OpenAI and Anthropic subscription economics spark secondary market

OpenAI’s ChatGPT Pro 20× and Anthropic’s Claude Max 20× both cost $200 a month, yet SemiAnalysis shows they would cost $14,000 and $8,000 respectively if fully utilized at API rates, meaning the plans become unprofitable after just 5.7%–10% utilization [2]. This margin squeeze is driving a nascent secondary market where firms resell discounted access and data‑locking services.

Pricing pressure and lock‑in strategy

SemiAnalysis’ testing revealed that the flat‑rate plans dramatically under‑cover the compute cost once usage exceeds modest thresholds, with OpenAI turning negative beyond 5.7% utilization and Anthropic after roughly 10% [2]. The companies therefore seek ways to lock customers in beyond raw model quality. Pydantic CEO Samuel Colvin explains that both labs are adding “code‑generation discounts” and planning to store full interaction traces, then offering those traces for free while prohibiting export [1]. This creates a data lock‑in: enterprises that build massive AI‑generated codebases will need the provider’s service to maintain and understand that code, giving the labs leverage to raise prices later.

Emergence of a secondary market

Because the primary subscriptions are cheap relative to actual compute, third‑party vendors have begun buying bulk seats and reselling them at a discount, while also packaging the trajectory‑storage feature as a value‑add. This mirrors the broader trend of companies shifting workloads to cheaper open‑source models—Lindy moved 100% of its traffic to DeepSeek V4, saving “millions of dollars” after abandoning Anthropic’s Claude [2]. The resale market thus offers a cheaper entry point for firms that cannot afford full‑scale usage, but it also spreads the lock‑in effect across a wider user base.

Competitive implications

The economics pressure forces OpenAI and Anthropic to balance growth against margin erosion. While subscription pricing fuels user adoption, the looming cost gap may push the labs to either raise fees, restrict usage caps, or accelerate the rollout of cheaper, tiered models. Competitors such as DeepSeek and other open‑source offerings stand to gain if enterprises adopt a “mix‑and‑match” strategy, routing high‑value queries to frontier models and routine work to lower‑cost alternatives [2].

What to watch

• Pricing revisions: Any announced changes to ChatGPT Pro or Claude Max subscription rates or usage caps.
• Trajectory‑storage rollout: Implementation timeline for the proposed “trace database” feature and any export restrictions.
• Resale activity: Emergence of new vendors offering discounted bulk subscriptions or bundled lock‑in services.

The clash between flat‑rate subscription appeal and the high cost of heavy AI usage is reshaping how frontier labs monetize their models, while a secondary market is already forming to meet demand for cheaper, locked‑in access. How the labs adjust pricing and data‑ownership policies will determine whether the lock‑in strategy sustains profitability or fuels further market fragmentation.