Renting AI from foreign providers is a national security risk, warns Cohere CEO

Cohere CEO Aidan Gomez argues that foreign dependence on frontier AI models is a structural national security risk, especially after the U.S. restricted access to Anthropic’s Mythos. He said democracies should build sovereign AI stacks with domestic control over chips, data centers, and models, and back a small set of trusted champions through alliances rather than trying to replicate the full ecosystem in every country. The article also notes Cohere’s planned acquisition of Aleph Alpha to form a roughly $20 billion transatlantic AI company.

Analysis

The tradeable implication is not “sovereign AI” in the abstract, but a slow re-rating of control-plane infrastructure: domestic cloud, on-prem deployment software, private networking, air-gapped security, and non-U.S. compute orchestration. The winners are likely the picks-and-shovels vendors that make models deployable inside regulated environments, while the losers are pure API-dependent model providers whose customer value proposition can be interrupted by policy rather than product quality. Over time, this should also broaden procurement budgets away from centralized frontier labs toward regional systems integrators and defense/critical-infrastructure vendors that can certify end-to-end control.

The second-order effect is budget fragmentation: governments and large enterprises will likely fund multiple smaller “good enough” models instead of one best-in-class foreign model, especially where latency, data residency, or export-control exposure matters. That is bearish for winner-take-all economics in frontier model platforms, but bullish for middleware, inference optimization, cybersecurity, and local data-center operators. The most important time horizon is 6-24 months, because policy change will first alter procurement language, then capex allocation, then vendor share.

The key risk is that sovereignty becomes a political slogan without incremental capex. If regulators simply require data localization while continuing to allow foreign-managed APIs, the economics barely change; if, however, allied governments begin subsidizing domestic infrastructure and mandating on-prem deployment for defense, health, and finance, the shift becomes real. The contrarian read is that this is not an outright anti-U.S. trade — it is a diversification trade, meaning the market may be underpricing non-U.S. beneficiaries that can sell “trusted” AI without needing to win the frontier race. The reversal trigger would be a loosening of export-control rhetoric or a breakthrough in cross-border trust frameworks that reduces the premium for control.

Key Decisions for Investors

• Long hyperscale data-center and sovereign-cloud beneficiaries on any pullback over the next 1-3 months; pair against pure frontier-model exposure. Thesis: procurement shifts favor owned infrastructure over rented APIs, supporting sustained capex and utilization.
• Long cybersecurity / air-gapped deployment enablers for 6-12 months. Focus on vendors with on-prem, regulated-environment workflows; expected upside comes from higher attach rates as customers internalize model hosting.
• Pair trade: long non-U.S. enterprise AI enablers vs short U.S. API-first model monetization proxies. Time horizon 6-18 months; the spread should widen if governments convert sovereignty rhetoric into procurement mandates.
• Use call spreads on defense/critical-infrastructure software names that sell into government and regulated sectors. Best entry is after any policy headline-driven selloff; catalyst window is upcoming budget cycles and tender awards.
• Avoid chasing headline-only sovereign-AI names until there is evidence of funded capex or binding regulation. The risk/reward is poor if the theme stays rhetorical rather than operational.