OpenAI uses aggressive discounts to capture Anthropic market share

The shift in customer loyalty

Major companies are beginning to shift their AI workloads from Anthropic to OpenAI primarily to reduce operational costs. For example, the HR software company Rippling saw its usage of OpenAI models jump to 52 percent, while its use of Anthropic dropped significantly. The company's CEO stated that OpenAI proved to be the cheaper option for their engineers.

Other significant shifts involve large-scale contracts where price was the deciding factor. Replit, a popular AI coding app, moved the majority of its AI spending to OpenAI after the company offered a price reduction of over 30 percent on a deal valued at more than $100 million annually. This move was prompted by Anthropic's refusal to extend previous discounts without requiring Replit to commit to higher spending.

The transition for these companies is becoming increasingly rapid due to the low technical barriers involved. Replit was able to complete the switch in just a few days for its products and a tiny amount of code modification. This speed is faster than the time it took for the company to negotiate its original contract with Anthropic.

Wholesale buyers as price sensors

Wholesale operators serve as a critical indicator of the current pricing war because they resell AI models to millions of developers. Companies like Replit and Cursor operate on the profit gap between their user fees and the costs they pay to the labs for tokens. Because their profit depends on this spread, they are the most price-sensitive customers in the industry.

These wholesale buyers can move their entire workloads to a different provider in a matter of hours. This agility allows them to react instantly to new discounts or price drops from the major labs. The report notes that Cursor, which was largely powered by Anthropic's Claude, has been moving business toward OpenAI and SpaceX's Grok models.

The move toward multi-sourcing is also becoming a common strategy for large-scale users. A 'hard cap' on discounts, which Anthropic uses once a customer hits a contract limit, often invites users to move the next workload to whoever is cheaper per task. This creates a constant pressure on the labs to offer more flexible and competitive pricing structures to retain their biggest accounts.

Margin vs. market share

The underlying economics of this competition differ significantly between the two rivals. Anthropic currently reports a gross margin of around 75 percent on its existing models, and its revenue is beginning to outrun its compute spending. Because Anthropic is preparing for a mega-AI IPO before Thanksgiving, it faces pressure to maintain these healthy margins while still showing growth.

In contrast, OpenAI is currently operating with projected compute costs that are well above its revenue for this year. This means OpenAI is willing to pay up and sacrifice its own margins to buy market share from Anthropic. By offering deep discounts, OpenAI is positioning itself to capture the volume necessary to justify its own upcoming IPO next year.

This dynamic creates a situation where Anthropic's high margins represent a direct opportunity for OpenAI. While Anthropic's revenue growth has slowed from doubling to 45 percent between the second and third quarters, OpenAI is using its financial runway to undercut them. This creates a strategic tension between protecting profit per user and securing the total number of users.

The broader competitive landscape

The pressure on frontier AI pricing is not just coming from OpenAI; it is coming from multiple directions. Microsoft has positioned itself as a customer advocate, routing its Copilot work to the cheapest model capable of doing the job. Meanwhile, Meta has cut its spending on Anthropic's Claude while building its own Muse models to compete directly in the market.

Even the largest partners are beginning to build their own alternatives to reduce their dependency on Anthropic. Companies like Cognition, Meta, and Microsoft are all developing internal tools to cut their Claude bills. This trend suggests that the high costs of frontier models are driving a move toward internal development and diverse sourcing.

Furthermore, open models are setting a price floor for the entire industry. This summer, DeepSeek offered coding output for about 28 cents, which was a fraction of the $25 cost for Anthropic's top model. This 'good enough' alternative at a much lower price point provides a constant downward pressure on the pricing of all frontier AI services.

What comes next for pricing

The current price war is likely to follow a playbook similar to the early days of cloud computing. As the market matures, we can expect to see a move toward flexible discounts that scale with usage volume. This allows labs to maintain a baseline of revenue while still offering the deep discounts required to win over wholesale and enterprise customers.

The outcome of this battle will depend on how the market reacts to the trade-offs between growth and margins. Anthropic must balance its need for high margins in its IPO prospectus against the risk of losing share to OpenAI's aggressive pricing. OpenAI must decide how much of its negative margin it is willing to sustain to establish itself as the dominant market leader.

Ultimately, the customers are the primary winners in this competition. As the labs fight for dominance, tokens will continue to get cheaper while usage expands. This cycle will likely continue until the scarcity of compute eases and the infrastructure becomes more standardized, potentially stabilizing the pricing landscape for everyone involved.