The Board Room
China is subsidizing AI models at 1/40th the cost of US equivalents per token
A startup in Lagos or Jakarta choosing which AI to build on faces a 40:1 price gap, and those models embed CCP-mandated ideological alignment by Chinese regulation. Simultaneously, Pentagon procurement reform just opened ~$1T in annual defense spending to commercial AI companies for the first time.
China's 40x AI Subsidy Wages a Global Platform Default War
China is subsidizing AI at 1/40th US cost per token to capture global platform defaults — the same 'involution' playbook that overwhelmed solar and EV markets. Pentagon procurement reform opens ~$1T to commercial AI. US government reversed: open-source AI is now a national security imperative, not a threat.
$700B in Hidden AI Infrastructure Commitments Create Systemic Margin Risk
Big Tech has quietly committed $700B+ in off-balance-sheet AI infrastructure leases. Oracle alone holds $260B. Meta's contractual commitments quadrupled to $131B in 12 months. Apple's $14B contrarian bet assumes models commoditize — two separate $1B world-model raises suggest the smart money agrees.
AI Value Chain Inverts: Integration Layer Captures Value as Models Commoditize
Palantir's 109% US commercial revenue growth vs. SaaS incumbents' ~10% proves the integration layer thesis. Foundation model makers are verticalizing into apps (Anthropic acquired Vercept, OpenAI Codex hit 2M+ WAU). VC consensus: proprietary data is the last defensible moat — software logic moats are nearly worthless.
Developer Supply Chain Under Industrial-Scale Attack
GlassWorm campaign weaponized LLM-generated code to seed 72 malicious IDE extensions and 151 GitHub repos in six weeks, using Solana blockchain for untakeable C2. Palo Alto Cortex XDR found to silently exempt ~50% of detections via hardcoded whitelists. Ransomware negotiators colluded with ALPHV BlackCat across $75M+ in payments.
AI Paradigm Bifurcation: World Models Challenge LLM Dominance
Yann LeCun left Meta, raised $1.03B at $3.5B valuation for AMI Labs targeting physical intelligence via JEPA architecture. Fei-Fei Li raised $1B separately. Physical Intelligence has $1B+ for robotics AI. Three billion-dollar bets that LLMs aren't the endgame — backed by NVIDIA, Toyota, Samsung, and Temasek.
China's 40x AI Subsidy Is a Platform Default War — and the US Is Losing on Diffusion
$700B in Off-Balance-Sheet AI Leases — The Industry's Hidden Margin Trap
GlassWorm + ALPHV Collusion: Your Developer Pipeline and Incident Response Trust Model Are Both Compromised
LeCun's $1B AMI Labs Splits AI Into Two Paradigms — Your Strategy Needs a Physical Intelligence Thesis
- Update: AI coding velocity trap gets hard data — Cursor users show 41% more commits but 38% more reverts, suggesting net reliable output improvement may be near zero while technical debt accumulates
- Meta becomes first major platform to roll back encryption — Instagram E2EE deprecated effective May 8; internal documents reveal the 2019 encryption push was known to be 'irresponsible' by Meta's own policy chief
- OpenAI Codex hits 2M+ WAU (4x YTD) and GPT-5.4 generated $1B in net-new annualized revenue within its first week — the coding agent category has crossed from experimental to revenue-generating at scale
- 47-day TLS certificate mandate starts its glide path — 200-day max effective March 2026, 100-day by March 2027, 47-day by 2029; organizations without ACME-based automation face recurring outages
- Frontier LLMs produce identical trading outputs across 150 experiments — GPT-5.4, Opus 4.6, Sonnet 4.6, Gemini 3.1 Pro, and Qwen3 Max all recommend the same trades, even at high temperature; any product thesis built on 'AI-powered unique insights' with commodity models is invalidated
- 8 competing 'AI-free' certification groups emerge with no unified standard — same pre-regulatory fragmentation that preceded organic food labeling and GDPR, signaling a two-tier market where 'human-made' commands a premium
- China approves world's first commercial invasive brain-computer interface (Neuracle Medical) — epidural approach less invasive than Neuralink, generating clinical data while US competitors remain in trials
- Anthropic built Claude Cowork (production desktop agent) in 10 days using AI-on-AI orchestration; simple markdown Skills files outperform custom-built MCP tool integrations — the 'connector economy' faces rapid depreciation
- Kimi/Moonshot AI's Attention Residuals achieve equivalent model performance at 20% less compute with <2% inference overhead — the first architectural change to residual connections in a decade, led by a Chinese lab
China is subsidizing AI at 1/40th US cost to capture the global platform default while American hyperscalers have quietly committed $700B in off-balance-sheet infrastructure leases on models that three separate billion-dollar 'world model' bets suggest may be transitional — and meanwhile, your developer tools are under industrial-scale supply chain attack from campaigns using Solana blockchain for untakeable command-and-control. The three actions this week: model your pricing against a 40x Chinese subsidy scenario, stress-test your P&L against AI infrastructure costs that don't decline, and hunt for GlassWorm indicators in your engineering environment today.