Source: Microsoft is testing out ways to replace OpenAI's models in Copilot with models from other AI labs including xAI, Meta, and DeepSeek
Aaron Holmes / The Information :
Context & Ripple Effects
Microsoft had previously deepened Copilot’s use of OpenAI models, including a GPT-4 Turbo Copilot update, but later reporting indicated it was exploring additional first- and third-party models for 365 Copilot amid cost and speed concerns.
This testing broadens that diversification effort beyond a single supplier: Microsoft is evaluating external labs alongside its reported MAI model work.
First-order effects
- Microsoft can benchmark OpenAI’s models against offerings from xAI, Meta, and DeepSeek for Copilot workloads rather than treating OpenAI as the default option.
- OpenAI faces a more contested position inside a major distribution channel, while the named alternative labs gain a potential route to enterprise-scale Copilot usage.
Second-order effects
- A multi-model Copilot architecture would give Microsoft more leverage over model pricing, performance, and availability, reinforcing the earlier push to add third-party models to 365 Copilot.
- Model providers seeking large enterprise deployments may have greater incentive to meet Microsoft’s integration, reliability, and cost requirements.
Third-order effects
- If large software platforms routinely switch models by workload, foundation models become more interchangeable infrastructure and buyers gain bargaining power over individual labs.
- The enduring competitive advantage may shift toward the platform layer—distribution, product integration, and routing across models—rather than exclusive reliance on one model partner.
The trend: This is one data point in the shift from exclusive foundation-model partnerships toward disciplined multi-model procurement by major AI product platforms.