OpenAI revenue hits $40B annualized; completes $7B share buyback

Full analysis

OpenAI says it's running at $40 billion annualized revenue, roughly double where it sat at the end of 2025. It closed a $7 billion employee share buyback, and SoftBank borrowed $10 billion against its OpenAI stake to plow another $10 billion back in. For anyone building on the OpenAI API, the question isn't whether the company is winning. It's what this revenue base and capital structure do to the pricing, reliability, and terms you depend on.

This is a Type 2 read for most builders. Nothing here forces a same-day move. But it should reset your cost model and your assumptions about API pricing over the next two quarters. The forcing function is quiet: revenue at this scale is exactly what a lab uses to justify aggressive price cuts against Anthropic and Google.

The Skeptic. Annualized revenue is the friendliest number in the drawer. Take a good month, multiply by twelve, call it $40B. Nobody has shown a margin, an operating loss, or a clean audited baseline for the "doubled" claim. And the SoftBank move is the tell: borrowing $10B against your own stake to reinvest $10B is one holder leveraging up to keep the round inflated, full stop. For a PM: imagine a startup taking out a loan against its own shares to buy more of its own shares. That's roughly the shape here. Revenue headlines don't answer the unit-economics question, and this one is loud enough to drown it out.

The Compute Pragmatist. Forty billion in revenue sounds like an endless GPU budget until you remember what inference at that scale costs. Every ChatGPT query and every API call burns compute now, continuously, and that bill grows with the top line. So the real fork is whether that revenue funds frontier training runs or just feeds the inference meter. The SoftBank $10B on top of the prior $40B round says OpenAI is still spending well past what it earns, which is consistent with doing both at once. For a PM: revenue is the money coming in the door; inference is the electric bill that never stops. Watch Azure capacity reservations, not revenue slides, to see where the compute actually goes.

The Safety Lens. Scaling to $40B changes the incentive math in both directions. OpenAI now has real money and reputation exposed to a bad deployment, which pulls some behavior toward caution. But the same pressure pushes toward shipping faster and approving enterprise customization that safety teams would rather slow down. The governance wrinkle is the debt: a creditor holding $10B against the stake wants monetization, and that interest is not aligned with anyone's harm-reduction agenda. For a PM: the people financing this want returns on a clock, and clocks push products out the door. The buyback also reshapes who stays. Retention now rewards the employees comfortable with commercial velocity.

The Builder. Set the capital drama aside. What changes on Tuesday? Your cost model, if you've anchored it to today's per-token pricing. At this revenue base, OpenAI has the room to cut prices hard to defend share, so model a meaningful drop over the next two quarters rather than a flat line. The buyback is quietly good news for you: fewer disgruntled engineers walking out the door means less internal chaos, which means your uptime SLAs and model behavior are less likely to lurch. The flip side of the SoftBank leverage is new monetization pressure, so watch for fresh rate tiers or feature paywalls aimed at mid-market teams. Don't sign a long inference commitment at today's rates.

Where they disagree. The Builder reads price cuts coming; the Safety Lens and the Skeptic read the same debt structure as pressure to extract more, sooner, through new tiers and paywalls. Both can be true: headline API prices fall while premium features and enterprise terms get pricier. The Compute Pragmatist and the Skeptic split on the numbers themselves. One treats $40B as real fuel; the other says we don't know the margin, so we don't know how much of that fuel is already spoken for by the inference meter.

What this actually hinges on: does revenue scale translate into lower prices for the workloads you run, or into a squeeze dressed as growth. You can't see OpenAI's margins, so stop trying to. Test the thing you can observe. Lock nothing long. Keep an Anthropic or Google fallback wired into your stack so a mid-market rate tier doesn't strand you. And put a price-review trigger on your roadmap for the next two OpenAI pricing updates, because the incentive to cut base rates and raise premium ones is now baked into the capital structure.

Prediction: By OpenAI's next major API pricing update on or before 2026-11-15, OpenAI will cut headline per-token prices on at least one flagship model tier by 20% or more, while introducing or raising a premium/enterprise feature charge.

Confidence: Medium. Revenue scale and share-defense incentive both point to base-rate cuts plus premium upsell.

Why: OpenAI has spent 2024 and 2025 cutting per-token prices repeatedly as it scales, and $40B in revenue gives it the room to keep doing so to hold share against Anthropic and Google, whose competing models keep landing near or below OpenAI's rates. The SoftBank debt loop adds pressure to monetize, and the cleanest way to do both at once is to drop headline prices while charging more for premium context, priority throughput, or enterprise features. The opposite (flat or rising base prices) is less likely because it would cede price-sensitive API volume to competitors right when OpenAI most needs the revenue story to keep compounding.

Revisit by 2026-11-15: We're right if OpenAI cuts a flagship tier's per-token price 20%+ and adds or raises a premium/enterprise charge. We're wrong if base per-token prices hold flat or rise across flagship tiers through that date.