Anthropic Started Late but Took the Lead in Enterprise AI Revenue
Claude Code changed the game. But understanding the $2 trillion IPO story means looking past explosive revenue growth to $518 billion in compute commitments and customer concentration.
Late to start, but it picked a different market from the beginning
OpenAI was founded in 2015 and opened the mass-market generative AI space with ChatGPT. Anthropic came six years later, in 2021, started by former OpenAI researchers including Dario Amodei and Daniela Amodei. On timing alone, OpenAI's first-mover advantage looked overwhelming.
But Anthropic avoided a head-on fight. Instead of chasing consumer app usage and download numbers, it focused on the work developers and enterprises actually pay for. According to internal projections obtained by Reuters in 2025, roughly 80% of Anthropic's revenue was already coming from enterprise customers. That choice shaped everything that followed.
The decisive product was Claude Code. It started as an internal tool built by Anthropic's own engineers, then grew into an agentic coding product capable of reading, editing and testing code over long, sustained tasks. Coding is a good market to prove AI's economics in: productivity gains are easy to measure, and purchase decisions are easy to justify against developer salaries.
Claude Code's success was a win for focus, not just one product
Anthropic's edge was never one or two model benchmarks. It was clarity about who the product was for. By narrowing its focus to developers, enterprise IT teams, and professionals in finance and law, the company aligned its model research, product design, security and deployment around a single direction.
Cowork extended that strategy beyond coding. As expectations grew that AI agents could take over work long handled by existing SaaS tools — document drafting, data analysis, legal work, finance, sales — shares of some software and information-services companies wobbled. Some argued the market reaction was overdone, but one thing was clear: enterprises had started treating AI not as a chatbot, but as an execution layer for actual work.
OpenAI changed course too. In March 2026, OpenAI executives reportedly told staff internally that the company would concentrate more resources on coding and enterprise products, cutting back on lower-priority "side quests." The fact that the market leader began following the challenger's playbook was itself evidence that Anthropic's positioning had worked.
The numbers get more extreme from here
| Metric | Value | Note |
|---|---|---|
| Anthropic 2025 revenue | $4.6B | About 12x year-over-year growth |
| Anthropic annualized revenue, end of July 2026 | $65B+ | More than 7x the pace at end of 2025 |
| OpenAI annualized revenue, September 2026 | ~$70B | Enterprise revenue more than doubled since July |
The key distinction here is between fiscal-year revenue and annualized run rate. Anthropic's $4.6 billion is actual full-year revenue for 2025. The $65 billion-plus figure, and OpenAI's roughly $70 billion, are run rates — a recent month or quarter's pace projected out over a year. They are not the same kind of number.
Even so, the direction is unmistakable. Anthropic's annualized revenue jumped from roughly $9 billion at the end of 2025 to more than $65 billion by July 2026. Bloomberg reported in mid-September that Anthropic's annualized revenue could top $100 billion before the year is out. OpenAI, meanwhile, was also closing in on roughly $70 billion by late September, with enterprise revenue more than doubling since July. The gap between the two companies isn't fixed. It's narrowing again.
But the more important number for the IPO isn't $4.6 billion. It's $518 billion.
Anthropic's net loss for 2025 was roughly $42 billion. On its face, that looks like a loss more than nine times revenue. But about $34 billion of that was a non-cash accounting charge tied to revaluing financing-related liabilities. Looking at the actual economics of the business requires separating out an operating loss north of $8 billion and $7.33 billion in compute and infrastructure spending.
The bigger issue is future commitments. According to the IPO prospectus reviewed by Reuters, Anthropic has locked in at least $518 billion in cloud, compute, equipment and infrastructure obligations over roughly the next decade. About 80% of that is either difficult to cancel or must be paid regardless of actual usage.
| Key line item | Size | What it means for investors |
|---|---|---|
| 2025 revenue | ~$4.6 billion | Starting point for explosive growth |
| 2025 operating loss | $8 billion+ | Compute costs still outpace monetization |
| 2025 compute/infrastructure spending | $7.33 billion | A cost structure bigger than revenue |
| Future infrastructure commitments | $518 billion minimum | Costs could remain even if demand cools |
| Top two customers' share of revenue | ~25% | Customer concentration risk |
In other words, Anthropic shouldn't simply be read as an "AI software company." It looks more like a capital-intensive platform business that locked in hundreds of billions of dollars in compute supply ahead of demand. If revenue keeps exploding, those commitments become a moat. If growth slows, the fixed, contract-like costs become the company's biggest weakness. </markdown>
Insight Times Editorial Desk
