Microsoft AI Head Suleyman Claims Anthropic Models Are Too Expensive, Accelerating Development of In-House MAI Series to Reduce Dependency
Mustafa Suleyman, head of Microsoft AI, stated that Anthropic models are extremely expensive, prompting many users to urgently seek alternatives. Microsoft currently pays substantial fees to Anthropic, and the team aims to reduce and ultimately eliminate related procurement costs.
To break free from external dependencies, Microsoft announced seven self-developed MAI series models at the June Build 2026 developer conference, including the MAI-Thinking-1, which has reasoning capabilities. The new models significantly lower Token costs while maintaining performance and can compete with Anthropic's flagship model, Claude 4.6 Opus, in tasks such as programming.
Internally, Microsoft has adjusted its toolchain to gradually reduce reliance on third-party tools like Claude Code, shifting towards its self-developed GitHub Copilot CLI to control Token consumption.
Source: Public Information
ABAB AI Insight
Mustafa Suleyman co-founded DeepMind previously, and this statement continues Microsoft's strategy of shifting from reliance on external models to full-stack in-house development. In April 2026, Microsoft revised its agreement with OpenAI, ending exclusive licensing and terminating revenue sharing, clearing the way for the independent development of the MAI series.
On the capital front, Microsoft is converting the substantial Token fees previously paid to labs like Anthropic into internal R&D investments, reducing long-term costs through self-developed models while strengthening the competitiveness of the Azure platform, aiming to create a dual advantage in cost and performance in the enterprise AI market.
This adjustment is similar to Microsoft's historical vertical integration in cloud services and reflects the current trend among major tech companies shifting from "buying models" to "building models + proprietary infrastructure." The AI infrastructure sector is currently undergoing a restructuring phase driven by cost pressures.
Essentially, this is a restructuring of the industry chain: the reliance on purchasing large models is migrating towards a self-developed closed loop, as high external costs force giants to concentrate pricing power from third-party labs to their own training, optimization, and distribution stacks, with capital continuously flowing towards tech platforms with full-stack capabilities.
ABAB News · Cognitive Law
The more expensive the model, the more valuable the autonomy; true sovereignty begins with stopping the payment of "taxes" to others.
Dependency on external sources is always a temporary remedy; major companies will ultimately turn the most expensive parts into their own assets.
Token costs are not expenses but strategic variables; whoever first converts spending into proprietary capabilities will hold long-term pricing power in the AI era.