By Hamza Faisal
OpenAI is winning the fight for enterprise coding customers the old-fashioned way: by being cheaper. A new report from The Information, titled "Anthropic's Margin is OpenAI's Opportunity," shows how aggressive discounting is shifting real workloads from Anthropic's Claude to OpenAI's ChatGPT.
The clearest example is Rippling, the HR software company. In May, Anthropic's models handled 57% of its engineers' coding work, while OpenAI's handled just 16%. Those shares have now flipped: OpenAI handles 52% of the work and Anthropic just 18%. Rippling chief executive Parker Conrad said the company switched because OpenAI's pricing was more competitive.
Replit tells a similar story at bigger scale. The coding platform moved most of its AI spending from Anthropic to OpenAI in recent months, under an annual contract worth more than $100 million. OpenAI offered a discount of more than 30%, according to The Information — after Anthropic declined to extend Replit's previous discount unless it committed to higher spending.
The deal terms behind the defection
Replit chief executive Amjad Masad first signalled the shift on stage at TechCrunch Disrupt in mid-October, saying the company was no longer working with Anthropic and that the decision came down to "cost, partnership, and performance." He added, bluntly: "I think ChatGPT is better now."
The split has proven messier than the headline. A person familiar with the deal told TechCrunch the Anthropic partnership remains in place, and Replit later clarified that it remains model agnostic: OpenAI is now the default for Replit Agent, but customers can still choose Claude. That distinction matters less than it sounds — most users stick with the default, so the default captures the bulk of the spending.
DataCamp shows the same playbook at a smaller scale. Its chief executive Jonathan Cornelissen said OpenAI offered a 40% discount on models used for internal product development, provided the company also used Codex — and estimated costs could fall by 80% to 90% if the offer extended to products it sells.
Why Anthropic cannot simply match the price
Anthropic's problem is that discounting is expensive when margins are already under pressure. The company is preparing for a large IPO as soon as next month, and its growth still held at 45% despite slowing from the second to the third quarter. Its computing costs in the second quarter were $5.6 billion — less than half of its $11.5 billion in revenue — while OpenAI has forecast that its computing spending this year will significantly exceed its revenue.
Anthropic also has a policy of ending discounts when customers reach their contract limits, which contributed to losing the Replit business. Meanwhile, some of its biggest customers — including Cognition, Meta and Microsoft — are developing their own technology or alternatives to reduce what they spend on Claude.
What the price war means for everyone else
Notably, no benchmark appears anywhere in the reporting on these switches. Replit builds a coding product, where model differences are supposed to matter most, yet the reported trigger was a withdrawn discount, not a failed evaluation. At the application layer, wrapped in scaffolding and retries, the two systems are apparently close enough that a 30% price gap outweighs any remaining capability gap.
For smaller companies, that is good news in the short term: a price war between the biggest model makers usually reaches everyone eventually. The open question is whether the discounts last — both labs are cutting prices ahead of public listings, and subsidised pricing rarely survives contact with public-market margins. For now, the AI coding war is being fought on the invoice, and OpenAI is winning it.
