OpenAI Is Gaining Ground on Anthropic with Business Users: What That Says About Vendor Lock-In

Key takeaways

• Enterprise AI customers are switching providers more readily than the "high switching cost" narrative suggests.
• If your own product or workflow is single-provider, you're carrying risk you don't need to carry.
• Multi-provider readiness (not necessarily multi-provider usage) is the practical hedge against a competitor's price cut or capability jump.
• The cost of testing a second provider is now low enough that "we've always used Provider X" is a weak reason to stay put.

What the data actually shows

New reporting indicates businesses are shifting between OpenAI and Anthropic more than the standard "enterprise AI is sticky" narrative would predict, with usage swinging toward whichever lab has shipped the most recent capable, competitively priced model. That's a meaningful data point for anyone assuming their current provider relationship is a long-term moat, on either side of that relationship.

For founders building on top of these providers, this cuts two ways. If you're deep in one provider's ecosystem, the market is telling you that your own customers might not stay loyal to you for the same reason they're not staying loyal to a single model lab: better price-performance elsewhere is worth the switch.

Why switching costs are lower than they look

The traditional argument for vendor stickiness in enterprise software (integration costs, retraining, contract terms) applies less cleanly to LLM APIs than it does to, say, a CRM or an ERP. Most LLM integrations sit behind a thin abstraction layer: a prompt, a request format, maybe a few provider-specific parameters. Rebuilding that layer for a second provider is real work, but it's a fraction of what it costs to migrate a database or retrain a workforce.

That lower switching cost is exactly why the OpenAI/Anthropic seesaw shows up in the data. When switching is cheap, price and capability differences translate into actual behavior change instead of staying theoretical.

The practical hedge: build for portability, not loyalty

You don't need to run a multi-provider production system to benefit from this insight. What you need is the ability to switch quickly if your primary provider raises prices, degrades on your specific tasks, or falls behind on a capability you rely on. That means:

• Keeping your prompts and business logic decoupled from provider-specific quirks where possible.
• Tracking at least one credible alternative's pricing and capability on a recurring basis, not just when something breaks.
• Running a small, current benchmark of your own representative tasks against at least one alternative provider every few months, so switching is a data-backed decision, not a scramble.

What to do with this if you're a founder or engineering lead

The lesson isn't "constantly chase the cheapest model." It's that provider loyalty without a periodic cost and capability check is a quiet source of margin leakage. If your competitors are watching this data and re-pricing their own stacks accordingly, and the data suggests plenty of businesses are, standing still is a decision too.

One line to take away: enterprise AI spend is proving less sticky than anyone assumed a year ago, so treat your provider relationship as a decision you revisit, not a default you inherited. Calcaas lets you compare provider pricing side by side so that revisit takes minutes instead of a full afternoon of spreadsheet work.