Analysis
OpenAI is pulling customers away from Anthropic by undercutting it on price, not just features, according to a report from The Information relayed by Daily AI Brief. Four accounts named this week show the gap is already showing up in production workloads, not just vendor pitches:
- Rippling -- CEO Parker Conrad said OpenAI's share of the HR-software company's coding workload jumped from 16% to 52% over recent months, while Anthropic's share fell from 57% to 18%.
- Replit -- the coding-agent startup shifted spending after OpenAI cut more than 30% off a contract worth over $100 million a year.
- DataCamp -- CEO Jonathan Cornelissen said OpenAI's pricing cut the company's model costs by 80% to 90%.
- An unnamed US bank with more than $7 billion in assets doubled its ChatGPT seat count to 500 at no extra charge, paying under $7 per seat against a usual price of at least $20.
Four accounts, one pattern
OpenAI's discounting leans on its Sol and Luna model lines, which the report describes as cheaper to run than earlier GPT generations -- the same cost edge that let the company cut prices without matching cuts to its own infrastructure spend. The timing is pointed: OpenAI has been racing to close a $30 billion funding round around the same period Pulse previously covered OpenAI's and Anthropic's valuation paths pulling apart, and Anthropic is heading toward a public listing where revenue growth and gross margin will get line-by-line scrutiny.
An investor memo cited in the report says Anthropic's growth rate slowed between the second and third quarter -- the first external signal that the discounting shows up in Anthropic's own numbers, not just OpenAI's win column. Neither company discloses customer-level pricing, so these four accounts are a sample, not a census; OpenAI has said publicly it is approaching $50 billion in annualized revenue, a disclosure Anthropic has not matched with comparable detail. The Information has also argued, in a separate analysis, that the margin math right now tilts toward OpenAI -- a dynamic this discounting data appears to confirm.
What the discounting really costs
For founders choosing a model vendor, the lesson is blunt: large API contracts are now negotiable in a way they weren't a year ago, and incumbents with switching leverage -- an existing relationship, a security review already passed -- can extract OpenAI-level discounts just by threatening to move. For GPs underwriting AI-wrapper startups, gross-margin assumptions tied to a single model vendor's list price are no longer reliable; Replit's $100 million contract moved on a single discount offer.
The catch for OpenAI is that winning share this aggressively is expensive, and it is doing so while still justifying a valuation north of a trillion dollars to the same investors funding its $30 billion round. A price war that compresses OpenAI's own gross margin -- even as it reports higher revenue -- is a worse outcome for its cap table than slower, higher-margin growth. Critics of the strategy would note that four customer anecdotes, however specific, are not proof of a durable shift; Anthropic could just as easily win back share with its own next round of price cuts or a model upgrade.
Watch three things next: whether Anthropic responds with public pricing moves ahead of any IPO roadshow, whether OpenAI's next revenue disclosure shows margin compression alongside the top-line growth, and whether more named enterprise customers go on record the way Rippling and DataCamp did this week.