Source: Replit recently moved most AI spend from Anthropic to OpenAI, which gave 30%+ off in a deal worth $100M+/year, after Anthropic discontinued its discount
OpenAI's efforts to undercut Anthropic on price are producing some results. — For instance, in May, Anthropic models powered 57% …
Context & Ripple Effects
Anthropic had been gaining initial AI-tool purchases in March, when Ramp data put it at about 73% of new-company spending, while OpenAI held a larger reported Q1 revenue base despite a deeply negative adjusted operating margin. The two companies also accounted for roughly 89% of annualized revenue among leading AI startups in May, concentrating the commercial contest between them.
Replit’s reported contract shift turns that broader rivalry into a large-account pricing test: a supplier’s willingness to maintain discounts can determine workload allocation alongside model preference.
First-order effects
- Replit reportedly moves most of its AI-model spending from Anthropic to OpenAI, making OpenAI the primary beneficiary of a contract valued at more than $100 million annually.
- Anthropic loses a major discounted workload after discontinuing its discount, while OpenAI accepts a discount of more than 30% to win the spend.
Second-order effects
- Anthropic faces pressure to weigh retention pricing against revenue per customer as large buyers can use OpenAI’s offer as procurement leverage.
- For OpenAI, the account adds contracted demand but sharpens the margin trade-off already visible in its reported negative adjusted operating margin.
Third-order effects
- Large AI customers are gaining bargaining power to route workloads toward the provider with the lowest acceptable price, rather than treating model choice as fixed.
- If comparable contracts are contested through discounts, the leading-model market may increasingly reward suppliers that can finance lower prices at scale, intensifying the revenue concentration among leading AI startups.
The trend: Enterprise AI procurement is becoming a price-sensitive contest for concentrated, high-value workloads, putting model-provider margins under pressure.