The Board Room
Microsoft killed its 'AI everywhere' strategy this week
Your two most urgent recalibrations: triage the AI roadmap to margin-positive outcomes only, and assume your entire codebase is one commodity scan away from full exposure. The era of shipping AI as a feature flag just received its death certificate from the company with the most distribution on earth.
Microsoft's AI Retreat Validates Outcome-Only Thesis
Microsoft confirmed broad AI distribution destroys margins. 81 products rationalized, inference costs dragging earnings. Anthropic's opposite bet — per-result pricing, focused agent outcomes — grew revenue 80x. The 30-point gross margin gap between AI-native (50-60%) and traditional SaaS (80-90%) is structural, not transitional.
Offense Commoditized at $30/Scan — Defenders Stuck at 55 Days
Zero-day discovery now costs $30-$150 per codebase. Red team agents achieve 95% domain dominance in under 6 minutes. Mozilla found 271 Firefox bugs in one Mythos pass. The defender's 55-day remediation average and 135 new CVEs/day mean the gap is mathematically unfixable with human-speed processes.
Agent Load Breaks Infrastructure — GitHub at 85% Uptime
GitHub's uptime fell to 85% — 2-3 hours daily downtime — under AI agent load 30x above architecture assumptions. CTO revised scaling target from 10x to 30x in 4 months. Competitors (GitLab, Vercel, Linear) absorbing same growth without failures. Mitchell Hashimoto publicly declared GitHub 'unfit for professional work.'
State AI Law Closes 'Algorithm Did It' Defense
Connecticut passed an omnibus AI law (131-17 House, 32-4 Senate) explicitly removing automated decision-making as a defense in discrimination cases. Federal regulation most likely arrives via NDAA, not standalone bill — meaning compressed timelines and less debate. The 12-18 month window to shape vs. react is closing.
Per-Seat SaaS Pricing Enters Terminal Phase
Stripe shipped 280 features for agentic commerce. HubSpot declared full API parity with UI as survival strategy. Anthropic moved to per-result pricing. When AI agents become primary SaaS consumers, per-seat models see higher utilization but flat revenue — a paradox with a 3-5 year fuse.
The AI Feature Sprawl Death Certificate — And the Margin Math That Killed It
Microsoft Proved the Negative
Microsoft's Copilot rationalization is the most instructive strategic signal in enterprise AI this quarter. A company with unlimited frontier-model access, 400 million Office users, and effectively unlimited capital concluded that broad AI distribution destroys value. Customer feedback produced the phrase 'functionally useless.' The earnings call confirmed that inference costs drag margins. Eighty-one distinct products were in flight. Nadella's sequence was consolidation under one executive (Andreou), then killing everything that failed both a customer-value test and a unit-economics test.
If breadth-first AI feature sprawl does not work inside the largest software distribution on earth, the question of whether it works inside a smaller one answers itself.
The Structural Margin Gap
A reasonable skeptic would call this a maturity problem that scale resolves. The skeptic is wrong, and the numbers say so. BVP puts AI company gross margins at 50-60% against 80-90% for traditional SaaS. Reasoning models consume 10-100x the tokens of the prior generation for the same user-visible answer. OpenAI's 1,000x cost reduction over 14 months was eaten by its own model advances. Per-token cost falls. Tokens consumed per task rise faster. A company shipping AI on every surface is running thirty margin-negative line items to fund the one that pays for itself.
The Opposite Bet Is Working
Anthropic's shift to per-result pricing, charging for outcomes rather than tokens, is the structural alternative. It works because the agents complete the work, with a 90% autonomy target for Claude Code, which makes the vendor's eat-the-cost risk acceptable. Focused 365 Copilot, the part Microsoft is keeping, grew paying users 33%. Narrow, high-value surfaces where customers pay on purpose outperform broad feature spray by every measure that matters in year two.
Portfolio Implications
Every AI feature shipped on inference without corresponding willingness-to-pay is a standing cost against non-existent revenue. The audit is straightforward. Map every AI-powered feature to customer-perceived value and to inference cost. Anything that fails both tests is a margin leak that compounds with scale. Microsoft absorbed it for 18 months. Most organizations cannot absorb it for one quarter.
The era of competing everywhere with undifferentiated AI is over. Microsoft proved it does not work with infinite resources, which is useful to know before spending finite ones.
- Update: Anthropic reported $30B ARR (80x Q1 growth) — leasing 100% of xAI's Colossus 1 at ~$5B/year because even Google Cloud + AWS cannot supply enough inference compute
- Google licensing Gemini through PE firms (Blackstone, KKR, EQT) for portfolio-wide deployment — trading per-deal margin for distribution velocity across thousands of enterprises
- Uber exhausted its entire 2026 AI budget by mid-April and is now cannibalizing hiring budgets to fund AI compute — confirms AI is displacing headcount as primary opex growth category
- Meta's internal AI token leaderboard gamed within weeks — engineers scripted millions of tokens for no productive purpose, proving input metrics for AI adoption are fundamentally ungovernable
- RAG accuracy collapses from 90.7% to 50.6% when corpus scales from 5K to 500K documents — the demo was never the product, knowledge-graph architectures emerging as successor
- Palo Alto Networks zero-day (CVE-2026-0300): unauthenticated RCE as root on PA-Series/VM-Series firewalls — no patch until May 13, CISA confirms active exploitation
- DPRK IT worker fraud industrialized at scale — 70+ companies (including Fortune 500) infiltrated via remote engineering roles, $1.2M generated from just two caught facilitators
- Pinterest hit first $1B quarter on 80B monthly visual searches — 24% higher conversion than social engagement, proving commercial-intent data is the new ad moat
- Google's WebMCP protocol turns every website into a callable service for AI agents — 6-12 month first-mover window before it becomes table stakes
- AI productivity gains plateau at 6 months in organizations that don't redesign operating models — the honest budget split is 50/50 tech vs. org redesign, most run 90/10
Microsoft just proved that distributing AI features broadly destroys margins even with unlimited resources, while AI-powered offense hit $30 per zero-day scan with 95% automated success rates — and the platforms your engineering relies on (GitHub at 85% uptime) are breaking under agent load nobody planned for. The three recalibrations this quarter are: triage every AI feature for margin contribution, assume your codebase is already scanned by commodity tools, and stress-test your infrastructure against 30x the load you architected for.