeek would keep developing open-source AI models while pursuing artificial general intelligence, according

The Implicator

DeepSeek Seeks $4 Billion in First External Round at $50 Billion Valuation

DeepSeek's valuation has climbed from $10 billion to as much as $50 billion in three weeks. The Hangzhou-based AI lab is now raising $3 billion to $4 billion in its first external round, with China's

The Implicator

True Footage Raises $40 Million to Build the Full Stack of Home Appraisals

True Footage, an Austin-based residential appraisal company, announced a $40 million Series C round on Monday, bringing its total funding past $105 million. Cox Enterprises led the round through its S

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai