Our Take
The G7 is naming the problem without yet naming the solution: you cannot build AI independence without controlling both silicon and minerals, and neither exists at scale outside U.S. or Chinese hands.
Why it matters
AI supply chains are now a geopolitical battleground. Governments are waking up to the fact that relying on U.S. models and Chinese inputs creates leverage points that no Western nation controls independently.
Do this week
Enterprise AI buyers: audit your model sourcing and rare earth dependencies now, before procurement freezes accelerate fragmentation into incompatible regional stacks.
G7 names the dependency trap
The G7 is explicitly confronting two structural vulnerabilities in Western AI infrastructure: reliance on U.S.-built AI models and dependence on Chinese supply chains for critical minerals required in chip manufacturing and battery production. The framing, per Fortune, positions China as "the elephant in the room" that member nations can no longer ignore in the context of AI resilience and economic sovereignty.
This is not a new observation. The supply-chain fragility has been documented repeatedly since 2022. What is new is the collective acknowledgment at the G7 level that partial solutions (subsidizing domestic chip fabs, promoting open-source models) do not solve the fundamental imbalance: the U.S. controls large-language-model development; China controls the minerals pipeline. Neither dependency is easily severed without massive capital reallocation and time horizons measured in decades.
Geopolitical supply chains reshape vendor strategy
The G7 statement signals that national AI policy will increasingly separate into two categories: models and minerals. U.S. AI vendors (OpenAI, Anthropic, Google) will likely face export restrictions or domestic-preference mandates in EU and allied procurement. Simultaneously, European and Japanese chipmakers will be pressured to develop rare-earth alternatives or secure locked contracts with friendly suppliers.
For enterprises, this means two compounding risks. First, the model layer is already concentrating: U.S. closed-model APIs dominate production workloads, and open-source alternatives (Meta's Llama, Mistral) are not yet proven at enterprise scale in regulated industries. Second, mineral access is becoming a policy lever. Companies dependent on Chinese supply chains for manufacturing will face pressure to diversify or relocate production, raising capex and extending timelines.
The pragmatic outcome is not AI independence but rather "managed fragmentation." The EU will fund local models (Mistral, ALEPH Alpha); Japan will secure rare-earth agreements with Australia and other allies; India and other middle powers will accept vendor diversity as the cost of sovereignty. None of this solves the problem faster than 5 to 10 years, during which U.S. and Chinese dominance only deepen.
Lock in model and supply relationships before policy accelerates
Practitioners should move quickly on two fronts. First, evaluate whether your inference workload can run on open models (Llama 3, Mistral) with acceptable latency and cost before enterprise licensing around closed models hardens into regional contracts. Second, map your supply chain for any hardware dependency on rare earths or strategic minerals; if you have one, begin conversations with your procurement team about geographic diversification or long-term fixed-price contracts now, before G7 policy translates into national-security reviews of vendor relationships.
The G7 statement is a slow-motion alarm, not an emergency. But it signals that the window for choosing your stack without geopolitical friction is narrowing. Act before fragmentation becomes mandatory rather than optional.