OpenAI transitions from non-profit to public company, filing for IPO

On June 8, OpenAI filed a confidential S-1 with the SEC. Goldman Sachs and Morgan Stanley are the underwriters, targeting an IPO window in September, with private market valuations already reaching $730 billion to $850 billion.

A long-time observer of the industry stared at this news for a while and realized the most absurd part wasn't the valuation or the underwriter lineup—it was that this company still has the words "non-profit" in front of its name.

Nine years ago, when OpenAI was launched by Musk and Altman with pledges to "ensure AGI benefits all humanity" and "never be bound by capital," it seemed like a different animal. Now it's about to ring the bell on Wall Street.

But here's what needs to be said first: OpenAI didn't become "non-profit" today. It's been sliding toward profitability for nine years, inch by inch.

In December 2015, OpenAI was established. The charter was unambiguous: non-profit organization, mission to ensure AGI benefits all humanity, research results open source. Musk, Altman, Hoffman and others committed to raising $1 billion. Under the "non-profit parent company," OpenAI created a subsidiary called OpenAI LP—a "limited-profit" entity. In other words: investors could receive returns, but capped at 100x maximum. That July, Microsoft's first $1 billion arrived.

Note the timing: ChatGPT wouldn't exist for another three years. Even before seeing any commercial breakthrough, OpenAI already understood that donations alone couldn't sustain AI development.

When ChatGPT exploded globally in 2023, the valuation soared to $29 billion. Microsoft added investment, with rumors of $10 billion to $100 billion. But something became crystal clear to everyone: training large models costs ten times more than initially imagined.

By 2024, Altman began openly signaling to the board, employees, and the public: the for-profit division needed to be formally spun off as a Public Benefit Corporation (PBC), with the non-profit parent retaining only partial equity while the rest became fully marketized. Altman was decisive. He removed board members who opposed, managed dissenting employees, and brushed aside Musk's public criticism and lawsuits—and kept moving forward.

On October 28, 2025, the restructuring was complete. Microsoft's stake diluted from 32.5% to 27%. OpenAI's for-profit division formally separated from the "non-profit shell," becoming an independent company ready for IPO.

Here's an uncomfortable truth: the statement "non-profit OpenAI filed for IPO" is itself a false premise. What actually filed is the "for-profit entity" that began building from 2019, started splitting apart in 2023, and achieved full independence in 2025. The "non-profit" wrapper was always just a trust-building label.

To use an analogy: imagine a company selling $100 of goods while spending $300 to produce them, taking a $200 loss—then asking investors to value it at $20,000.

But most people who play on Wall Street aren't normal. In the March 31 funding round, OpenAI raised $122 billion at an $852 billion valuation—Silicon Valley's largest single private funding round, with Amazon alone committing $50 billion.

OpenAI's annual recurring revenue grew from $2 billion in early 2024 to $4 billion by year-end, then $10 billion by mid-2025, and $12 billion by July 2025. That's 6x growth in 18 months. No SaaS company has ever achieved this rate. Salesforce took seven years to go from $100 million to $1 billion. OpenAI did $1 billion to $10 billion in under a year.

So profitability doesn't matter—the speed of growth is all that counts. Grow fast enough, and losses are simply fuel.

Altman's narrative has shifted. It's no longer "we built ChatGPT." It's now: "AI is the next electricity. OpenAI is the next power grid."

Which country's power grid isn't unprofitable initially? They all are. But once built, there's no way around them. So the $600 billion in compute commitments, $50 billion Amazon investment, and $200 billion in future cash burn get reframed as "infrastructure investment," not "losses."

OpenAI's real competition comes from elsewhere. A book about Hassabis captured a revealing detail: the day OpenAI emerged in 2015, no one was more devastated than Hassabis.

Google had acquired DeepMind in 2014, and Hassabis negotiated unusually strict terms: Google couldn't interfere with research direction, Google couldn't bypass DeepMind's ethics committee. In essence, Hassabis wanted to close the door and methodically get AGI right.

Then OpenAI appeared and changed everything. By calling themselves "non-profit," Musk and Altman turned building AGI into an open arms race. What Hassabis planned to do in ten years got compressed into three. The Nobel laureate spent those years shuttling between Silicon Valley and London, fighting with Pichai, battling Google Brain for resources, forced from pure research toward Gemini—a "must-ship" commercial product.

During that period, Hassabis said something: "We didn't need to rush this much."

Now Altman faces the same pressure. Ten years ago, he used "non-profit" to force Hassabis to accelerate. Now Anthropic, xAI, and DeepSeek are doing it to him. The AI arms race has no off switch. Stop, and you're finished.

What Altman fears most now isn't losses—it's being overtaken. Go public first, secure $100+ billion in real capital, lock in compute capacity first, secure talent first, ship "GPT-X" first. Everything else can wait.

A $11.5 billion loss on an $850 billion valuation looks insane by traditional finance. But Altman isn't using a traditional financial lens. He's asking: there are only so many seats at the AGI table. Whoever claims one first becomes the next decade's Microsoft.

Current losses? Amazon's $50 billion and the IPO's $100+ billion are real money—someone else is paying.

OpenAI's current CFO Sarah Friar stated internally that the company lacked IPO-readiness conditions before 2026: compliance, process, organizational structure all insufficient, financial pressure extreme, with $200 billion needed to burn and a five-year commitment to $600 billion in cloud compute.

So why rush now? Multiple forces converge.

First, the Fed cycle. Interest rate cuts have progressed, money is still relatively loose. But ahead lie 2027 elections, tariffs, geopolitical tensions, AI regulation—any could slam the window shut.

Second, valuation. The $852 billion round is already funded; IPO pricing can only rise, not fall. Delay another year and if GPT-5 or GPT-6 don't represent qualitative leaps, valuation deflates. A valuation is a bubble; its value lies in cashing out before it bursts.

Third, competitors. Anthropic is already on the filing path, roughly the same timeline. If Anthropic goes public first to strong market reception, OpenAI going later gets directly benchmarked. If Anthropic flops, delaying hurts OpenAI more. The optimal play is getting there first, locking down the "AI's first major IPO" narrative.

Altman sees the window itself: miss now and the next three years may not offer a better one. Organization, compliance, losses—solve those post-IPO with market cap leverage.

China's internet industry knows this playbook cold. Luckin Coffee, Pinduoduo, Meituan, Didi—how many went public truly "ready"? All had to become ready *after* going public.

Except this is at $850 billion scale. If the forced cleanup fails, what crashes isn't one company. It's the entire AI valuation architecture.

When that happens, NVIDIA, Broadcom, TSMC, SK Hynix—everything riding OpenAI's rise—goes down together.

After stripping away the "non-profit," "open-source," "for humanity" mythology, consider what probably unfolds if OpenAI successfully goes public and raises $100+ billion in real capital.

Most significantly: the US-China AI valuation gap and frontier model performance differences will widen further, driven by capex and compute disparity.

Nine years ago, OpenAI's founding mission was to "ensure AI benefits all humanity." Nine years later, the company is heading to Wall Street to ring the bell, sell itself for $850 billion, and set off a cascade of consequences that no one can predict.