Microsoft’s $190 Billion In-House AI Bet: A New Strategic Course but Investors Want Hard Returns

Microsoft's AI revenue soars to $37B annualized, but $190B capex and margin pressure drag stock down 6.3% as it shifts to building own models.

Microsoft is charting a radically different course in artificial intelligence, pivoting from bankrolling external labs to building its own models from the ground up. The company hopes this push toward self-reliance will cut costs and cement its independence, but the market is demanding clear proof that the massive spending will generate sustainable profits — a tension that has driven the stock sharply lower.

The numbers tell a story of a business firing on all cylinders operationally yet struggling to convince investors that the payoff is near. Revenue for the third fiscal quarter of 2026 came in at nearly $83 billion, with the AI business now running at an annualized clip of $37 billion. Yet the investment required to get there is eye-watering. Chief Financial Officer Amy Hood has flagged capital expenditure of $190 billion for the current fiscal year, and the resulting depreciation on new data centers has pushed the gross margin to 67.6% — the lowest level since 2022.

Shareholders have responded by heading for the exits. The stock closed the week at €361.70, a drop of 6.3% in five sessions, extending a year-to-date decline of roughly 10%. That has taken the shares below their 200-day moving average of €391.25, a technical level that now acts as a ceiling for any near-term rally. Even a stunning $315 billion surge in market value during May has been largely erased.

The company used its Build developer conference to showcase a new generation of in-house technology. The “MAI” family comprises seven language models developed entirely internally, led by MAI-Thinking-1, which has beaten rivals such as Claude Sonnet 4.6 in independent tests. Additional models are tailored for specific tasks: MAI-Voice-2 handles more than 15 languages, while MAI-Transcribe-1.5 supports 43. New Work IQ APIs give corporate clients access to an intelligent data layer for automated agents, and the entire suite is being tightly integrated with Microsoft’s Copilot assistant.

Alongside the AI push, Microsoft unveiled Majorana 2, a quantum chip it says is 1,000 times more reliable than its predecessor. The company has set a target of building a first scalable quantum computer by 2029.

Analysts remain broadly optimistic despite the share price weakness. Barclays reiterated its “overweight” rating with a price target of $545. Morgan Stanley’s Keith Weiss argues the huge infrastructure investments lay the groundwork for substantial future revenue growth, and he notes the consensus target sits at $561. That optimism is underpinned by a commercial backlog of $627 billion — a massive reservoir of contracted revenue that has yet to be recognized.

The deeper strategic shift is unmistakable. Microsoft’s AI chief, Mustafa Suleyman, has openly stated that the company wants to become one of the world’s top four AI labs, a status it does not currently hold. A key part of the plan is to sever the expensive relationship with Anthropic, slashing and eventually eliminating the license fees the company pays. By developing custom chips and linking its own models directly to Copilot, Microsoft believes it can escape the cost trap that has squeezed margins. The next few earnings reports will be crucial in showing whether the billions being poured into in-house AI and quantum hardware can deliver the earnings performance the stock needs to reclaim lost ground.