Among the four tech giants, Microsoft is "lagging behind" (3 core pain points)

1. Over-reliance on OpenAI; in-house models falling behind

- Microsoft's MAI series of in-house large models are far inferior to GPT-4o, Gemini, and Llama 3 in performance, and it still relies on OpenAI to power Copilot.

- The exclusive partnership seems like an advantage, but OpenAI is also seeking compute power from AWS, Google, and Oracle, making "exclusivity" exist in name only.

- The progress of its in-house AI chips, Maia 100/200, is slow, and commercialization is weak, resulting in deep dependence on NVIDIA.

2. Product experience: "Plug-in AI" criticized

- Copilot feels more like an Office/Windows plug-in rather than a natively rebuilt product; Windows AI features are criticized as "laggy, useless, and crammed everywhere."

- In March 2026, Microsoft began cutting redundant Copilot entry points in Windows, admitting to previous "over-integration and poor experience."

- In comparison: Google Gemini, Meta Llama, and Anthropic Claude are more aggressive in multimodal capabilities, reasoning, on-device AI, and open source.

3. Compute power and execution: Resource misallocation, unstable delivery

- High-end GPUs are prioritized for OpenAI, leaving Azure's small and medium-sized AI customers unable to get resources, with market share taken by Oracle and CoreWeave.

- In mid-2024, it temporarily paused AI infrastructure expansion, causing its market share to drop from 38% to 18%.

- Capital expenditure surged (USD 37.5 billion in a single quarter), but the ROIC of its AI business is only 20%, far below the company's overall 35–40%.