OpenAI President Brockman: The company may be ready for an IPO by 2027
Greg Brockman rarely admits failure in public. But at the WSJ Tech conference on Wednesday, he made a rare concession.
"Q1 was pretty tough, and Q2 was also pretty tough," said the co-founder and president of OpenAI. "When it comes to programming, we weren't even in the conversation. We had Codex, but no one was talking about it." He was describing a period when the company was left behind by competitors. Anthropic's Claude Code captured 54% of the enterprise programming market, with annual recurring revenue surging from roughly $9 billion at the end of 2025 to $440 billion to $470 billion. In terms of overall share of enterprise AI spending, Anthropic's proportion soared from about 10% in early 2025 to over 65% in February 2026, overtaking OpenAI. Brockman made no attempt to hide this.
Then he shifted his tone: by the third quarter, "the situation was completely different". The launch of GPT-5.6 Sol and ChatGPT Work brought what he called a "revenue inflection point", with OpenAI's annualized revenue growing 70% quarter-on-quarter, its enterprise business revenue doubling, and the share of revenue contributed by enterprise customers jumping from the previous level of roughly 30% to a majority. He cited the example of JPMorgan Chase to illustrate what kind of clients the company is winning. When asked if he was ready for a 2027 IPO, he replied: "I think so. Yes."
This narrative has its elements of truth. An analysis by OpenRouter of about 120,000 enterprises that use tools from both companies shows that by September, the spending share between OpenAI and Anthropic had been "roughly equal"; while at the beginning of the year, Anthropic accounted for three-quarters of that share. The absolute size of enterprise revenue is also expanding: by the end of September, OpenAI's annualized revenue was close to $500 billion, up 77% year on year. Brockman's claim that "the enterprise business has really taken off" is not empty talk.
But "being ready for an IPO" is a far more complex judgment than "doubling enterprise revenue".
The growth in enterprise revenue masks the structural costs incurred for such growth
OpenAI's enterprise business is indeed accelerating. The problem is that the cost of this acceleration and its sustainability have been folded into the phrase "painstaking work behind the scenes" in Brockman's narrative.
A widely discussed but never finalized issue is the revenue accounting standard. OpenAI adopts the net method in cloud distribution arrangements, recognizing only its own share in transactions; Anthropic uses the gross method, recording the full amount paid by customers through cloud platforms as revenue. OpenAI argues that Anthropic's gross method has inflated its reported revenue by tens of billions of dollars. This dispute itself does not constitute a ruling against either party, but it points to a deeper vulnerability: when an AI lab's revenue is highly dependent on cloud giants that are simultaneously investors, computing power suppliers and channel partners, the boundary between "revenue growth" and "cost expenditure" becomes blurred. Every dollar of growth in enterprise revenue is also creating computing power orders and channel commissions for the same set of counterparties.
Brockman attributed the surge in enterprise revenue to "a series of previous investments finally paying off", but what OpenAI has not disclosed is how much of these investments are locked in the future in the form of committed expenditures. As of July 2026, OpenAI's committed cloud computing infrastructure expenditure up to 2030 is about $7.5 trillion; Bernstein, after reconstructing the model under the assumption that contracts are delivered on time, estimates that the actual expenditure is about $6.14 trillion, peaking at about $1.84 trillion in 2031. These figures are in the same order of magnitude as the $895.4 billion computing power commitment disclosed in Anthropic's prospectus. The growth in enterprise revenue is real, but it occurs against the backdrop that capital expenditure is also expanding in units of hundreds of billions of dollars. Doubling revenue and doubling bills are two sides of the same coin.
The private market has already re-priced, and the public market will only be more stringent
The most notable part of Brockman's speech is not what he said, but the context he did not mention.
A few days before he made these remarks, OpenAI's latest private financing target was lowered from $300 billion to about $200 billion, and its valuation target was reduced from $1.4 trillion to $1.25 trillion. By the end of October, the company had received about $110 billion in investment commitments, only 55% of the lowered target. Some investors have reduced their subscription size. This is not a marginal adjustment, but a directional signal: in the private market, the pricing of OpenAI has begun to swing back from "narrative premium" to "cash flow constraint".
The constraints of the public market will only be tighter than those of the private market. OpenAI's operating loss in the first quarter of 2026 was about $9.3 billion, cash consumption was about $3.7 billion, and the adjusted operating profit margin was about -122%; in the second quarter, revenue was $6.7 billion, but operating loss expanded to about $12.3 billion, which is equivalent to burning more than $1.8 for every $1 earned. Internal forecasts show that cumulative negative free cash flow from 2026 to 2030 will reach $2.78 trillion, and expenditure on computing power and infrastructure alone will reach about $8.56 trillion by the end of 2030 — exceeding the cumulative revenue forecast of $8.4 trillion over the same period. A research report released by HSBC in November 2025 has already warned that even by 2030, it will still be difficult for OpenAI to achieve profitability.
Brockman said he "has never been more confident", which holds true at the micro level of the enterprise business. But IPO pricing is not pricing for confidence in a single quarter. It is the discounting of free cash flow over the next decade. When the end point of the discount is still far out of reach, and the committed expenditure has been locked as an irrevocable obligation in the contract, the first question the public market will ask will not be "how much has enterprise revenue grown", but "how far is it from no longer needing external financing".
The organizational structure has cleared legal obstacles, but does not answer governance questions
The restructuring in October 2025 did pave the way for the IPO at the technical level. Previous investments by investors were uniformly converted into ordinary equity, the cap on shareholder returns was lifted, the for-profit entity was restructured into OpenAI Group PBC, and the non-profit OpenAI Foundation holds about 26% of the equity and has ultimate control, with Microsoft holding about 27%. From the SEC's perspective, the legal structure is in place. OpenAI even secretly submitted its S-1 draft as early as June.
But the distance between "being able to submit an application" and "being able to successfully price the offering" is not determined by legal documents. The arrangement that the foundation holds control can be regarded as an institutional guarantee for mission commitment in the private market; in the public market, it will be re-examined as a governance variable: when the controlling entity prioritizes non-profit missions, to what extent can the interests of minority shareholders be protected? Altman himself does not hold equity in the restructured company, which is a plus in the mission narrative, but an outlier that needs to be explained in the principal-agent framework of the public market. These are not negative questions, but they are unavoidable questions in the IPO roadshow. Brockman's "ready" answer addresses operational confidence, and the pricing logic at the governance level requires a different set of narratives.
Anthropic's IPO window has just been punctured, and OpenAI's 2027 is not a blank calendar
Brockman's mention of 2027 came right after Anthropic's IPO probability experienced a sharp drop.
On November 5, the Nasdaq fell by more than 5%, and the Philadelphia Semiconductor Index plummeted by more than 12%. On Kalshi, the probability that "Anthropic will submit an IPO application before November 30" dropped from 84% to 60%, and the probability before the end of the year fell from 90% to 74%. This sell-off took place after the midterm elections, where the Democrats took control of both the Senate and the House of Representatives, turning data center construction from election rhetoric into real uncertainty at the approval level. Dina Powell McCormick, President of Meta, put it bluntly at the same WSJ Tech event: after the election results were announced, building data centers will "become more difficult". OpenAI quickly released a data center policy proposal on Thursday, promising to terminate non-disclosure agreements with towns and pay the full cost of the electricity and infrastructure it uses. This is an AI company's immediate adaptation to changes in the political environment, but policy proposals can be written overnight, and the approval cycle and construction cycle of data centers cannot.
Anthropic's IPO is widely regarded as a demand test for the entire AI sector: if the public market is willing to accept a lab with quarterly revenue of $170 billion and nearly $900 billion in committed expenditures at a valuation of $1.5 trillion to $2 trillion, then OpenAI's listing path will have a referable anchor. Conversely, if Anthropic's pricing is poorly received, OpenAI will face not a question of "whether it is ready in 2027", but a question of "under what conditions and within what window can pricing be completed". Brockman said that the company will go public at the "right time", and this passive voice is accurate: whether the timing is right is not entirely determined by OpenAI's level of preparation, but also by how much patience the public market is willing to pay for the losses and commitments of AI labs at that moment.
Brockman's candor deserves to be taken seriously. He admitted falling behind in the first half of the year, admitted that Codex "no one talks about it", and admitted that the company was seen as a consumer-grade enterprise a year ago. These acknowledgments lend a certain degree of credibility to the narrative of the "Q3 inflection point". But when the president of a company says "we are ready" after private financing has shrunk by a third, after a competitor's IPO window has been punctured by the market, and when the company's cumulative funding gap is measured in hundreds of billions of dollars, the weight of this statement far exceeds the scope supported by "doubling enterprise revenue".
2027 is not a date. It is a condition. Whether the condition is met is not up to Brockman, but up to the public market.
This article is from the WeChat official account "Tech Business", written by Zisen and Bowen, and published with authorization from 36Kr.