Microsoft Pivots to Self-Reliance: Own AI Models Replace Anthropic, Quantum Chip Accelerates

Microsoft orders switch from Anthropic's Claude to in-house MAI models by 2026, while spending $80B on AI infrastructure and unveiling Majorana 2 quantum chip.

Microsoft is walking a high-wire act between fiscal discipline and monumental investment. The tech giant has ordered a halt to costly external AI tools like Anthropic’s Claude Code by June 2026, while simultaneously pouring more than $80 billion into AI data-center infrastructure. That tension is playing out internally as well: the company removed a long-standing question about pay fairness from its employee survey, sparking criticism even as participation rates held at 71%.

“We are over-dependent on OpenAI,” CEO Satya Nadella has admitted, and the new directive from Mustafa Suleyman, head of Microsoft AI, drives that point home. Each developer using external tools cost Microsoft between $500 and $2,000 a month. Suleyman has mandated a switch to the company’s own “MAI” model family, reasoning that internal models can deliver better performance at lower cost. Several business units had already blown their annual AI budgets just months into the fiscal year.

The centerpiece of that family is MAI-Thinking-1, a reasoning system that in internal blind tests surpassed Anthropic’s flagship Claude Opus 4.6 on coding tasks — and at a cheaper price. It was trained exclusively on Microsoft’s own data, without distilling OpenAI’s technology, after a renegotiated partnership with OpenAI granted the right to build independent models. The broader MAI lineup includes MAI-Code-1-Flash for developers, MAI-Transcribe-1.5 covering 43 languages, and specialized tools for image and speech processing.

Analysts are taking notice. TD Cowen maintained a Buy rating and a $540 price target after the Build announcements, calling the shift a “new phase” in Microsoft’s AI development. Barclays rates the stock Overweight with a $545 target, citing an accelerated quantum computing program and the pivot to AI agents as a core business model. The firm noted that GPU capacity is being redirected from external partnerships to in-house research.

On the hardware side, Microsoft unveiled the Majorana 2 quantum chip, promising a thousand-fold improvement in qubit reliability and a parity lifetime of 20 seconds. The company now targets a commercially viable quantum computer by 2029 — years ahead of earlier forecasts. Wedbush Securities called the news a “validation” of the quantum industry, though skeptics are demanding independent proof of the claims.

Enterprise adoption is marching forward. Infosys, TCS, and Wipro together have deployed more than 300,000 Microsoft 365 Copilot licenses in the past six months. A new integration between Pinecone Nexus and Microsoft OneLake is said to cut large language model token consumption by over 95%, making AI agents 30 times faster.

The stock tells a more cautious story. Microsoft shares trade near $441, roughly 9–10% below where they started the year and below the 200-day moving average of about $458. In recent weeks the price was essentially flat at €368.50, with a neutral relative strength index of 53.3. The stock has gained nearly 5% over the past 30 days, but that still leaves it 23% below its 52-week high of $478.10. For all the technological ambition, the market is waiting for the payoff to hit the bottom line.