OpenAI's "glossy facade" exposed: massive $12.3 billion loss, core executives fleeing, and even price cuts can't compete with Chinese models?

Just yesterday, OpenAI CFO Sarah Friar held an all-hands meeting and, for the first time, gave a clear timeline for going public.

The leading U.S. AI company plans to hold its initial public offering in 2027, and could move the IPO earlier if business continues to accelerate. The slides she presented at the meeting were full of impressive numbers: quarterly revenue run rate grew 35% quarter-over-quarter, enterprise business run rate grew 50%, and weekly active users for AI coding and office products surpassed 20 million.

But in reality, OpenAI's underlying concerns have already surfaced, showing up in several areas:

First, OpenAI's quarterly revenue was only $6.7 billion, up just 18% quarter-over-quarter, below some analysts' expectations. At the same time, operating losses widened sharply from $9.3 billion in the first quarter to $12.3 billion.

Second, Anthropic, long seen as the chaser, posted quarterly revenue of $11.6 billion, surpassing OpenAI for the first time in its history—and by a wide margin.

Third, since the start of 2026, roughly a dozen department heads have left OpenAI, with the densest wave occurring this month.

Fourth, although OpenAI announced significant price cuts on its models, Chinese model vendors are also offering "high-quality, low-cost" options. More and more U.S. companies are shifting work to DeepSeek or Kimi, further eroding OpenAI's customer base and potential clients.

Is the IPO a new round of fundraising?

At this meeting, OpenAI finally locked in its listing timeline. Friar was measured in her remarks: "The IPO is not the finish line—it's a milestone, another round of funding. We raised $122 billion in March, which gives us flexibility."

OpenAI has been preparing for its listing for a long time: in June, it confidentially filed its prospectus with the U.S. SEC; in March, it completed a $122 billion funding round at a post-money valuation of $852 billion; and just this Monday, it closed a $7 billion secondary share sale, allowing current and former employees to cash out at the $852 billion valuation.

According to earlier reports from multiple media outlets, Sam Altman has been pushing for a listing as early as the fourth quarter of this year, and has explicitly stated that a valuation below $1 trillion is "unacceptable." Friar held the opposite view—after carefully reviewing the company's spending commitments, she privately advised waiting until 2027, arguing that OpenAI is not yet ready to bear the strict disclosure standards of a public company.

So the external advisory team presented OpenAI's management with two options: list earlier but at a lower valuation, or wait until 2027 to secure the target valuation. It appears Altman ultimately accepted Friar's recommendation.

Friar also mentioned competitor Anthropic's listing at the all-hands meeting: "As you all know, we've filed confidentially, and Anthropic has filed too. They could unveil that confidential document in the coming weeks and go public as early as September. That's fine—we run our own race."

Anthropic's growth puts OpenAI in an "awkward" spot

That said, Anthropic's results are indeed striking. Two leading AI companies on the same track going public around the same time will inevitably be compared by the market. So relative to Anthropic's explosive growth, OpenAI's growth looks particularly awkward.

Anthropic's Q2 revenue was $11.6 billion, up roughly 143% quarter-over-quarter from $4.73 billion, and up more than 14-fold year-over-year from $787 million in the same period last year. This is the first time in Anthropic's history that it has surpassed OpenAI in quarterly revenue. As of the end of July, its annualized revenue run rate reached $65 billion, seven times higher than at the end of last year.

What's even more striking is the profit line: Anthropic reported adjusted operating profit of approximately $559 million, becoming the first frontier AI lab to announce a quarterly operating profit. Meanwhile, OpenAI posted a massive $12.3 billion loss for the quarter—a stark contrast.

But it must be emphasized that Anthropic is still a private company. These figures come from investor documents and informed sources, not audited public financials, and they are preliminary and subject to revision. More critically, there's the issue of accounting methodology: Anthropic's adjusted profit excludes stock-based compensation, and the exact calculation method has not been fully disclosed.

The two companies' definitions of "revenue" are also not entirely aligned. Some analysts point out that Anthropic counts end-customer spending through cloud distributors as revenue, rather than the actual revenue it receives from Amazon or Google, while OpenAI's reporting is closer to net figures. In other words, the $11.6 billion versus $6.7 billion comparison cannot simply be equated with a gap in competitiveness or market share. But the gap in growth rate is real—and growth rate is exactly what investors care about most when pricing an IPO.

A wave of core executive departures

An important backdrop to this all-hands meeting is that several of OpenAI's core executives have announced their departures, so Friar needed to provide a clear listing timeline to steady the ship.

Since the start of 2026, roughly a dozen department heads have left OpenAI, with the densest wave occurring this month. On August 11, Chief Operating Officer Brad Lightcap, who had been with the company for eight years, announced on X that he was leaving to "do something new." He had worked with Altman at Y Combinator and joined OpenAI in 2018, serving first as CFO for four years. His biggest achievement was expanding the go-to-market team from about 50 people to over 700 in eighteen months.

Two days later, on August 13, Chief Revenue Officer Denise Dresser announced her departure after just eight months on the job. She had joined in December from her role as CEO of Slack, and her successor is Dali Rajic, former president and COO of cybersecurity company Wiz. In the same week, OpenAI's ethics lead, safety lead, and an executive who called himself "Chief Futurist" also left.

Going further back, in July, the company's de facto number two, Fidji Simo, CEO of the applications business, stepped down due to a severe worsening of a chronic illness and transitioned to a part-time advisory role, with her duties taken over by co-founder and president Greg Brockman. In April, four people left at once, including Kevin Weil, vice president of the science division, and Kate Rouch, chief marketing officer, who departed to undergo cancer treatment.

OpenAI president Brockman tried to downplay the matter in media interviews: "We get so much attention that every departure gets magnified in a way it wouldn't otherwise." He stressed: "The company has gone through different eras, and each era has had a different leadership team. I'm the constant. Altman is the constant."

However, having both the COO and CRO leave within the same week, just as the company is preparing to face quarterly scrutiny from public markets, is likely to shake morale. PitchBook senior analyst Harrison Rolfes commented: "A CRO leaving in under a year is telling, because she was hired to turn product demand into a sustainable enterprise revenue organization, and that requires sustained effort. If this kind of repeated turnover continues, the organizational execution needed for an IPO will be harder to guarantee, and investors are probably watching more closely now."

Another pressure point comes from safety. Earlier, OpenAI disclosed that its autonomous AI agent broke out of its sandboxed environment during a cybersecurity test, breached the open-source platform Hugging Face, and retrieved test answers. The company subsequently paused training on some frontier models, expanded monitoring, and confirmed in its latest blog post that it would delay the release of new models like Astra.

In OpenAI's own words, "We've temporarily slowed the pace of expansion." In a race where release speed determines survival, hitting the brakes takes courage—but doing so in the same month as slowing growth, widening losses, and a competitor overtaking it doesn't look good.

Chinese models pressure the U.S.

There are many reasons for OpenAI's slowing growth: ChatGPT consumer demand has peaked (monthly active users hit 1 billion in June, with weekly actives around 900 million, leaving limited room for expansion), enterprise customer spending has turned cautious, and executives have been leaving in droves. But equally important is the price pressure from Chinese open-source models, which is now weighing directly on OpenAI's income statement.

On July 30, three weeks after the GPT-5.6 series launch, OpenAI announced significant price cuts on two non-flagship models: Luna's input price was slashed from $1 per million tokens to $0.20, an 80% cut, with output dropping from $6 to $1.20. Mid-tier Terra saw input fall from $2.50 to $2 and output from $15 to $12, a 20% cut. Only the flagship model Sol kept its price. Of course, this round of cuts also put Anthropic on the spot—Claude Sonnet 4.6's per-token price is now higher than the discounted Terra.

OpenAI's official explanation is "passing efficiency gains on to customers," but the timing says it all. In one week in July, Chinese models accounted for 57% of token volume called by U.S. enterprises on the OpenRouter platform—up from just 4.5% in the first half of 2025. Andreessen Horowitz's analysis is even more direct: 80% of U.S. startups are already running at least one Chinese model.

Ultra-high cost-effectiveness is the most direct reason. DeepSeek V4 Pro's output price is $0.87 per million tokens, Moonshot's Kimi K3 is $15, and Zhipu's GLM series is around $4.40. By comparison, Anthropic's flagship Fable costs $50 for output. The discounted Terra at $12 is still fourteen times DeepSeek's price.

Enterprises are indeed struggling with token bills, and Chinese open-source models give them a high-quality, low-cost option. Uber burned through its entire 2026 AI coding tool budget in the first four months, and its COO publicly stated that engineers' high token consumption is "increasingly hard to justify." The tech giants that were all-in on TokenMaxx earlier this year can no longer sustain the pace of token spending.

Citigroup reportedly shut off employee access to the most expensive models. DoorDash CTO Andy Fang confirmed that the company has shifted some "lower-tier" coding tasks to China's Kimi, with the result being "better quality at lower cost." AI startup Lindy has moved 100% of its traffic from Claude to DeepSeek.

Price cuts for growth: a double-edged sword

Price cuts can indeed drive usage and accelerate OpenAI's growth. But for a company still posting massive losses and rushing toward an IPO, it's a double-edged sword.

In fact, OpenAI's losses are widening faster than its revenue is growing. The $12.3 billion quarterly operating loss was 32% larger than the previous quarter, while revenue grew only 18% in the same period. This loss includes stock-based compensation. Looking back, OpenAI's full-year 2025 operating loss was $20.9 billion on revenue of $13.07 billion. Some estimates suggest cash burn could reach $27 billion in 2026 and be even higher in 2027.

It's worth emphasizing that the $12.3 billion operating loss reflects current operating costs—compute leasing to run ChatGPT, R&D personnel salaries, stock-based compensation—and does not include capital expenditures like buying land and building data centers. The two are accounted for separately; capital expenditures are typically amortized over years and not fully charged to quarterly operating losses at once.

So a more accurate interpretation is: OpenAI's widening quarterly loss is driven mainly by "marginal compute costs" for serving users rising with user growth, compounded by hefty R&D salaries—R&D spending alone hit $8.6 billion in Q1. Of the $12.3 billion quarterly operating loss, the main components are compute and labor costs on the operating side, not capital expenditures directly booked.

OpenAI plans to spend $665 billion on compute through 2030, which will continue to pressure its financials in the coming years. The Stargate project has announced planned capacity of nearly 7 gigawatts with over $400 billion in committed investment over three years, and there are reports of a data center partnership with Nvidia worth up to $250 billion. These contracts won't shrink just because Chinese models cut prices.

However, price cuts are a double-edged sword for OpenAI's IPO narrative: on one hand, lower prices can expand usage and retain price-sensitive customers who are preparing to migrate to Chinese models, because in the pre-IPO story, growth usually matters more than short-term margins. On the other hand, this directly compresses the gross margins investors are watching closely. Winning price-sensitive customers and proving to future shareholders that the company can make money are clearly in tension.

Three-way squeeze weighs on the listing

As the pioneer and leader of the generative AI era, OpenAI is at an exceptionally critical juncture, bearing heavy pressure from multiple fronts.

It just completed a $122 billion funding round in March, so it has no immediate cash crunch. Its annualized revenue run rate surpassed $40 billion in July, doubling from the end of last year. More importantly, enterprise business now accounts for over 40% of revenue, and coding products like Codex are growing rapidly, competing fiercely with Anthropic in the enterprise market.

By any tech company standard, this is extraordinary growth—unless compared to Anthropic. Yet OpenAI faces a three-way squeeze: above it, Anthropic's strong earnings reversal and sprint toward listing; below it, Chinese open-source models undercutting price-sensitive customers at a fraction of the cost; and in the middle, massive ongoing losses and hundreds of billions in compute commitments.

Price cuts are OpenAI's most effective response right now, but they don't solve the problem. Because the real pricing power in this price war is no longer in Silicon Valley's hands. If a free, downloadable, locally deployable Chinese open-source model can handle even 80% of a user's tasks, then every price cut OpenAI makes is moving it closer to a price anchor it cannot control.

Friar said the IPO is not the finish line, just another round of funding. That's not wrong. But for a company that will face quarterly public market scrutiny starting in 2027, the real question isn't when it goes public—it's whether, on the day it does, it can present a credible path to profitability.