Enterprise AI loyalty is a myth—companies are model tourists, not committed customers.
The Summary
- OpenAI is closing the gap with Anthropic in enterprise adoption, with businesses switching providers as each lab ships newer models
- Customer "stickiness" in enterprise AI is proving far weaker than investors assumed, creating volatility in market share
- The willingness to swap providers suggests AI models are commodifying faster than the labs can build moats
The Signal
The enterprise AI market is revealing an uncomfortable truth: businesses treat frontier models like SaaS trials, not strategic infrastructure. New data shows OpenAI regaining ground against Anthropic in business deployments, but the real story is the churn. Companies are switching between OpenAI and Anthropic with each model release, behavior that would make any B2B CFO nervous.
This isn't how enterprise software is supposed to work. When a company commits to Salesforce or SAP, switching costs keep them locked in for years. Training data, workflow integration, employee habits—all of it creates friction that protects revenue. But frontier model providers are learning that their moats are shallow.
"Businesses are willing to flop back and forth as each lab releases new models."
The pattern is clear: Claude 3.5 drops, enterprises migrate to Anthropic. GPT-5 ships, they migrate back to OpenAI. The next Gemini release will trigger another round of musical chairs. What looks like market volatility is actually price discovery—the market is telling us that model capabilities are the only differentiator that matters, and capabilities are temporary.
This has second-order effects that ripple through the agent economy:
- API lock-in is dead. Developers building on these models know to abstract their provider layer from day one
- Model routers are the new kingmakers. Services that automatically switch between providers based on performance and price are becoming critical infrastructure
- Inference costs matter more than brand. When outputs are comparable, the cheapest tokens win
For the labs, this means the race isn't just about capability anymore. It's about building something enterprises can't easily replace. OpenAI is betting on custom models and deeper workspace integrations. Anthropic is pushing constitutional AI and safety guarantees as differentiators. Neither has cracked the code yet.
The Implication
If you're building agent infrastructure, don't marry a single model provider. The enterprises with the most leverage are the ones who've built their systems to be model-agnostic. Watch for the emergence of middleware companies that manage model selection and routing—that's where the real stickiness will develop.
For the frontier labs, the message is harder: shipping new models isn't enough. You need lock-in that survives the next benchmark leap from your competitor. Otherwise, you're just renting customers between releases.