As OpenAI and Anthropic race toward public listings, their financials tell opposite stories. One company has found AI profitability through enterprise discipline. The other is still asking markets to believe it will. NYSE in New York - Photographer: Michael Nagle/Bloomberg
© 2025 Bloomberg Finance LP
This week, the two world’s most valuable private AI companies moved toward public markets within days of each other. OpenAI filed confidentially for its IPO, working with Goldman Sachs and Morgan Stanley toward a listing as early as September 2026 at a valuation above $1 trillion. Simultaneously, Anthropic, as part of its latest funding round, disclosed to investors that it projects $10.9 billion in revenue for the second quarter of 2026, more than doubling Q1’s $4.8 billion, and expects its first-ever operating profit of $559 million for that period. That one-two punch frames the debate about AI profitability with one company asking public markets to fund additional years of mounting losses, while the other is arriving with a profitable quarter already in hand.
Investors keep reaching for Amazon as a point of comparison. The company lost billions for years before becoming one of the most profitable in history and its template of spending aggressively, capturing the platform shift and harvesting the returns is used as the basis for the AI bull case. The analogy is misleading and the numbers behind it show where AI profitability will come from. Amazon accumulated roughly $3 billion in cumulative losses over six years before turning its first annual profit in 2003. OpenAI is on course to accumulate hundreds of billions in losses before reaching positive cash flow around 2029 or 2030. The scale differs by a factor of 100, the cost structure differs in kind and the path to profitability, for those who will find it, runs through enterprise, not consumer adoption.
As I wrote last December, OpenAI’s infrastructure ambitions were already straining credibility against its revenue base. The company was exiting 2025 with roughly $20 billion in annual revenue while committed to $1.4 trillion in infrastructure spending, since revised down to approximately $600 billion through 2030, according to CNBC. Even the trimmed figure dwarfs anything a comparable-stage technology company has ever committed to operational spending.
In April 2026, Anthropic surpassed OpenAI in annualized revenue run rate, reaching $30 billion, up from $1 billion fifteen months earlier. More significant than the run rate itself is the structure behind it. Approximately 85% of Anthropic’s revenue comes from enterprise and developer customers. OpenAI's mix runs in the opposite direction, with roughly 85% tied to ChatGPT consumer subscriptions and roughly 95% of those users paying nothing. OpenAI's computing expenditure will reach $121 billion in 2028 alone, with a projected loss of $74 billion that year. Anthropic, by contrast, projects $17 billion in positive cash flow in 2028 on $70 billion in revenue, with gross margins approaching 77%.
The divergence traces to the client mix. Enterprise customers generate three to five times more revenue per token than consumer users, their query patterns are more deterministic and therefore cheaper to serve and their contracts are sticky. Over 500 companies now spend more than $1 million annually on Anthropic’s Claude platform and eight of the Fortune 10 are customers. That is the foundation of a profitable business. A free-tier consumer base of 900 million weekly users generating massive inference costs without proportionate revenue is not.
MORE FOR YOU
The IPO filings now in motion will force a reckoning that private markets have long deferred. OpenAI is preparing to ask public investors to value the company at more than $1 trillion while projecting a $14 billion loss for 2026 and no profitability before 2029 or 2030. The listing will test how much faith investors still have in the AI boom, with some already questioning whether generative AI can deliver returns that justify the trillions being poured into the sector. OpenAI’s own CFO, Sarah Friar, has flagged reservations about the timing, noting the company is not yet ready for the scrutiny of public markets.
Anthropic’s IPO story is materially different. The Q2 operating profit figure includes model training costs, the expense most often cited as the structural barrier to AI profitability. The milestone arrived two full years ahead of what Anthropic had told investors last summer. The important qualification is that the operating profit projection excludes stock-based compensation, which at a company that has raised billions in private capital could be significant enough to erase the margin on a GAAP basis. Anthropic may also not stay profitable for the full year, given planned spending increases on compute and model training. What the quarter demonstrates, however, is that the business model can generate operating profit at sufficient revenue scale. That is a valuable thing to show to the public markets.
The issue for OpenAI is structural. Amazon survived its loss years because it generated positive operating cash flow almost throughout, since customers paid before suppliers, keeping the company solvent without continuous capital raises even during the dot-com crash. OpenAI has no equivalent mechanism. Its burn is tied to the operational cost of running inference at scale and is dependent on continuous access to private capital at a pace without precedent. Amazon raised approximately $8 billion over its lifetime. OpenAI, only in its latest round in March, has raised $122 billion in private capital and still needs more to fuel its data center build plans. HSBC analysts have estimated a $207 billion funding shortfall against its growth plans. Public markets, with their quarterly earnings cadence and analyst scrutiny, are a less forgiving environment for that kind of structural gap than the sovereign wealth funds and private equity firms that have funded the AI build-out so far.
As I noted in January when predicting AI’s trajectory in 2026, the era of pure scaling is giving way to something more disciplined: enterprise buyers demanding measurable returns and selective capital deployment replacing the build-first posture of the prior three years. That transition is showing up in the financial divergence between these two companies. When S-1 filings put actual numbers in front of institutional investors, the divergence will be impossible to ignore.
Anthropic’s trajectory toward 77% gross margins by 2028 is closer to the economics of enterprise software than to AI infrastructure. The relevant IPO comparison may be Salesforce or ServiceNow, not Amazon's retail operation. OpenAI's public market case, by contrast, rests on the thesis that agentic AI, the autonomous systems executing multi-step tasks, will become the structural unlock that shifts its consumer reach into enterprise revenue. That thesis is plausible, but the shift has not yet happened at scale. Meanwhile, Anthropic's faster revenue growth narrows the window for OpenAI to establish the valuation anchor it wants before its rival files.
The AI profitability race comes down to whether and how quickly the cost of serving intelligence can be brought below the revenue generated by deploying it. As of this week, one of these companies has demonstrated that it can. The other is asking public market investors to believe it will, on a timeline that stretches past the end of the decade. That is the wager now on the table, instead of the Amazon story. It is something considerably harder to price and considerably more consequential for the future of AI investment.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。