Lead:
Microsoft, OpenAI's biggest partner and 27% shareholder, is tightening control of its AI destiny. On February 12, 2026, Microsoft AI chief Mustafa Suleyman (formerly of DeepMind) told the Financial Times the company will build in-house frontier models - starting with MAI-1 - so it can be self-sufficient even as the OpenAI partnership continues. See the report at Financial Times.
What changed:
The October 28, 2025 restructuring gave Microsoft a 27% equity stake in OpenAI, rights to use OpenAI technology through 2032, and a large Azure commitment from OpenAI (about $250 billion). The deal also loosened exclusivity so Microsoft can compete when needed. Translation: they remain partners, but with more freedom to act. Read more at GeekWire.
Define the jargon, fast:
Foundation/frontier model - a very large, general-purpose AI system that powers many apps (copilots, agents). 'Frontier' just means the newest, biggest, and most expensive models. See Financial Times.
MAI-1 - Microsoft’s planned high-end model intended to reduce reliance on GPT. Timing is expected in 2026; specs are not public. Reports suggest very large scale, but Microsoft has not confirmed details. See Financial Times.
The money shot:
Last quarter Microsoft logged $37.5 billion in capital expenditure (capex), with roughly two-thirds devoted to AI infrastructure (GPUs and data centers). Commercial remaining performance obligations (the revenue backlog from signed contracts to be recognized over roughly 2.5 years) hit $625 billion - and about 45% of that ties to OpenAI, according to CFO Amy Hood. Markets noticed: on January 29 Microsoft lost around $357 billion in market value in a day. See the company filing at Microsoft Investor.
OpenAI's reality:
OpenAI's revenue is scaling fast (annualized run rate above $20 billion as of late 2025), but costs are rising too - spending projected to total about $115 billion through 2029. Ads started appearing in ChatGPT on February 9, 2026; soon after, a researcher publicly quit and criticized the move. Only about 5% of ChatGPT's roughly 800 million users were paying as of mid-2025, which helps explain the ad experiment. For revenue context see Yahoo Finance.
Why founders should care:
Vendor risk is now quantified. If roughly 45% of Microsoft's $625 billion backlog hinges on one partner, relying on a single model or vendor is not a strategy - it's a liability. Build an abstraction layer now. See discussion at The Motley Fool.
Costs and latency will move as infrastructure gets reshuffled. Budget for 20-50% variance in API costs or rate limits over the next 12 months. That $37.5 billion capex is real. See Microsoft Investor.
Action plan (90 days): audit every SLA for third-party model calls; list your top 10 AI APIs; dual-track at least one open model (for example Llama or Mistral) and one non-OpenAI closed model (for example Anthropic or Google); add a model-router to switch providers; pin data residency and failover policies in writing. See practical notes at Financial Times.
Bottom line:
Microsoft is still in OpenAI's corner, but it's building a plan B at frontier scale. If your product depends on a single model or vendor, treat this as a wake-up call and build redundancy now.
Get daily insider tech news delivered to your inbox every weekday morning.
