RED indicators: 4 (OpenAI circular structure), 2 (capex-demand gap) AMBER-RED: 1 (depreciation — 6-yr life while Amazon shortens), 6 (RPO commitment overhang) GREEN: 3 (insider — routine tax withholding/gifts, not discretionary selling)
Last updated:
Human
The code got written by AI; the decision to cut 20,000+ people and pour $100B into compute was made in a boardroom. AI is the story; capital reallocation is the act.
Microsoft is where the 'AI productivity' claim collides most visibly with the layoff ledger, because both numbers are enormous and both are the company's own. Nadella says AI writes up to 30% of the code; the same year, Microsoft cut roughly 20,000 jobs and spent past $80B on AI data centers — while explicitly exempting the AI teams from the cuts. This dossier tracks the delta between 'AI made us efficient' and 'we moved the money and the headcount toward the bet.'
[Claims][Filings / Data]
"20–30% of Microsoft's code is now written by AI."
— Satya Nadella, LlamaCon fireside with Zuckerberg, April 2025
The same period: ~15,000 roles cut across 2025, 8,750 voluntary buyouts (April 2026, 'Rule-of-70,' explicitly framed against the $80B AI capex), and ~5,000 more in July 2026 (Xbox + commercial sales). AI/Copilot teams were EXEMPTED from both the March 2026 hiring freeze and the buyout program — so AI-written code did not shrink the engineering org; the cuts fell on sales, Xbox, and non-AI units.
CNBC 2025-04-29; SEC 10-K/10-Q (MSFT) capex; 2026 layoff reporting
AI-native go-to-market — 'when AI can qualify leads and handle the first layers of enterprise discovery, you need fewer people doing that.'
— Microsoft restructuring rationale, 2026
Commercial-sales headcount cut while FY2026 capex runs past $100B into AI data centers (FY2025: $80B committed, $64.6B deployed, +45% YoY). The reallocation is explicit: capital and headcount out of people, into compute.
SEC filings (MSFT capex); Bloomberg capex analysis 2026
The AI buildout is sound investment.
— Microsoft investor materials
Microsoft funds above 65% of its capex beyond operating cash flow (the Bubble Watch 'Debt' signal), and its useful-life extension on AI hardware flatters reported earnings (the 'Depreciation' signal). The AI Impact story and the Economics fragility read are the same balance sheet, read two ways.
SEC 10-K (MSFT) cash-flow + depreciation notes — see The Report
The Three Inversions
The Efficiency Lie Sourced
The centerpiece. 'AI writes 30% of the code' is offered as the reason for a leaner Microsoft; the ledger shows a capital reallocation. Microsoft grew its AI teams and its $100B compute line while cutting ~20,000 people from sales, Xbox, and support — and shielded the AI org from every cut. The productivity is real. The layoffs are a decision about where to spend the savings: on the bet, not on the people who used to do the work.
The Safety Ruse Sourced
Microsoft backs the industry push for a single federal AI framework that preempts state law — the posture that favors scaled incumbents over smaller players who cannot absorb 50-state compliance. The Senate stripped a preemption provision 99–1 in 2026, but the lobbying and the super-PAC money behind it continue. Honest nuance: the industry is NOT monolithic — Anthropic gave $20M restricted to public education and argues governments should be able to block dangerous AI, so 'safety regulation' is not a blanket ruse.
The Energy Myth Sourced
Microsoft pledged carbon-negative by 2030; its FY2025 emissions rose 25% (16.2M → 20.3M metric tons CO2e) on the AI data-center build, and it has signalled it may scale the pledge back. It restarted Three Mile Island nuclear for AI power and, like its peers, contracted gas generation. The pledge and the buildout now point in opposite directions. See The Power Draw for who absorbs the grid and water cost. — see full analysis →