✦ The Future BetsNarrow moat

Microsoft (MSFT) — the future bets

Microsoft's future is one bet at four altitudes — thirty million agent seats billing today, an OpenAI position that swung $11 billion between two years, seven homegrown models, and a qubit that finally held still.

Microsoft's future bets are less colorful than Google's — no robotaxis, no glucose monitors — because they are all the same bet at different altitudes: who owns the intelligence layer of work. Agents on top, models underneath, a stake in the frontier, and qubits in the basement.

What the future bets did to reported earnings, FY2026 ($B)+$6.5BGain on the OpenAI position-$4.8BLoss a year earlier$24.1BRevenue booked from OpenAIForm 10-K FY2026, Note 3 and the related-party disclosure
The OpenAI position alone swung reported earnings by more than $11 billion in a year, and Microsoft now publishes a non-GAAP measure whose only job is to take it back out.

Four are live. The one that already bills is Copilot turned agent platform: more than thirty million paid seats by the close of fiscal 2026, roughly ten million of them added in the June quarter1, with agents spreading through the Fortune 500. The one that already shows up in the accounts is the OpenAI position: the October 2025 restructuring left Microsoft with an equity-method interest of about 25% on an as-converted basis, IP rights through 2032 and OpenAI's commitment to buy an incremental $250 billion of Azure2 — and in fiscal 2026 it produced $24.1 billion of revenue and a $6.5 billion net gain, against a $4.8 billion loss the year before3. The hedge is MAI: Mustafa Suleyman's superintelligence team, formed in November 2025, unveiled seven in-house models at Build 2026, built from scratch so the landlord need never rent a brain again4. And the long shot is quantum, where the contested Majorana program answered its doubters with a second chip whose qubits hold coherence for about twenty seconds — and a fault-tolerance target pulled forward to 20295.

Notice what is missing: consumer moonshots. Microsoft bets where it already collects rent, and each wager reinforces the others — agents sell seats, seats justify capex, the models (rented or homegrown) fill the capex, and the stake pays either way. The honest grading: the agent push is measurable now and growing violently; the OpenAI position is now large enough that Microsoft publishes a non-GAAP measure whose only purpose is to take it back out; MAI is seven models old; the qubits are physics with a calendar. Watch the seat count, the revenue Microsoft books from OpenAI against the funding it puts in, and whether MAI models start quietly replacing OpenAI's inside Copilot — that swap, if it comes, re-prices the whole relationship.

Moat trajectory: Widening

Widening on the strength of the one bet that reports: Copilot added roughly ten million paid seats in the June quarter to pass thirty million, agents spread through the Fortune 500, and the OpenAI position swung from a $4.8 billion loss to a $6.5 billion gain while booking $24.1 billion of revenue. MAI and the qubits are too young to grade — but nothing in the portfolio went backward this year.

The number that tests this moat
Reported
Net gains and losses from the OpenAI investment
+$6.5B in FY2026, against -$4.8B in FY2025

The future bets already move reported earnings by billions before any of them ships a product: Microsoft now publishes a non-GAAP measure whose only purpose is to strip OpenAI out. This year's gain came largely from being diluted in OpenAI's recapitalisation. Read the adjusted line, and watch how far the two diverge.

Source: Microsoft Form 10-K, FY2026 — Note 3 ↗
✦ Future bets — beyond today's moat
✦ FUTURE BET 01
Copilot & the Agent Push
Twenty million paid seats and agents in nine of ten Fortune 500s — the container strategy now sells intelligence by the seat.
✦ FUTURE BET 02
The OpenAI Stake & the New Terms
Microsoft booked a $6.5 billion gain on OpenAI largely because it was diluted — and $24.1 billion of revenue from the same company it competes with.
✦ FUTURE BET 03
MAI & the In-House Brain
Suleyman's team was 'set free' to build Microsoft's own brain — seven models from scratch, so the rent on intelligence has a ceiling.
✦ FUTURE BET 04
Quantum & the Topological Bet
The qubit design physics doubted now holds still for twenty seconds — a concentrated bet that error-proof hardware beats everyone's error-correcting…
References
  1. ReportedMicrosoft 365 Copilot reached more than 30 million paid seats by the close of fiscal 2026, roughly 10 million of them added in the June quarter.
    Microsoft, Q4 FY2026 earnings release (Form 8-K, Exhibit 99.1) — quarter ended June 30, 2026: revenue $90,007M (+18%); operating income $40,603M (+18%); net income $35,766M GAAP and $35,286M non-GAAP; diluted EPS $4.81 and $4.74; Microsoft Cloud revenue $59.3B (+27%); commercial remaining performance obligation +84% to $678B; Azure and other cloud services +43%; Productivity and Business Processes $37,847M (+14%); Intelligent Cloud $39,306M (+32%); More Personal Computing $12,854M (-4%); Windows OEM and Devices -7%; XBOX content and services -10%; a $3.2B gain on the investment in Anthropic and lower-than-expected Voluntary Retirement Program expenses contributing $0.27 of EPS against guidance; Azure revenue surpassed $100 billion for the first time and Microsoft 365 Copilot reached over 30 million paid seats; $10.2B returned to shareholders — Q4 FY2026 (quarter ended June 30, 2026) · publ. July 29, 2026 · source ↗
  2. ReportedThe October 2025 restructuring: OpenAI recapitalized as a public benefit corporation, Microsoft holding an as-converted interest with IP rights through 2032 and an incremental $250 billion Azure commitment.
    Microsoft — 'The next chapter of the Microsoft-OpenAI partnership' (Oct 28, 2025): OpenAI recapitalized as a PBC; Microsoft holds ~27% on an as-converted basis (~$135B at the ~$500B valuation); IP rights through 2032; OpenAI committed to an incremental $250B of Azure — Announced Oct 28, 2025 · publ. Oct 28, 2025 · source ↗
  3. ReportedAn approximate 25% as-converted interest; $24.1 billion of revenue from OpenAI in fiscal 2026; a $6.5 billion net gain against a $4.8 billion net loss the year before.
    Microsoft Form 10-K, FY2026 — OpenAI disclosures: an equity-method investment representing an approximate 25% interest on an as-converted basis, accounted for by hypothetical liquidation at book value; revenue from commercial arrangements with OpenAI of $24.1 billion in fiscal 2026 and accounts receivable of $6.0 billion at 30 June 2026; total funding commitments of $13.0 billion of which $11.9 billion funded; other income included $6.5 billion of net gains in fiscal 2026 and $4.8 billion of net losses in fiscal 2025, the gains relating primarily to the dilution gain from the OpenAI Recapitalization — FY2026 (ended June 30, 2026) · publ. July 29, 2026 · source ↗
  4. ReportedThe MAI superintelligence team, formed November 2025, unveiled seven in-house models at Build 2026.
    Press coverage of Microsoft Build 2026 — the MAI superintelligence team (formed Nov 2025 under Mustafa Suleyman) unveiled seven in-house models (incl. MAI-Thinking-1, MAI-Voice-1); Suleyman: 'set free' to pursue superintelligence — Nov 2025 – Build 2026 · publ. 2026 · source ↗
  5. ReportedMajorana 2: coherence of about twenty seconds; the fault-tolerance target pulled forward to 2029.
    The Quantum Insider — Majorana 2 (Build 2026): lead-superconductor redesign, coherence from milliseconds to ~20 seconds (~1,000x), fault-tolerance target pulled from 2033 to 2029, after the 2025 topological-claims debate — Build 2026 · publ. Jun 2, 2026 · source ↗
Sources
Generated September 22, 2026