The Gist: Executive Overview

The Pentagon's designation of Anthropic as a supply chain risk and London's immediate recruitment effort reveal how security restrictions are fragmenting the allied AI ecosystem rather than consolidating it. This dynamic extends beyond bilateral tensions — China is deploying provincial industrial policy through frameworks like OpenClaw in Shenzhen to build domestic AI adoption clusters resilient to U.S. export controls, while Japan positions itself as a middle ground where Western AI talent can access Asian manufacturing without navigating the full complexity of U.S.-China decoupling.

The pattern suggests national security classifications are evolving into industrial policy instruments. Allied governments are discovering that offering more permissive regulatory environments than Washington — while remaining within the Western sphere — creates opportunities to attract frontier AI capabilities. Meanwhile, U.S. export controls may be accelerating rather than hindering competitor development by forcing autarkic innovation within large markets where state coordination substitutes for market adoption.

The historic memory chip shortage driven by AI data centres represents a fundamental shift in semiconductor dependencies, moving from logic chip concerns to high-bandwidth memory where South Korea dominates production. This coincides with premium valuations for data centre cooling infrastructure (KKR's multibillion-dollar CoolIT exit) and the emergence of worker housing as a commercial opportunity for remote AI facility construction. The convergence signals that AI scaling trajectories face compounding physical bottlenecks across energy, memory, thermal management, and labour — not the single-constraint optimisation problems labs have successfully addressed previously.

These constraints are already creating allocation conflicts. Memory manufacturers face strategic choices between high-margin AI customers and consumer electronics producers, potentially pricing consumer device makers out of supply. The resolution of these tensions will determine both AI scaling roadmaps and competitive dynamics in hardware-dependent sectors, while concentrating strategic leverage in the small number of firms controlling critical chokepoints.

Major platforms' AI products demonstrably fail to prevent documented harms with clear causal pathways to injury — facilitating illegal gambling to vulnerable users, enabling mass deanonymisation attacks, and serving as unregulated therapy tools for trauma survivors. These failures occur despite voluntary safety commitments and existing content moderation systems, exposing a structural misalignment where safety resources flow to speculative future risks while measurable present-day harms receive inadequate attention.

The pattern extends to accountability gaps in capability claims. Block's justification for eliminating 4,000 positions based on AI productivity gains faces employee disputes that the technology cannot perform eliminated work, yet no mechanism verifies whether automation assertions match operational reality before consequential workforce decisions. Similarly, AI systems perform functions subject to professional licensing and liability in traditional contexts — clinical therapy, employment decisions — but face no equivalent requirements when delivered via chatbot.