The Board Room
Microsoft's 34% crash — its worst quarter since 2008
The market has stopped rewarding AI faith and started demanding receipts, but the CEOs actually producing those receipts are concluding they need dramatically fewer people. Your capital plan and org chart are both built on assumptions that expired this week — stress-test them simultaneously, not sequentially.
AI Investment Thesis Cracks Under Triple Shock
Microsoft down 34% since October (worst since 2008), rate expectations flipped to 52% hike probability in 30 days, and H100 GPU rentals now exceed 2022 launch prices. AI capex faces simultaneous investor revolt, rising cost of capital, and appreciating compute costs — a triple squeeze that invalidates most FY26 plans.
AI Labs Become Vertical Category Killers — Security First
Anthropic's leaked Claude Mythos — described internally as posing 'unprecedented cybersecurity risks' — triggered a sector-wide security selloff. RSA 2026 field intelligence confirms: every security product is becoming a commodity API call in agentic workflows within 1-3 years. This pattern will repeat across every knowledge-work software vertical.
CEO Workforce Compression Goes Public
Dorsey told JPMorgan Tech100 investors that using coding agent Goose convinced him he could nearly halve Block's workforce. Databricks CEO Ghodsi echoed the same pattern. Hashimoto disclosed a parallel agent workflow where agents plan while he codes and code while he reviews. The CEO class is setting 12-18 month headcount expectations based on personal AI usage.
GPU Depreciation Models Break — H100s Are Appreciating Assets
H100 rental prices have reversed sharply upward since Dec 2025, now exceeding Oct 2022 launch-era values. Reasoning and agent demand plus better inference software have made 4-year-old chips more capable than any depreciation schedule assumed. Every cloud GPU contract and FY26 infrastructure budget line is potentially mispriced. Google is funding Anthropic's data center buildout, accelerating the capex arms race.
Meta Assembles Every Layer of the Next Platform
Meta simultaneously scaled brain-reading AI from 4 to 720 subjects (1K→70K voxels), filed FCC paperwork for prescription smart glasses with Wi-Fi 6, committed to 7 new gas power plants, published self-improving hyperagent research, and had Zuckerberg privately coordinating with Musk on DOGE and OpenAI IP. This is a coherent platform-consolidation play across hardware, compute, AI, and political capital.
The AI Investment Thesis Just Hit a Wall — And the Compute Economics Make It Worse
AI Labs Just Became Existential Threats to Entire Software Verticals — Cybersecurity Is the Canary
CEO Workforce Compression Has a Name, a Number, and an 18-Month Clock
- Update: SoftBank secured a $40B unsecured bridge loan from JPMorgan and Goldman to fund its $30B OpenAI commitment — leverage concentration at this scale signals peak AI capital cycle and sets AI valuations disconnected from market equilibrium
- SAP acquired Reltio (master data management) — signaling that clean, unified, governed data estates are now the competitive prerequisite for AI value extraction, not a back-office nice-to-have
- 30% of LLM-generated code contains vulnerabilities while AI tools increase competition entry by 42% without improving success rates — more code from more people with no quality improvement is expanding your attack surface faster than security teams can match
- Open-source models close to within 5% of frontier on coding: GLM-5.1 scored 45.3 vs. Claude Opus 4.6's 47.9 — the gap is functionally irrelevant for most production workloads, eroding closed-model API pricing power
- Midjourney — profitable and bootstrapped — is being squeezed by Google to the point where founder may take VC for the first time, confirming that product-market fit and profitability are necessary but insufficient for survival against hyperscaler competition
- Update: FBI director's personal and official email verified breached alongside Handala's destructive wipe of tens of thousands of Stryker medical devices — state-sponsored cyber ops have shifted from espionage to destructive attacks designed for maximum operational damage
- Whoop is defying FDA guidance on medical features in wearables — a harbinger of the regulatory collision between AI-powered health tech and existing approval frameworks that will affect every company in the space
- Update: Delve's allegedly fraudulent SOC2 and ISO27001 certifications for LiteLLM (3.4M daily downloads) — Karpathy assessed the attack code was 'vibe coded' by AI, meaning the barrier to supply chain attacks has collapsed to near-zero skill requirements
The AI industry just split into two simultaneous realities that your strategy must reconcile: investors are punishing AI spending without receipts (Microsoft down 34%, rate expectations flipping to hikes, H100 costs rising above launch prices), while CEOs who actually use AI agents are publicly concluding they need half the headcount (Dorsey at Block, Ghodsi at Databricks). Meanwhile, Anthropic's leaked cyber-capable model triggered the first selloff of an entire software vertical by an AI lab — a pattern that will repeat across legal, finance, and healthcare within 2-3 years. The organizations that win the next 18 months are those stress-testing their capital plans against higher rates and rising compute costs, measuring agent productivity gains in their own engineering teams this quarter, and building API-composable products before foundation model labs commoditize their vertical from above.