Microsoft
Microsoft runs a two-track model strategy — its own MAI family plus third-party frontier models — rather than a single in-house stack. Build 2026 (June 2) introduced seven MAI models, led by MAI-Thinking-1, a 35B mixture-of-experts reasoning model trained on commercially licensed data with no OpenAI distillation (97.0% AIME 2025, self-reported), and MAI-Code-1-Flash, a 5B coding model live in all GitHub Copilot tiers; the new Copilot announced September 25 still routes agentic work to 'frontier models like Astra and Fable' from OpenAI and Anthropic. Q4 FY2026 (July 29) pushed full-year Azure revenue past $100B (Azure +43% YoY), with M365 Copilot paid seats doubling quarter-over-quarter to 30M+ and the AI business at a $37B run rate (+123% YoY). On September 25 Microsoft unveiled a Copilot 'super app' — Home (chat plus Cowork task delegation), Code (natural-language app building on GitHub Copilot technology) and Autopilot (the renamed Scout, always-on cloud agents), with agentic work billed by usage; Home and Code reach Frontier-program customers in the coming weeks and Autopilot enters private preview at month-end, so no adoption data exists yet.
Scenarios
$627B cloud RPO and $37B AI run rate growing at 123% YoY mean Microsoft has locked in demand that exceeds current capacity — a rare position for any hardware-constrained vendor. Copilot at ~6.7% penetration of 450M commercial seats (30M paid, doubled in a quarter) has 10-15x runway if enterprise adoption follows typical software S-curves.
$190B CapEx at 123% AI run-rate growth implies Microsoft is betting the balance sheet on sustained exponential demand — if growth decelerates even to 60-70% YoY, the infrastructure will be overbuilt. GitHub Copilot usage-based pricing from June 1 creates churn risk. Capacity constraints are now a growth ceiling, not just a demand signal.
Key Factors to Watch
- ●MAI-Thinking-1 (35B MoE, zero OpenAI distillation, 97.0% AIME 2025) and MAI-Code-1-Flash (5B, 51% SWE-Bench Pro) ship June 2 — first in-house frontier models, live in all Copilot tiers
- ●Q4 FY2026: full-year Azure past $100B (Azure +43% YoY), Copilot 30M paid seats (doubled QoQ), AI run rate $37B (+123% YoY); capacity-constrained through 2026
- ●Agent 365 GA May 1 ($15/user/mo) governs agents across three clouds; Copilot pricing shifts to usage-based AI Credits June 1
- ●Work IQ APIs GA June 16 — M365 semantic layer (A2A/MCP/REST) open to third-party agents on consumption billing; positions M365 as the default context substrate for any enterprise agent
Score History
| Date | Score | Direction | Note |
|---|---|---|---|
| 2026-09-26 | 8.3 | Very Positive | W39 mentioned-deep — HOLD 8.25. Copilot super app (Home, Code, Autopilot) announced September 25 in early access and private preview only, no adoption data. Corrected the summary's opening claim that Microsoft is the only hyperscaler shipping first-party frontier models — the new Copilot itself routes agentic work to OpenAI and Anthropic models. |
| 2026-05-16 | 8.3 | Very Positive | W20 refresh after 52d — score 8.0→8.25 (are 8→9). Q3 FY26: Azure +40% (vs +26% prior), AI run rate $37B (+123% YoY), Copilot 20M seats (+33% seq), $627B cloud RPO. |
| 2026-03-08 | 8.0 | Very Positive | Score 8.2->8.0 (dr 3->4). External research cross-ref: Copilot 3.3% penetration, 40-60% discounting, per-seat pricing model risk from AI agents |
| 2026-03-08 | 8.2 | Very Positive | Initial assessment from batch 1 research |
Software Peers
Last researched: 2026-09-26
This is research and analysis, not financial advice. Scores reflect AI impact potential, not investment recommendations.