AI models are becoming the new semiconductors — assets governments want to control at the border.
Anthropic recently restricted two of its most advanced models to US users only, citing national security guidance. The move follows a pattern: first chips, now models. The US is extending its technology blockade* deeper into the AI stack.
*A technology blockade, or export control, is a government restriction on selling or transferring certain technologies to foreign countries or users. The US has used this tool aggressively on advanced chips since 2022.*
The broader trend is called AI fragmentation — where the US, EU, and China each develop, restrict, and silo access to their own AI systems.
What fragmentation means for markets
When models get locked behind national borders, companies outside the US face a harder choice: build their own (expensive, slow) or work around restrictions (limited, unreliable). Either way, the layer that sits above any specific model — the infrastructure that connects data, workflows, and decision-making across an organization — becomes more valuable, not less.
This is the platform thesis*: that durable value in AI belongs not to whichever model wins the benchmark race, but to whoever provides the connective tissue that makes AI usable at scale, regardless of the underlying model.
*A platform thesis is a bet that infrastructure supporting an ecosystem benefits no matter which specific technology dominates within it.*
The historical parallel: during the electrification era, the companies selling wires and switchgear often outlasted the ones fighting over AC versus DC power standards.
What changes — and what doesn’t
Export controls change access. They don’t change the underlying demand for AI integration, workflow automation, or decision-support tools. If anything, fragmentation adds complexity — and complexity increases demand for tools that manage it.
My take:
AI blockades don’t slow AI adoption. They slow the adoption of specific models. For investors focused on the platform layer — where the work of actually using AI gets done — the export control era may look less like a headwind and more like a tailwind.
This is educational content, not investment advice.