OpenAI Revenue Revised Down $20B to ~$50B Annualized

OpenAI told investors its annualized revenue is "approaching $50 billion," about $20 billion below the figure that had been circulating. Part of the gap is accounting: Anthropic counts revenue that flows through AWS and Google Cloud; OpenAI books direct only. The real 2025 revenue figure is roughly $13 billion, so the $50 billion annualized number is a recent monthly run-rate extrapolated, not a year anyone has banked. The IPO slipped from 2025 to early 2027.

The math is tight. OpenAI raised $122 billion in March and spends significantly more than it earns, almost entirely on compute: training new models and running inference on every query customers send. The bull case is that inference costs fall fast enough, via NVIDIA's B200 chips and OpenAI's own custom silicon, to flip the margin before the IPO window. If that hardware roadmap slips, the 2027 window moves again or prices at a fraction of the private valuation. Cost curves don't arrive on schedule just because a company needs them to.

The accounting split has a practical implication for enterprise buyers. Anthropic's revenue running through AWS Bedrock and Google Cloud means a lot of real AI spend already sits inside a hyperscaler contract, with its data terms, audit logs, and indemnification. A CTO signing a multi-year deal gets procurement cover through that intermediation that OpenAI's direct contract may not match. That's worth something independent of the valuation question.

For developers and operators building on these APIs, the more immediate question is what pre-IPO pressure does to pricing. A company this far underwater, 15 months from a public offering, has one direction to push margin. Expect cheap tiers to get squeezed, rate limits to tighten, and top-model access to migrate behind higher-priced tiers. The practical hedge is boring and correct: keep prompts portable, run the same eval suite against Claude and Gemini this quarter, and know your switching cost before the pricing email arrives.

The call: OpenAI will raise prices or cut capacity on at least one developer API tier before its early-2027 IPO, with the change landing by 2027-03-31. Confidence is medium. The burn and IPO clock point one direction, but a fast drop in inference costs could defer it. Anthropic and Google competing on price is the one force that could hold OpenAI's hand. We're wrong if all current paid API tiers keep their October 2026 prices, model access, and rate limits through that date.

OpenAI told investors its annualized revenue is "approaching $50 billion," about $20 billion under the $70 billion figure that had been floating around. The gap is partly accounting: Anthropic counts money that flows through AWS and Google Cloud as its revenue, OpenAI doesn't. The IPO slipped from 2025 to early 2027. For anyone who buys or builds on these models, the question is simple. Does a company burning this hard, chasing a public offering, keep API prices where they are?

This is a read, not a decision you have to make today. Nothing here is hard to undo. You can switch model providers in an afternoon if the economics turn. What's actually at stake is whether the pre-IPO squeeze on OpenAI lands on your bill in the next 18 months. Nothing sets a hard deadline, but the early-2027 IPO window is the clock that matters.

The Skeptic

A $20 billion gap that lived as consensus until someone checked is not a footnote. It tells you how little real diligence sits under these numbers. Note what $50 billion annualized means against roughly $13 billion in actual 2025 revenue. That is a run-rate snapshot of the best recent month, multiplied up, not a year anyone has banked. The bull case needs revenue to nearly quadruple and hold. That has to be proven. OpenAI raised $122 billion in March. It spends far more than it earns. The 2027 IPO exists to refinance that burn in public markets. Tight math, and the "AI is inevitable" story is carrying the weight that unit economics should.

The Compute Pragmatist

The interesting number is the spread between $13 billion in 2025 revenue and "significantly higher spending." That gap is almost entirely compute, both training the next models and running inference, which is the cost of answering every query a customer sends. At $50 billion annualized, OpenAI is still underwater on infrastructure alone. The $122 billion raise is a bet that inference costs fall fast enough, via NVIDIA's B200, faster chip-to-chip links, and OpenAI's own custom silicon, to flip the margin before the IPO. If NVIDIA's roadmap slips or the custom chips underperform, the 2027 window pushes again or prices at a fraction of the private mark. Cost curves don't bail out software on a reliable schedule.

The Enterprise Buyer

The accounting split is a buying signal. Anthropic books revenue through AWS and Google Cloud; OpenAI books direct. That means a lot of real AI spend runs through the hyperscaler you already have a contract with, with its data terms, its audit logs, its indemnification. If you buy Claude through AWS Bedrock, you get procurement cover OpenAI's direct contract may not match. For a CTO signing a multi-year deal, that intermediation is a feature, not an accounting quirk. The flip side: a vendor burning toward an IPO is a vendor whose pricing and contract terms will move. Lock in terms now, ask for price protection, and don't assume today's tier survives the rationalization.

The Builder

Forget the valuation drama. What changes on Tuesday? Probably nothing yet. But a company this far underwater, 15 months from a public offering, has one direction to push margin, and it runs straight through developer pricing. Expect the cheap tiers that power your prototypes to get squeezed, rate limits to tighten, and the best capability to sit behind the priciest tier. The practical move is boring and correct: keep your prompts portable, keep an eval harness that runs the same tests against Claude and Gemini, and know your switching cost before the pricing email lands, not after.

Where they split

The Skeptic and the Compute Pragmatist agree the math is tight but disagree on what saves it. The Skeptic says revenue has to quadruple and that's unproven. The Pragmatist says the rescue is the cost side collapsing, not revenue climbing, and that depends on NVIDIA and custom chips hitting their dates. Those are different bets with different failure points.

The second split is the Enterprise Buyer versus the Builder on what the intermediation means. The Buyer sees the hyperscaler relationship as safety and leverage. The Builder sees any pre-IPO vendor, direct or intermediated, as a pricing risk to hedge against. Both are right, which is why portability is the answer that satisfies both.

What it hinges on

One belief does most of the work: whether OpenAI can hold or raise prices into an IPO without bleeding developers to Anthropic and Google. If inference costs fall fast, they don't need to squeeze you. If costs stay stubborn, the squeeze lands on API tiers before early 2027. The council leans toward some tightening, because the burn is real and the IPO clock is real, and margin pressure on a software business with this cost structure goes to the product people pay for.

Before you commit more of your stack to one provider, run the same eval suite against two alternatives this quarter and write down the switching cost. That's the whole hedge.

Prediction: OpenAI will raise prices or cut capacity on at least one of its developer API tiers before its early-2027 IPO, with the change landing by 2027-03-31.

Confidence: Medium. The burn and IPO clock point one way, but a fast inference-cost drop could defer it.

Why: OpenAI spends far more than the roughly $13 billion it earned in 2025, and the 2027 IPO forces it to show a credible path to margin on a public timetable. For a business whose biggest cost is running the models, the fastest lever on margin is the price buyers pay, which means API tiers get repriced or rate-limited. The opposite outcome, prices held flat, only happens if inference costs fall fast enough via new NVIDIA hardware and custom chips to fix margin without touching the price list, and hardware roadmaps rarely arrive early. Anthropic and Google competing on price is the one force that could hold OpenAI's hand, which is why this is Medium, not High.

Revisit by 2027-03-31: We're right if OpenAI raises per-token prices on any paid API tier, introduces a higher-priced tier that gates current top-model access, or cuts rate limits on an existing paid tier by then. We're wrong if all current paid API tiers keep their October 2026 prices, model access, and rate limits through 2027-03-31.