OpenAI Eyes 5% Stake for Government Support: Mutual Gain or Backdoor Bailout?

A Headline That Keeps Getting Reinterpreted

According to Bloomberg, OpenAI is considering offering the Trump administration roughly 5% of its equity to ease regulatory pressure from Washington. You might have missed the directional nuance in that framing — this isn't the government demanding a stake; it's the company voluntarily offering one in exchange for political support and cooperation.

Understanding this move requires some background on OpenAI's corporate structure. Founded in 2015 as a nonprofit, OpenAI introduced a "Capped Profit" subsidiary structure in 2019, allowing outside investment while limiting returns. In late 2024, OpenAI announced plans to convert fully into a for-profit company — a process involving complex equity restructuring. Questions remain unresolved: how does the nonprofit parent's oversight transition, and how are the rights of original donors protected? It's precisely during this conversion window that a government equity stake becomes operationally feasible. Offering 5% to the government can be read as trading political cover for regulatory leniency during the transition.

Bilibili creator Chris Norlund offered a pointed analysis of this development in his latest video. He highlighted a recent precedent: Trump had publicly criticized Intel's management and job losses in harsh terms — but after the U.S. government took a stake in Intel, the president's tone toward the company immediately turned positive. The contrast makes clear just how valuable the political dividend of giving up 5% could be.

What's more, this may not be limited to OpenAI alone. There are proposals requiring all tech companies to contribute 5% of their equity into a domestic sovereign wealth fund. A Sovereign Wealth Fund (SWF) is a state-owned investment fund — think Norway's Government Pension Fund or Singapore's Temasek. The U.S. has never had a federal-level SWF, making the Trump administration's push to create one a significant policy shift. Forcing all tech companies to cede 5% would face substantial legal obstacles: antitrust scrutiny, existing shareholder agreements, and the challenge of defining what counts as a "tech company" — which is precisely why such a broad mandate would face far greater resistance than a single company's voluntary arrangement.

Sharing Upside, or Sharing Downside?

At the heart of this potential deal is a fundamental question: is this about sharing profits, or about getting the government to eventually share the company's risk?

OpenAI CEO Sam Altman has leaned into the former narrative. In a Financial Times op-ed, he argued that AI models are trained on the accumulated output of society — news coverage, cultural content, and more — and that everyone deserves a share of the returns. This argument touches on one of the most contentious legal battles in the AI industry: the New York Times, Getty Images, and others have already sued OpenAI and Microsoft over training data copyright, with the core dispute being whether using copyrighted content to train AI models constitutes "fair use." Altman's move to frame this legal gray zone as a moral case for "shared benefits" is a rhetorical strategy that converts potential legal liability into a policy proposal — and if the government holds equity, it would face a conflict of interest in any future copyright litigation. Polling data suggests the public does harbor similar expectations.

Chris Norlund offers a colder read: the real goal is to secure smoother regulatory approvals and a clear political runway. And if the AI boom is fundamentally a bubble, pulling the government in means the government absorbs part of the eventual crash. In his words, this is essentially a request for a government bailout.

He also noted an interesting detail: Altman has recently been repeatedly calling for an "AI council" and a formal regulatory framework in his public writing. This is interpreted as Altman wanting to be the first inside the room when the rules are written — so he can shape regulations that benefit his company while raising barriers for competitors.

The Burn Rate and the Margins: A Balance Sheet That Doesn't Add Up

Another major thread in this debate is whether these AI giants can actually make money.

Citing Financial Times data, Norlund noted that OpenAI burned through approximately $20.9 billion in 2025; another frequently cited figure puts revenue at around $13.07 billion against losses of $38.5 billion. From this, Norlund draws a key conclusion: the problem with these companies is that costs scale nearly linearly with revenue — there's no evidence that margins can improve, regardless of how many specialized chips are deployed or what the next-generation architecture looks like.

The video also presents the bull case. Wall Street analyst Dan Ives argues that hyperscalers' $700 billion in investment is "underpinning the AI revolution," and that we're simply in the first phase of infrastructure buildout — like the early days of constructing the Las Vegas Strip, with monetization inevitably coming later. This echoes the classic "picks and shovels" logic in tech investing, and recalls similar defenses during the dot-com bubble (1995–2001): the fiber optic infrastructure laid by telecoms sat largely idle after the crash but did eventually underpin the streaming and cloud computing booms two decades later. The critical difference, however, is that fiber's marginal cost could be amortized at scale once demand arrived, whereas AI inference costs today remain nearly linear with usage — no significant economies of scale have emerged yet. If AI infrastructure economics differ fundamentally from fiber, the "first phase" analogy may be flawed at its core. Ives also noted that Microsoft has largely captured the enterprise market, with roughly 5% of customers now on an AI transformation path.

Norlund isn't buying it — and he notes a telling coincidence: the analyst who keeps championing AI's future is about to leave his current role to launch what he calls a "modernized merchant bank." The timing is cited as a subtle signal of wavering confidence within the industry.

Karp's Sardonic Observation: Why Enterprises Resist Paying for AI

Palantir CEO Alex Karp's comments on Squawk Box became a viral talking point.

With a touch of irony, Karp described the prevailing corporate attitude: companies feel they're "wasting time on tokens, getting no value, while their assets are being taken." To understand this resistance, it helps to know how AI is currently billed. A token is the basic unit a large language model uses to process text — roughly corresponding to a word or sub-word fragment. Most AI services charge by token consumption rather than by task outcome. This pricing logic stems from the marginal cost structure of inference: every generated token consumes GPU compute, a quantifiable cost, but output value is difficult to standardize. Compared to traditional software seat licenses or SaaS feature subscriptions, token-based billing is highly opaque to non-technical decision-makers and makes it extremely difficult for enterprise buyers to build reliable ROI models.

Norlund's take: enterprises resist because these AI companies don't charge based on outcomes or results — large models inherently hallucinate and cannot realistically offer outcome-based guarantees. So the model becomes: encourage users to consume as many tokens as possible, use their code, data, and ideas to train the model, then pocket most of the benefit.

He further questions the quality of these products, citing an observation that both Anthropic's Dario Amodei and OpenAI's Altman love saying they "can't wait to see what people build with this" — which, read in reverse, suggests even they don't know what value this technology will ultimately create. They're effectively outsourcing innovation experimentation to society at large.

The Political Math Behind the "Nationalization" Narrative

Karp also raised a much bigger topic on air — claiming that for the past six months, he's been warning industry titans that "we're going to get nationalized," and that while no one believed him at first, support for nationalization is now growing.

Norlund views this "nationalization" narrative with considerable skepticism, especially given Palantir's business profile. Founded in 2003, with the CIA (through In-Q-Tel) among its early backers, Palantir's core products — Gotham and Foundry — are designed for government intelligence analysis and enterprise data integration. U.S. government contracts still accounted for roughly 55% of its total revenue in 2023. Palantir is structurally one of the biggest beneficiaries of the deep fusion between AI and state power; its surveillance and tracking capabilities only become more valuable as government expands its reach. When Karp positions himself as a prophet warning about nationalization risk, he is simultaneously one of the biggest potential winners of that scenario — deeper government involvement in AI could expand Palantir's market. So when those at the top of the food chain mix nationalization talk with public-interest framing, their real motivations deserve scrutiny.

Norlund also runs the numbers from an investor perspective: if the equity dilution isn't borne by Altman personally, existing investors' stakes get diluted by this deal. Unless OpenAI can clearly tell investors "we secured X approval through this, which will generate Y additional revenue," the arrangement is hard to justify.

There's also the subtler risk of political cycles. Norlund notes that the more OpenAI ties itself to Trump, the greater its exposure — the next president is likely a Democrat, and every entity deeply associated with Trump could face a reckoning. In his view, OpenAI is "voluntarily painting a huge target on its own back."

Conclusion: A Public Standoff Over Bubbles and Bailouts

Taken together, the debate over OpenAI's potential equity deal reflects three overlapping anxieties in the AI industry: whether the business model can ever close, how to navigate regulatory and competitive pressure, and who ultimately absorbs the bubble risk.

Whether it's Altman's "shared benefits" narrative, Karp's "nationalization warning," or analysts' "first phase of infrastructure" defense — all of them point to the same unresolved core question: with massive ongoing losses and no clear path to margin improvement, these companies are trying to achieve certainty by anchoring themselves to the government. Whether that anchoring is a genuinely win-win institutional arrangement or a backdoor bailout remains very much an open question.

A note on sourcing: the views in this article are largely drawn from and attributed to a single creator's video analysis, which reflects strong personal opinions (including characterizations like "bubble" and "bailout"). Financial figures cited originate from public reporting by Bloomberg and the Financial Times. Readers are encouraged to consult a broader range of independent sources before drawing conclusions.

Key Takeaways