Regarding the timing of the IPO, there is an internal disagreement at OpenAI: Altman hopes for the fourth quarter, while the CFO believes it is not yet ready.

Does a shift in AI CFO power suggest a change in the company’s strategic focus?

Internal tension at OpenAI between an aggressive expansion path and financial prudence is starting to come into view.

According to The Information, CEO Sam Altman privately said he hopes to complete an IPO as early as this year’s fourth quarter, while CFO Sarah Friar earlier this year shared a sharply different view with colleagues, believing the company is not yet ready to list.

Friar’s concerns point straight to the company’s financial reality: OpenAI has committed to spending more than $600 billion over the next five years to rent servers, while warning investors that its cash burn through 2030 will exceed its prior forecast by more than double—meaning the pace of spending could surpass $200 billion.

A person who spoke with Friar said she still has doubts about whether slower revenue growth can support such a spending commitment. At the same time, competitor Anthropic is eroding OpenAI’s market share, further intensifying external pressure.

In response to reports that a rift has emerged in the relationship between the two, Altman and Friar jointly released a statement saying the two are “completely aligned” on the point that “securing compute for the long term is OpenAI’s core strategy,” and stated that over the past year and more, both have “directly participated in every major compute decision.”

Differences over the IPO timeline

Despite Friar’s reservations, OpenAI has already begun work on preparations ahead of an IPO. The company has retained two law firms—Cooley and Wachtell Lipton Rosen & Katz—and has entered into informal discussions with the IPO underwriting teams at Goldman Sachs and Morgan Stanley.

Altman’s desire to go public also carries clear competitive motivation. He privately said he wants OpenAI’s IPO to happen before Anthropic completes its own, and that the latter is currently discussing going public in the fourth quarter of this year. If the IPO proceeds, the deal could rank among the largest IPOs in history.

Friar’s reluctance to rush into a listing is not without precedent. Last November, in an interview with The Wall Street Journal, she publicly said the IPO is “not currently on the table,” because the company is still working to “adjust the company to a state that matches its current scale.”

The internal power structure is quietly changing

At the organizational-structure level, an unusual change has already taken place: since August last year, Friar has stopped reporting directly to Altman, and instead reports to Fidji Simo, the head of the applications business. It is rare in large companies for the CFO to report directly to executives other than the CEO.

Multiple people who work closely with the two said Altman has excluded Friar from certain important discussions involving financial planning.

One example is that Altman recently discussed server spending with a major investor in OpenAI without inviting Friar, whereas she had attended a meeting on the same topic earlier.

A participant said her absence was “noticeable and awkward.” Another person said that earlier this year, at an executive-level meeting involving major financial decisions, Friar was likewise not invited.

The high-priced compute commitments hide financial risk

OpenAI’s current compute spending plan is unprecedented in scale.

OpenAI has signed server lease agreements totaling about $665 billion, covering:

Oracle (about $300 billion, five-year term, starting in 2027);

Microsoft ($250 billion, through 2032);

Amazon Web Services ($138 billion, eight-year term);

CoreWeave ($22 billion, five-year term);

Cerebras ($10 billion) and several other partners.

These commitments are not ordinary cloud computing contracts. Friar has explained that AI data center construction cycles last several years, so OpenAI must reserve capacity in advance.

“I have to make a decision today to ensure we have enough compute in 2028, 2029, and even 2030—if I don’t place orders today, the data centers won’t exist.”

In one case, as The Information previously reported, OpenAI signed a risk-sharing agreement with Oracle regarding cost overruns for data center construction—an unusual term in cloud customer contracts. At present, OpenAI has shelved its previous plan to build its own data centers.

Pressure from both slower revenue growth and higher cash burn

The external competitive landscape is tightening faster. Anthropic has surpassed OpenAI in selling AI models to enterprises and application developers, and Google’s Gemini continues to erode ChatGPT’s dominant position in the consumer chatbot market.

OpenAI raised its revenue expectations for the next five years by 27% this year, but in the same period it privately told investors that its cash burn through 2030 will be more than twice the forecast made last summer.

In addition, the company informed investors that last year’s gross margin was lower than expected because user demand exceeded expectations, forcing the company to temporarily procure compute at higher prices.

In February this year, Anthropic cofounder and CEO Dario Amodei was direct about the dangers of investing too far ahead in data centers during an interview on the Dwarkesh Patel podcast—words that strongly implied OpenAI:

“If I misjudge—by even just a year—you go bankrupt. My impression is that some companies haven’t really done the math seriously, and they don’t truly understand the risks they’re taking on.”

This closely echoes the concerns that have surfaced internally from Friar. A person who works closely with both described the CFO’s situation this way: “She’s dealing with a founder who has big ambitions and wants to go all-in on spending. That’s a very hard job.”