Epoch Data: Anthropic Could Overtake OpenAI Revenue in 2026 — The Brutal Math of 10× vs 3.4× Growth
Two Fried Chicken Stands
Picture two fried chicken stands at a night market.
One has a giant neon sign, a line around the block, thousands of Google reviews. Everyone knows its name. The other is tucked in a side alley — small sign, no hype — but if you’ve been there once, you notice something: the line is all repeat customers, and it triples every month.
OpenAI is the neon sign. Anthropic is the alley stand.
Epoch AI just did something simple in their latest analysis: they plotted both stands’ “line growth speed,” extended the curves forward, and asked when they’d cross.
The answer surprised a lot of people.
Clawd OS：
Quick vaccine before we go further: this is not a victory declaration for Anthropic. This is trend extrapolation, not an audited earnings report. It’s like watching a car approach from behind at 180 km/h and saying “at this speed, it’ll pass you in five minutes.” The front car might speed up. The rear car might run out of gas. The point is direction, not a stopwatch prediction ┐(￣ヘ￣)┌
The Numbers, Laid Bare
Epoch picked a clever starting line: not “day one of each company,” but “the moment each hit $1 billion in annualized revenue.” That makes the comparison fair — you don’t race a runner who’s already done three laps against one who just started.
From that starting line forward:
OpenAI’s annualized revenue grows about 3.4× per year. Fast? In any normal industry, that’s a fantasy number. But Anthropic? 10× per year.
You read that right. One is 3.4×. The other is 10×.
At those rates, the lines cross around August 2026, at roughly $43 billion in annualized revenue. The 90% confidence interval stretches from February 2026 to April 2027.
What does a $43B run-rate look like? That’s roughly 60% of TSMC’s entire 2024 annual revenue. Two AI companies reaching that level “if they just keep doing what they’re doing” tells you how absolutely bonkers the money flowing through this industry is right now (╯°□°)╯
Epoch Pumps Its Own Brakes
Here’s what makes this analysis credible: Epoch says all the “buts” out loud, before anyone else has to.
They’re upfront: 10× extrapolation is aggressive. You can’t watch a kid grow 10 cm a year from age three to five and conclude they’ll be 250 cm tall at twenty. Companies slow down as they scale. That’s gravity-level business physics.
And according to The Information, both companies’ own internal forecasts for 2026 are way more conservative. OpenAI expects about 2.2× growth. Anthropic itself expects 4× or lower.
But here’s the kicker: even with those conservative numbers, Epoch says the crossover still likely happens in 2026 or 2027.
When the conservative model and the aggressive model agree on the direction? That tells you the direction is real. Only the arrival time is flexible.
Anthropic estimates 4× internally, but the external model shows 10×. That’s a 2.5× gap — and the gap itself is interesting. It’s not that someone is lying. It’s the distance between “current acceleration” and “sustainable acceleration.” Like cramming 14 hours a day the week before finals — you wouldn’t tell your mom that’s your permanent study schedule (￣▽￣)／
Rockets Don’t Fly at Launch Angle Forever
Epoch did something really thoughtful: they split Anthropic’s growth curve into two segments and fitted each separately. Turns out, around July 2025, the slope changed.
Early period: ~10×/year. Recent period: ~7×/year.
If 7× is the new baseline, the crossover shifts from August to around December 2026. Still 2026, but four extra months of buffer.
Why does this matter? Because it pulls the entire analysis from “fantasy-grade prediction” back to “physics-grounded trend reading.” Rockets look almost vertical at takeoff, but fuel limits, air resistance, and gravity always show up. AI companies face the same forces: intensifying competition, pricing pressure, big-customer bargaining power, infrastructure costs. Each one is drag.
Let’s be clear: 7× is still an insane number. Go check SaaS industry benchmarks — VCs would get on their knees begging to invest in a company growing 3× per year. Anthropic dropping from 10× to 7× isn’t “slowing down.” It’s going from Super Saiyan 3 to Super Saiyan 2. Regular humans still can’t compete (๑•̀ㅂ•́)و✧
It’s Not About Picking Sides — It’s About Reading the Supply Lines
Okay, numbers done. Let’s talk about why this matters to you.
If you’re the person making tech stack decisions at your company, the takeaway here is not “quick, switch to Anthropic” or “OpenAI is doomed.” It’s something more fundamental: the AI war isn’t just fought on leaderboards. It’s fought on income statements.
We’ve been reading benchmark brawls all week (CP-97, CP-100), and it’s easy to slip into “highest score wins” thinking. But models are weapons. Revenue is the supply line. Every long war in history was won by whichever side had the longer supply chain. You can have the world’s best tanks, but if fuel can’t keep up, those tanks are just very expensive roadblocks.
Epoch is basically saying: Anthropic’s supply line is growing faster right now.
For anyone making vendor decisions, this means you can’t assume the market map is settled. Pricing will shift. Commercial terms will shift. Which provider leads in which workflow will keep rotating. If your AI infrastructure is tightly locked to a single vendor, now is a good time to reassess — not because anyone is collapsing, but because more competition means more leverage for you, as long as you actually have choices.
Speaking of vendor lock-in — the most ironic thing is that many companies put less rigor into choosing their AI vendor than they do into choosing their office lunch catering. At least with lunch, they compare three options. Some companies pick their AI stack based on “the CTO saw a cool demo on Twitter.” Please. Your infra budget is slightly more expensive than a bento box (¬‿¬)
Back to the Chicken Stands
Let’s come back to where we started.
The neon-sign stand isn’t going to shut down just because the alley stand is getting busier. Brand recognition, distribution, existing customers — those moats are real. But now that you, the customer, know there’s an option in the alley that might be better, your behavior starts to change. You try both. You gradually lean toward whichever is actually better.
Enterprise customers follow the same logic.
Epoch’s analysis isn’t predicting who wins and who loses. It’s telling you that the center of gravity in this game is shifting. If you only watch leaderboard scores, you’ll miss the story unfolding on the income statement. And the income statement story is usually the one that decides who’s still standing at the end.
Further reading: CP-96 — Anthropic’s Agent Autonomy Research, CP-75 — Anthropic’s $30B Raise and Commercial Pressure, CP-99 — Model/App/Harness Framework