AI coding agents are turning into a real business for OpenAI and Anthropic

OpenAI and Anthropic appear to have crossed a critical threshold: their AI coding agents are generating substantial revenue, and both companies have quietly restructured their pricing to capture it. Simon Willison writes for his weblog that the two leading AI labs have found product-market fit, driven by a shift toward enterprise customers paying full API prices for agent tools.

From flat fees to usage-based billing

Until recently, enterprise customers at both companies benefited from generous flat-rate deals. Anthropic’s enterprise plan once promised “enough usage for a typical workday.” That changed in November 2025, when Anthropic switched to a model of $20 per seat per month, plus full API pricing on top. OpenAI made a comparable change in April 2026, aligning Codex pricing with API token costs for all enterprise plans, including education, healthcare, and government customers.

The timing matters. Both companies also released new, more expensive frontier models in April 2026. OpenAI’s GPT-5.5 costs twice as much per token as its predecessor. Anthropic’s Opus 4.7 costs roughly 1.4 times more than Opus 4.6 once its new tokenizer is factored in. Enterprise customers who signed year-long contracts are now renewing under the new pricing structure.

Why coding agents changed the economics

Willison illustrates the scale of the shift with his own usage data. He pays $200 per month across both platforms on consumer plans. Running the same workload at API prices would have cost him $2,180 in the past 30 days alone. For companies deploying these tools across large engineering teams, the costs multiply accordingly.

Coding agents consume far more tokens than a standard chat interaction. They work autonomously on complex tasks, running for extended periods. That makes them expensive to run but also highly valuable to well-paid professionals, particularly software engineers. Anthropic’s Claude Code and its consumer-facing sibling Claude Cowork, as well as OpenAI’s Codex, are the products driving this demand.

Large companies are already feeling the pressure. Uber reportedly exhausted its full-year AI budget within a few months of 2026, mostly due to Claude Code usage. Microsoft cancelled a number of Claude Code licenses ahead of its June 30 financial year-end, partly for cost reasons. Willison interprets both cases not as signs of failure but as evidence that the pricing is landing exactly where it should: just expensive enough to cause hesitation, but still worth paying.

The infrastructure costs at the labs are enormous. A recent SpaceX S-1 filing revealed that Anthropic agreed to pay $1.25 billion per month for cloud computing capacity, through May 2029. The deal appears to cover inference rather than model training, suggesting that serving users at scale now requires spending at that level.

Anthropic is rumored to reach $10.9 billion in revenue in the second quarter of 2026, potentially marking its first profitable quarter. Both companies are also hiring aggressively for enterprise sales roles. About a third of OpenAI’s 703 open positions and roughly 27 percent of Anthropic’s 390 openings relate to enterprise sales and support.

Both companies are preparing for IPOs. Audited figures in their upcoming S-1 filings will provide the first clear picture of whether this revenue surge is as substantial as current signals suggest.