OpenAI is in talks with investors over a new private round at a valuation of about USD 1.2 trillion. Whether that number holds depends on whether its initial public offering, now pushed to 2027, can clear the same level. The proposed valuation would add roughly USD 350 billion to the USD 852 billion post-money mark set by its USD 122 billion round in March, the largest private financing in Silicon Valley history.
The move exposes the central contradiction of the AI build-out: revenue compounds, losses compound with it, and the model needed to justify USD 1.2 trillion leaves almost no room for error. With annualised revenue running near USD 25 billion and 2026 losses forecast in the tens of billions, the market is being asked to price the company like a mature technology giant whose cash burn has no precedent among listed firms.
OpenAI filed a confidential registration statement with the US Securities and Exchange Commission between late May and June, with advisers initially targeting a listing as early as the third quarter. By late June the company leaned towards pushing the offering to 2027. Chief executive Sam Altman has told investors he would not take the company public below a USD 1 trillion valuation, a threshold his team treats as a floor. Advisers put two paths to management: list sooner at a lower valuation, or wait until 2027 and aim for up to USD 1 trillion. A USD 1.2 trillion private round effectively forces the second path. One investor in both OpenAI and Anthropic said subscribing to the latest round requires assuming an IPO valuation of USD 1.2 trillion or more, meaning the round is not just new money but a bet that public markets will accept unprecedented pricing for a company losing tens of billions a year.
The stakes for existing backers are large. Microsoft has invested more than USD 13 billion for about 27 per cent, the largest single institutional stake. Nvidia has committed about USD 30 billion, mostly in compute credits, part of more than USD 40 billion it has poured into AI labs including Anthropic. For both, a trillion-dollar listing would convert their holdings into two of the most valuable strategic positions in corporate history and validate the AI infrastructure thesis behind Azure, Copilot and graphics chip demand.
SoftBank anchored the March round and took a USD 40 billion bridge loan to fund its commitment, an arrangement that carries its own time pressure. That round was staged: SoftBank’s USD 30 billion commitment deploys in three quarterly tranches through 2026, while a substantial part of Amazon’s USD 50 billion anchor is said to depend on OpenAI listing or hitting an artificial general intelligence milestone. Nvidia’s USD 30 billion is mostly compute credits offsetting GPU infrastructure spending rather than cash. A large share of the headline USD 122 billion is conditional, deferred or vendor-linked, a nuance that matters when markets ask what actually backs the valuation.
Revenue growth is the bull case in its purest form. Annualised revenue reached about USD 25 billion by February 2026, up roughly 92 per cent over the prior twelve months, with first-quarter revenue of USD 7.5 billion pointing to a full-year target of USD 30 billion. Other estimates put the run rate higher, near USD 40 billion by mid-2026 against about USD 20 billion at the end of 2025. Enterprise is the engine, at more than 40 per cent of revenue and expected to match consumer by the end of 2026.
The loss trajectory is the counterweight. OpenAI’s first-quarter 2026 operating loss was about USD 9.3 billion, widening to roughly USD 12.3 billion in the second quarter, with full-year operating losses guided at USD 27 billion to USD 33 billion. Including fair-value charges, some forecasts show GAAP losses above USD 60 billion. Inference cost, the expense of actually running models, reached USD 8.4 billion in 2025 and is expected to climb to USD 14.1 billion in 2026. The company plans 30 gigawatts of compute capacity by 2030, has locked in eight of those gigawatts and says its nearest competitor will not reach that level until late 2027.
Operations are more nuanced than the spending headlines. Gross margin improved to about 39 per cent in the first quarter of 2026 from about 33 per cent a year earlier, as model efficiency cut the cost of each query. The problem is not that unit economics are collapsing but that compute and talent spending rise with revenue. Every efficiency gain is immediately reinvested in the next capacity and research cycle, so operating losses widen even as per-query margins improve.
On USD 25 billion of annualised revenue, a USD 852 billion valuation implies a price-to-sales ratio near 34 times. If USD 30 billion of revenue meets a USD 1.2 trillion valuation, the multiple is 40 times. Even on the more optimistic USD 40 billion run rate it is 30 times. No listed software company sustains such a multiple without a credible path to profit, and OpenAI’s losses are expected to widen over the next year rather than narrow.
On reach, OpenAI still dominates. By February ChatGPT had 900 million weekly active users and crossed one billion monthly actives in May, with more than 50 million consumer subscribers and over nine million paying enterprise users. The company says 92 per cent of Fortune 500 firms now use ChatGPT.
The growth narrative is under strain. Sensor Tower data shows ChatGPT’s share of global AI assistants fell below 50 per cent in May to about 46 per cent, down from 87 per cent in early 2025. Menlo Ventures’ mid-2026 enterprise report shows Anthropic overtaking OpenAI for the first time in enterprise large language model API spending, at 34.4 per cent against 32.3 per cent. Anthropic’s annualised revenue passed USD 30 billion in April 2026, beating OpenAI, and its second-quarter revenue alone exceeded USD 11.5 billion. The pressure has driven visible strategic churn inside OpenAI, which revised its product roadmap twice in six months, first against Google and then against Anthropic, and dropped projects including the Sora video launch and an adult chatbot. Some investors have told the company that such frequent turns risk scattering its focus just before a listing. One early backer said the company has a billion-user business growing 50 to 100 per cent a year and still talks mostly about enterprise and code, which is a badly distracted firm. Jai Das, president of Sapphire Ventures and an investor in neither company, went further, calling OpenAI the Netscape of AI, a reference to the browser that once dominated before Microsoft routed around it and AOL bought it.
Management pushed back hard. Chief financial officer Sarah Friar cited the USD 122 billion round as proof of confidence, with backers including SoftBank, Amazon, Nvidia, Andreessen Horowitz, Sequoia and Thrive Capital among more than 25 investors. She said the claim that investors do not support the strategy contradicts the facts, and that the round was the largest ever, oversubscribed, closed at record speed and supported by a broad global investor base.
The question investors must settle is whether OpenAI’s losses are cyclical expansion costs that fall away at scale or a structural feature of the foundation-model business that cannot self-correct. The evidence points to structural, and that distinction is why USD 1.2 trillion is hard to support. Cyclical losses have three markers: short-term drivers such as inventory build or one-off capacity expansion, a proven mean-reversion pattern, and at least three comparable historical cycles. OpenAI has none. Its losses are driven by two reinforcing forces. Inference cost scales with usage, so every added user and every prompt raises compute spending, meaning revenue growth itself pushes costs up. And the company has locked a fixed cost base through its 30-gigawatt commitment, a strategic decision that must be honoured regardless of demand. That is not temporary expansion but a cost floor above any plausible near-term profit.
This is also not a simple invest-first, monetise-later story with a visible inflection. Even as gross margin improved from 33 to 40 per cent, the company reinvested every point of efficiency into the next model and capacity. In the three years since ChatGPT launched, revenue grew exponentially and losses grew with it. The pattern has not reverted, it has intensified. A cyclical read needs evidence that spending will flatten while revenue compounds, and the 30-gigawatt roadmap points the other way.
What the market has not fully priced is the second-order effect. A USD 1.2 trillion valuation bets not only on OpenAI’s revenue growth but on the entire AI capital expenditure cycle expanding without a demand shock. If enterprise AI spending slows, the enterprise business now at 40 per cent of revenue cannot keep compounding, and the fixed cost base will not shrink with it. In that scenario losses widen as growth decelerates, revenue multiples compress, and this private round becomes the high point rather than a springboard. The risk is asymmetric, since private rounds can mark valuations up fast and public markets can mark them down just as fast when the growth-to-spend trade reverses.
The strongest bull case is simple: every previous platform shift, from personal computers to the internet to smartphones, produced one dominant company worth far more than sceptics thought possible. ChatGPT’s 900 million weekly users and 92 per cent Fortune 500 penetration look less like a bubble and more like infrastructure-grade adoption. If OpenAI becomes the operating system for enterprise knowledge work, USD 1.2 trillion is not expensive, it is the entry price. Microsoft’s 27 per cent and Nvidia’s USD 30 billion are votes that this is the destination rather than a fantasy.
The bear case is that OpenAI is a commodity layer in a crowded, well-funded market. Anthropic is growing faster in enterprise. Google is recovering. Model capability is converging, pushing the industry towards price competition, and price competition is fatal when the cost base is measured in gigawatts. The Netscape comparison bites precisely because Netscape was also the dominant platform of its moment with apparently unassailable adoption, until the browser became a free feature bundled into an operating system.
The falsification signals are quantifiable. If gross margin fails to expand above 50 per cent, if operating losses do not narrow below half of revenue within four quarters while revenue growth stays above 50 per cent year on year, then the losses are structural rather than cyclical and the USD 1.2 trillion valuation will not hold. Margin expansion alongside sustained growth is the only proof that scale economies are real. Without it the reinvestment cycle never ends and multiples must compress.
In the near term, the round and the IPO delay are a bet on time. Waiting until 2027 buys OpenAI another year to show enterprise revenue matching consumer, to prove margin improvement, and to let the loss trajectory fade in investors’ rear-view mirror. It also gives Microsoft and Nvidia another year of mark-to-market gains on stakes that remain illiquid, and gives SoftBank time to deploy its staged commitment without the pressure of an imminent listing.
The base case is a 2027 IPO at USD 850 billion to USD 1 trillion, with the USD 1.2 trillion private round as an anchor that makes a discounted raise politically impossible for management. The bull case is accelerating enterprise AI spending, a margin inflection above 50 per cent, and a listing above USD 1.2 trillion that makes today’s sceptics retreat. The bear case is slowing AI capital expenditure demand meeting a rigid cost base, forcing a private valuation reset below USD 852 billion or a flat first-day open. Watch whether the quarterly revenue run rate tracks the USD 30 billion to USD 40 billion target, whether gross margin breaks 40 and then 50 per cent, whether enterprise revenue catches consumer, whether AI assistant traffic share stops falling, and the pace of SoftBank’s staged funding. If any two of those turn negative for two consecutive quarters, the USD 1.2 trillion figure is a private-market artefact, not a public-market reality.
OpenAI’s valuation is not based on current profit, or even next year’s expected profit. It rests on a belief that the AI build-out has no brakes and that the capital expenditure cycle, the model race and the enterprise adoption curve will all climb together. What USD 1.2 trillion is really pricing is that belief rather than a set of financial statements. And belief, unlike revenue, can change its mind.
Editor’s note: This is an adapted translation of the original Sohu Technology report. It has been trimmed and restructured for readability for an international business audience.