Carnegie report: middle powers and an 'AI Breakout Posture'
The Carnegie Endowment for International Peace proposes an "AI Breakout Posture" for middle powers such as Germany, Britain, India and Japan, which the report says cannot afford a sovereign frontier-model moonshot yet cannot stake national security on US suppliers never cutting them off; it argues that all three conventional paths — sovereign build-out, pure adoption of foreign models, and negotiating guaranteed access — fail. Under the proposal, these states would not launch a frontier sprint today but would pre-build the prerequisites to do so on short notice: latent compute that can be redirected on demand (a single frontier training run may need only a single-digit share of global capacity), private-sector and talent capacity that can be mobilized quickly, and legal authority plus cross-party commitments, all while staying closely aligned with the United States rather than posturing against it. The report also flags coding agents such as Claude Code and Codex as emerging strategic goods, and contrasts India's tolerated nuclear latency with Iran's exposed and vulnerable program to argue against both overconfidence and fatalism.
It moves the middle-power question from slogans to cost accounting: the scarce resource may not be the money to build a frontier model, but the political will to pay now for a capability that may never be used.