The Most Expensive Dependency

That number — the share of Microsoft's coverage about money, not products — has more than doubled in two years. In the same period, the share classified as product launches dropped from 33% to 23%. Microsoft is being covered more for what it spends than for what it ships. On April 2, the company's AI chief explained why.

The Admission

Mustafa Suleyman, Microsoft's head of AI, told the Financial Times that the company is "not able to build models in the very largest scale yet" — but that its "computation ramp is coming" to enable it later in 2026. On the same day, Microsoft launched MAI-Transcribe-1, MAI-Voice-1, and MAI-Agent-1 — in-house models for specific tasks. Small models. Specialized models. Not frontier models.

Microsoft can build tools. It cannot build the thing the tools run on. After an investment in OpenAI that now stands at approximately $135 billion — the largest corporate AI position in history — the company that funded the frontier still cannot produce it.

The Profile

Microsoft's edge profile over eight quarters:

Read the two columns together. Financial rises from 15% to 37%. Launch falls from 33% to 23%. The crossover happened in Q3 2025, when financial coverage first exceeded launch coverage. By Q1 2026, the gap is 14 percentage points — Microsoft is covered nearly twice as much for spending as for shipping.

And note the new column. "Competitor" appeared at 9% in Q1 2026 — a category that didn't exist for Microsoft in any prior quarter. The company that was everyone's partner is now, for the first time, being covered as someone's rival.

The Hedging

In July 2019, Microsoft invested $1 billion in OpenAI. By January 2023, the total was reportedly $10 billion. The thesis: fund the best AI lab, integrate its models into Office, Azure, and GitHub, let the partnership deliver the capability. For two years, it worked — 33% launch coverage in Q1 2024. Then OpenAI's board fired and rehired Sam Altman in a single weekend in November 2023, and Microsoft had no warning. The company began hedging a dependency it had just discovered was fragile.

• May 2024 Sources: Microsoft is training MAI-1, an in-house model with ~500 billion parameters. The first serious internal effort.
• AUG 2024
  In its annual report, Microsoft lists OpenAI as a competitor in AI offerings. The partner is now a rival — in a regulatory filing.
• Mar 2025 Sources: Microsoft is testing ways to replace OpenAI's models in Copilot. Separately, it completes training a family of models codenamed Maia.
• SEP 2025
  Two stories in the same month. Sources: Microsoft will use Anthropic's models for some AI features. And: a clause rescinds Microsoft's access to OpenAI's most powerful technology if OpenAI achieves AGI.
• Nov 2025 Suleyman lays out plans to develop in-house models. The aspiration is stated publicly for the first time.
• FEB 2026
  Suleyman says Microsoft is pursuing "true self-sufficiency" in AI. The word "self-sufficiency" is a dependency admission dressed as ambition.

The sequence: 2024, train a model. 2025, test replacements, add Anthropic, discover the AGI clause limits your access. 2026, declare self-sufficiency as a goal. April 2026, admit you're not there yet.

You don't pursue self-sufficiency from something you control. You pursue it from something you depend on.

Compared to What

Microsoft holds $135 billion in OpenAI — a 27% stake. By any financial measure — multiples of what it paid — this is a successful investment. But the coverage asks a different question. The $135 billion bought distribution rights — the ability to embed OpenAI's models into Microsoft's products. It did not buy the capability to build those models.

Microsoft's financial edge share — 37% — is worth benchmarking:

Among the major AI companies, only Microsoft has a financial share that exceeds its launch share. Every other entity is covered more for what it ships than for what it spends. The investment didn't fail — it produced an enormous financial return. But it produced a dependency, not a capability.

The CFO confirmed the rebalancing from a different angle. Amy Hood paused some data center expansion — a capital allocation decision visible in the coverage shift quarters before she announced it.

37%

37% means that more than a third of how the world's reporters describe Microsoft's actions are about money — deals, investments, capital allocation, restructuring. In a company that employs 200,000 people and ships products used by billions, the thing the press covers most is the spending. Not because reporters are lazy. Because the spending is what's structurally interesting about Microsoft right now. The company's most consequential decisions are financial, not technical. Where the capital goes, what it buys, and what it doesn't buy.

What it doesn't buy, it turns out, is the ability to build frontier AI models. That's what Suleyman said on April 2. The coverage showed it first.