Anthropic reaches $47 billion in revenue run-rate and overtakes OpenAI: the AI race has a new leader
🔎 $47 billion in a month: the figure that changes everything
In May 2026, Anthropic announced a revenue run-rate of $47 billion. This means that if the company maintained its current billing pace over twelve months, it would generate $47 billion in revenue. For a company founded in 2021, this is unprecedented.
This figure was made public simultaneously with the announcement of a $65 billion Series H funding round, valuing Anthropic at $965 billion. The company thus overtakes OpenAI in both valuation and revenue, a first since the launch of ChatGPT in late 2022.
But a run-rate is not actual revenue. And the comparison with OpenAI is more complex than it appears. Breakdown.
The key points
- Anthropic announces a $47 billion revenue run-rate in May 2026, compared to $14 billion in February 2026 and approximately $9 billion at the end of 2025.
- A $65 billion Series H values the company at $965 billion, surpassing OpenAI.
- Growth is driven by the massive adoption of Claude Code and enterprise contracts, with the largest accounts doubling in less than two months.
- OpenAI contests the direct comparison: the distinction between gross revenue and net revenue in AI cloud distorts the picture.
- Anthropic has posted annual growth of over 10x for three consecutive years, a pace never before seen in enterprise software.
Recommended tools
| Claude Opus 4.7 | Anthropic's flagship model, agentic and adaptive | Starting at $20/month (June 2025, check on claude.ai) | Complex tasks, code, reasoning |
|---|---|---|---|
| GPT-5.5 | Leader of the agentic leaderboard (98.2) | Starting at $20/month (June 2025, check on openai.com) | Agentic benchmark, advanced automation |
| Gemini 3.1 Pro | Best general LLM (92) | Free with limits / Pro starting at $20/month (June 2025, check on gemini.google.com) | General use, Google integration |
The revenue trajectory: from 3 to 47 billion in one year
Anthropic went from an ARR of $3 billion to $47 billion in twelve months. That is a 15.7x increase, which according to Tech Insider constitutes the fastest growth in enterprise software history.
The curve is visibly accelerating. At the end of 2025, ARR was hovering around $9 billion. In early April 2026, NerdLevelTech reported a crossing of the $30 billion mark. In May, the $47 billion threshold was reached according to Simon Willison.
In fifteen months, Anthropic multiplied its revenue by 30. No SaaS company — not even Salesforce in its best years — has come close to this speed.
But a run-rate measures a snapshot of billing, not recognized revenue. The difference is crucial.
The precise timeline
- End of 2025: ~$9 billion ARR (Analytics Insight)
- February 2026: $14 billion run-rate (Simon Willison)
- April 2026: $30 billion ARR, surpassing OpenAI's $25 billion (WinBuzzer)
- May 2026: $47 billion run-rate (IBTimes UK)
Between February and May 2026, the run-rate tripled. This sudden acceleration deserves critical examination.
Gross vs net revenue: why OpenAI is contesting the comparison
The central controversy of this announcement relates to the nature of the reported revenue. Digg reports that while users celebrate the enterprise demand, other voices describe this dynamic as a bigger bubble.
In the AI cloud, the distinction between gross revenue and net revenue is fundamental. Gross revenue represents the total amount billed to customers before deducting infrastructure costs. Net revenue is what remains after paying compute providers.
Anthropic hosts a significant portion of its traffic on Google's TPUs, as part of a compute partnership that tripled its compute capacity. When Anthropic bills an enterprise client $1 million per month for API access to Claude Opus 4.7, a significant portion of that amount goes directly back to Google to pay the compute bill.
If Anthropic reports gross revenue, the $47 billion figure is impressive but the actual margin is compressed. If it's net revenue, it's a revolution.
OpenAI, which operates its own infrastructure via Microsoft Azure with different compute agreements, would tend to report a net revenue that is closer to its gross. The comparison would then be skewed.
Anthropic has not publicly detailed the proportion of gross vs net in its May 2026 announcement. This is a critical point of attention for evaluating the actual health of the company.
What cloud history teaches us
Google Cloud reached $10 billion in quarterly revenue in 2024, or about $40 billion annually, after more than 15 years of existence. AWS took 10 years to reach a comparable run-rate.
Anthropic claims it will reach $47 billion in 5 years. Even taking into account the exponential growth inherent to AI, the parallel with historical cloud trajectories calls for caution.
Claude Code : the hidden growth engine
According to IBTimes UK, this growth is directly linked to the adoption of Claude Code, Anthropic's coding agent. Claude Code is not just a simple chatbot: it is an agentic tool that executes code, navigates codebases, and resolves tickets autonomously.
The connection to the agentic ranking is clear. Claude Opus 4.7 (Adaptive) ranks fourth with a score of 94.3, behind GPT-5.5 (98.2) and Gemini 3 Pro Deep Think (95.4). But the benchmark score does not capture the product integration.
Anthropic has doubled its largest enterprise accounts in less than two months according to NerdLevelTech. This means that companies testing Claude Code have moved to massive organizational deployment contracts.
The economic model is powerful: a developer using Claude Code generates tens of thousands of API calls per day. Multiplied by companies with 5,000 to 50,000 developers, the revenue per account explodes.
This dynamic explains why Anthropic and OpenAI are each launching their enterprise joint-venture to deploy AI in SMBs and large corporations, with $10 billion committed.
Series H: 65 billion to 965 billion valuation
Announcing the revenue run-rate is not separate from the fundraise. It's a package deal. ABHS rapporte that the Series H raised 65 billion dollars at a 965 billion valuation.
This timing is not innocent. Presenting a 47 billion run-rate just before raising funds maximizes the pressure on investors to join the round. It's a standard practice in VC, but the scale is unprecedented.
Anthropic is now targeting 900 billion dollars and beyond, a round that had been anticipated but whose reality exceeds initial projections.
The question is whether this valuation is justified by the fundamentals or if it relies on a run-rate that could deflate.
AI Valuation Comparison (2026)
| Company | Valuation (2026) | Revenue run-rate | Multiple |
|---|---|---|---|
| Anthropic | 965 billion $ | 47 billion $ | ~20x |
| OpenAI | ~300 billion $ (est.) | 25 billion $ | ~12x |
| Google Cloud (AI segment) | N/A (integrated Alphabet) | ~40 billion $ (est.) | N/A |
| AWS (AI segment) | N/A (integrated Amazon) | ~30 billion $ (est.) | N/A |
A 20x multiple on a run-rate is aggressive but not absurd for a company growing 10x annually. That said, this growth still needs to be maintained.
The Chinese counter-attack: Kimi K2.6 and the open-weight game
While Anthropic and OpenAI wage a gross revenue war in closed-source, a third player is changing the game. Moonshot AI has raised $2 billion and its model Kimi K2.6 dominates the open-weight segment with an agentic score of 88.1 in the self-host version.
Kimi K2.6 also ranks at 84 in general, at the same level as DeepSeek V4 Pro (High). This is a strong signal for companies that want to deploy AI without relying on a single provider.
For technical teams, this option is serious. Running Kimi K2.6 locally or in self-host offers total control over data and compute costs. If you are exploring this path, the guide to the best LLMs to run locally is a concrete starting point.
The pressure of open-weight could force Anthropic to reduce its margins to maintain its growth. Which brings us back to the question of gross vs net revenue.
Is this growth sustainable?
Three factors suggest an inevitable slowdown.
First, the installed base. Anthropic has already convinced early adopters and large tech companies. The next customers will be SMEs and traditional businesses, whose sales cycles are longer and budgets tighter.
Second, compute saturation. The TPU partnership with Google has tripled capacity, but compute is not infinite. If demand continues to increase at the current rate, Anthropic could run into physical infrastructure constraints.
Third, competition from local models. With Kimi K2.6, Z.AI's GLM-5 (82 in agentic, 83 in general), and DeepSeek V4 Pro, companies have credible alternatives that erode Anthropic's pricing power.
10x annual growth over three years is a documented fact. But going from 47 to 470 billion in run-rate in a single year would be a different story. Idle.n points out that this is the first time a direct competitor to ChatGPT has taken the lead in revenue. The risk for Anthropic is becoming the number one target.
What this means for developers and businesses
For developers, the lesson is clear: the Claude ecosystem has become too big to ignore. Claude Opus 4.7 and Claude Sonnet 4.6 (81.4 agentic, 83 general) are now de facto standards in many companies.
But relying on a single vendor is a risk. The healthy practice is multi-vendor: GPT-5.5 for pure agentic tasks, Claude Opus 4.7 for code and reasoning, Gemini 3.1 Pro for integration into the Google ecosystem, and an open-weight model like Kimi K2.6 for sensitive workloads.
For businesses, the decision to invest heavily in Claude Code must be evaluated in light of pricing sustainability. A run-rate that triples in three months may indicate explosive adoption, but also promotional rates that will not survive the next market phase.
❌ Common mistakes
Mistake 1: Confusing run-rate with recognized revenue
A revenue run-rate of 47 billion does not mean Anthropic has collected 47 billion. It is an annualized projection based on the billing of a given month. The actual recognized revenue (GAAP revenue) is potentially very different. Always check whether the announced figure is a run-rate, an ARR, or recognized revenue.
Mistake 2: Directly comparing gross and net revenue between vendors
If Anthropic reports gross revenue (before Google compute cost) and OpenAI reports net revenue (after Microsoft compute cost), the comparison is misleading. Always ask about the nature of the revenue before drawing competitive conclusions.
Mistake 3: Extrapolating exponential growth indefinitely
10x annual growth over three years is extraordinary. Extrapolating it to predict 470 billion in run-rate in 2027 ignores market, compute, and competitive constraints. Exponential curves always eventually flatten out.
❓ Frequently Asked Questions
Is Anthropic really generating more revenue than OpenAI?
In terms of reported run-rate, yes: 47 billion versus approximately 25 billion for OpenAI in February 2026. But the comparison is debated because the revenue calculation methodology (gross vs net) likely differs between the two companies.
What is the Series H?
It is a funding round, the eighth for Anthropic. This Series H raised 65 billion dollars at a valuation of 965 billion dollars, making it one of the largest rounds in tech history.
Why is Claude Code driving growth?
Claude Code generates a volume of API calls per user that is much higher than classic chat usage. A developer can make tens of thousands of requests per day, which turns each user into a high-value account. The doubling of enterprise accounts in two months confirms this effect.
Which Anthropic model to choose in 2026?
Claude Opus 4.7 (Adaptive) for complex tasks and agentic code. Claude Sonnet 4.6 for a good performance/cost balance. Claude Opus 4.6 remains viable for less demanding workloads. The choice depends on your call volume and the complexity of the tasks.
✅ Conclusion
Anthropic has crossed a symbolic threshold by surpassing OpenAI in revenue run-rate, but the 47 billion figure must be read with the lucidity it deserves: it is a projection, potentially in gross revenue, in a market where the distinction between real growth and artificial inflation of metrics is blurry. The current leader is not necessarily the leader of tomorrow.