TL;DR
Anthropic hit $30 billion in annualized revenue in April 2026, passing OpenAI’s $25 billion for the first time since ChatGPT launched. Anthropic got there while spending roughly 3x less on model training. Claude Code and enterprise API usage are the two engines behind this growth, and an IPO targeting October 2026 could raise $60 billion or more. One caveat: the two companies use different revenue accounting methods, so the exact gap is debated.
From $1 Billion to $30 Billion in 15 Months
In December 2024, Anthropic was at $1 billion annualized. By July 2025, that had grown to $4 billion. When the year ended, they were running at $9 billion.
By February 2026, the number hit $14 billion. A month later, $19 billion. And as of April 2026, it’s $30 billion — a 1,400% year-over-year increase that makes most SaaS growth curves look flat.
This is the first time since ChatGPT launched in November 2022 that any competitor has matched OpenAI’s revenue, let alone surpassed it.
A caveat worth mentioning: Semafor reported on April 10 that the two companies use different revenue accounting methods. Through cloud partners like Azure, OpenAI counts only its roughly 20-cent cut per dollar as revenue (net), while Anthropic reportedly counts the full dollar (gross). On an apples-to-apples basis, the gap might be narrower than the headline numbers suggest. Still, even accounting skeptics agree Anthropic’s growth rate is outpacing OpenAI’s.
Where the Money Comes From
Anthropic’s revenue mix looks nothing like OpenAI’s consumer-heavy model.
Enterprise API: 70-75% of Revenue
The bulk of Anthropic’s income comes from enterprise and startup API calls on a pay-per-token model. Over 1,000 business customers now spend more than $1 million annually on Claude — that number doubled from 500 in less than two months. Eight of the Fortune 10 are active Claude customers.
The API pricing itself is competitive but not cheap. Claude Opus 4.6 costs $5/$25 per million tokens (input/output), Sonnet 4.6 runs $3/$15, and Haiku 4.5 comes in at $1/$5. But enterprises keep paying, because usage keeps climbing.
Claude Code: The $2.5 Billion Engine
Claude Code has become Anthropic’s fastest-growing product. Launched in May 2025, it hit $2.5 billion in annualized revenue by February 2026, and that figure has more than doubled since the start of the year.
Business subscriptions to Claude Code quadrupled in the same period, and enterprise use now represents over half of all Claude Code revenue. According to the Pragmatic Engineer’s 2026 developer survey (15,000 respondents), Claude Code leads as the “most loved” AI coding tool at 46%, with Cursor at 19% and GitHub Copilot at 9%. (We covered the broader JetBrains AI coding tools survey earlier this month.)
The product has 18% market adoption among developers as of January 2026, a 6x increase from roughly 3% in mid-2025.
Consumer Subscriptions: The Smaller Piece
Unlike OpenAI, where ChatGPT consumer subscriptions have historically driven a large chunk of revenue, Anthropic gets about 80% from enterprises. Consumer subscriptions (Claude Pro, Claude Team) contribute, but they’re secondary to the API and Claude Code businesses.
The Efficiency Gap: How Anthropic Spends Less
Anthropic spends a fraction of what OpenAI burns on training and is still growing faster.
| Metric | Anthropic | OpenAI |
|---|---|---|
| 2026 ARR | $30B | $25B |
| Projected 2026 training spend | ~$12B | $32B |
| Projected 2030 training spend | ~$30B/yr | $125B/yr |
| 2026 projected cash burn | Not disclosed | $25B |
| Revenue per training dollar | ~$2.50 | ~$0.78 |
OpenAI plans to spend $32 billion on model training in 2026 alone, with cash burn projected at $25 billion this year and $57 billion in 2027. Their total infrastructure commitments are estimated at over $1 trillion through 2035, including the $500 billion Stargate project alone.
Anthropic’s training budget is roughly a third of that. The company has been more selective about compute allocation, relying heavily on Google’s TPU infrastrucure rather than building out massive GPU clusters.
This efficiency advantage comes partly from architecture decisions, partly from the Google partnership, and partly from Anthropic’s approach to scaling, which favors targeted improvements over raw compute increases.
The Google-Broadcom Infrastructure Play
On April 7, Anthropic announced an expanded partnership with Google and Broadcom for 3.5 gigawatts of next-generation TPU capacity, expected to come online starting in 2027. This adds to the 1 gigawatt already coming online in 2026 under the Google Cloud agreement from October 2025.
To put that in perspective, 3.5 gigawatts could power a mid-sized city. Most of this infrastructure will be located in the United States, extending the $50 billion American AI infrastructure commitment Anthropic made in November 2025.
Analysts at Mizuho estimate Broadcom will earn $21 billion in AI revenue from the Anthropic relationship in 2026, growing to $42 billion in 2027. The partnership is a bet on custom silicon over commodity GPUs, and the revenue numbers suggest it’s paying off.
Why OpenAI Is Falling Behind on Revenue
$25 billion in annualized revenue is massive by any standard. But relative to Anthropic, OpenAI’s growth has decelerated, and there are a few clear reasons.
ChatGPT subscriptions ($20/month for Plus, $200/month for Pro) drove early growth, but consumer revenue is harder to scale than enterprise API contracts. Anthropic’s enterprise-first approach creates stickier, higher-value relationships.
The burn rate gap compounds this. OpenAI’s $25 billion cash burn in 2026 means they’re spending roughly $1 for every $1 they earn. Anthropic’s undisclosed but reportedly much lower burn rate gives them more runway per dollar of revenue.
Then there’s product sprawl. OpenAI has been shipping rapidly: the ChatGPT Super App (merging ChatGPT, Codex, and the Atlas browser), Sora (now killed), GPT-5.4, the upcoming GPT-5.5 “Spud.” This breadth costs money and dilutes focus. Anthropic’s product line is narrower: Claude models, Claude Code, and the API.
Killing Sora and watching the $1 billion Disney deal collapse didn’t help the revenue story either. Smart operationally, but it doesn’t inspire confidence in ancillary bets.
The IPO Race
Both companies are eyeing public markets, and the timing has become a competitive factor.
Anthropic is targeting an October 2026 IPO, with Goldman Sachs and JPMorgan Chase competing for underwriting roles. The company could raise $60 billion or more at a valuation between $380 billion and $500 billion. With eight of the Fortune 10 as customers and $30 billion in ARR, the S-1 will tell an unusual story for an AI company: one backed by actual revenue.
OpenAI is also laying groundwork for a U.S. listing, with reports suggesting they’ll reserve part of the IPO for individual investors. Their most recent private valuation hit $852 billion after the $122 billion funding round that closed in late March. But the revenue gap and the burn rate could make the public market conversation awkward.
SpaceX’s planned $1.75 trillion IPO might steal some of the oxygen, too. 2026 is shaping up to be the biggest year for tech IPOs since the dot-com era.
What This Means for Developers
If you’re building on top of AI APIs, Anthropic’s revenue lead has practical implications.
A company generating $30 billion with a manageable burn rate is more likely to maintain stable API pricing than one burning cash at a 1:1 ratio. If you’re picking an AI provider for a production system, Anthropic’s financials look healthier for long-term bets.
Claude Code is increasingly the default for teams picking an AI coding tool. With 18% developer adoption and growing, plus strong satisfaction scores in the Pragmatic Engineer survey, the momentum is real.
On the enterprise side, Anthropic’s revenue mix means features like SOC 2 compliance, audit logging, custom model tuning, and SLA guarantees will keep getting investment. Revenue funds research too. Anthropic’s efficiency means more of each dollar goes toward model improvements rather than infrastructure overhead.
The risk? Vendor concentration. If 80% of your AI spend goes to one provider and they raise prices post-IPO, you don’t have great leverage. OpenAI, Google, and open-source models still matter as alternatives, and diversifying your model layer is worth the effort even if Claude is your primary tool today.
FAQ
How much revenue does Anthropic make in 2026?
Anthropic hit $30 billion in annualized run-rate revenue as of April 2026, up from $9 billion at the end of 2025. This represents roughly 1,400% year-over-year growth.
When did Anthropic surpass OpenAI in revenue?
Anthropic’s reported ARR passed OpenAI’s $25 billion sometime in March-April 2026. However, Semafor reported that the two companies use different revenue accounting methods (gross vs. net), so the exact gap depends on how you count cloud partner revenue.
How much does Anthropic spend on training compared to OpenAI?
Anthropic spends roughly 3x less on model training than OpenAI. OpenAI has disclosed plans to spend $32 billion on training in 2026, while Anthropic’s budget is reported at around $12 billion, aided by their Google TPU infrastructure partnership.
When is Anthropic’s IPO?
Anthropic is reportedly targeting October 2026 for its IPO, with Goldman Sachs and JPMorgan competing for underwriting roles. The company could raise $60 billion or more at a valuation between $380 billion and $500 billion.
What percentage of Anthropic’s revenue comes from enterprises?
Approximately 80% of Anthropic’s revenue comes from enterprise customers. Over 1,000 businesses spend more than $1 million annually on Claude services, and eight of the Fortune 10 are active customers.
Bottom Line
For three years, the AI industry treated OpenAI as the default leader. That assumption no longer holds on revenue. Anthropic built a $30 billion business by selling API tokens to enterprises and a coding tool to developers.
The question now is whether OpenAI can close the efficiency gap before both companies go public. At $32 billion in training spend versus Anthropic’s estimated $12 billion, OpenAI pays nearly three dollars in training costs for every dollar that Anthropic spends. That math gets harder to defend in an S-1.
We’ll know a lot more by October.