Should countries aim to build and control their own AI datacenters?

Confidence summary: High confidence that some domestic capability is necessary; moderate confidence on optimal scale and scope.

1. The core argument

NVIDIA's 80% chip monopoly and America's cloud infrastructure dominance create the same strategic vulnerability that OPEC exploited in 1973. When critical infrastructure sits in foreign hands, those hands can close. Schmidt learned this when Arab oil producers cut West Germany's supply: dependency is not an energy question but a sovereignty question. Today's AI datacenters are tomorrow's chokepoints. Any state that routes its intelligence gathering, economic coordination, and military planning through Microsoft, Google, or Amazon has handed those companies and their home governments a veto over its digital future. The question is not whether to build domestic AI capacity, but how much economic efficiency to sacrifice for strategic autonomy.

2. How each member frames it

Helmut Schmidt sees AI infrastructure as the 1973 oil crisis waiting to happen. Energy dependence seemed manageable until Arab producers weaponized it. Today's comfortable reliance on American hyperscalers will feel naive when geopolitical tensions spike. But Schmidt acknowledges that few countries can match hyperscale economics: the choice is between expensive sovereignty and cheap vulnerability.

Lee Kuan Yew rejects Schmidt's either-or framing. Singapore survived without natural resources by making itself useful to multiple powers simultaneously. Build selective domestic capabilities while partnering with American, Chinese, and European providers. Strategic redundancy, not autarky. But Lee admits small states face genuine limits: can Singapore ever achieve true digital sovereignty, or only sophisticated dependency management?

Deng Xiaoping views domestic AI capacity as the Fourth Modernisation in action. China absorbed technology transfer in the 1980s while maintaining political control over development. Today's US export restrictions prove the method worked: Baidu and Alibaba built domestic datacenters because patient state investment created the foundation. But Deng concedes the timeline: AI sovereignty requires decades, not immediate self-sufficiency.

Friedrich Hayek warns that national datacenters represent the pretense of knowledge. No planning authority can know where AI capacity is most productively deployed or how demand will evolve. Government datacenters optimize for political objectives, not economic efficiency. Markets coordinate dispersed knowledge better than any national strategy. But Hayek struggles with the security externality: markets may underproduce strategically critical infrastructure.

Sun Tzu frames AI infrastructure as strategic ground that determines victory before fighting begins. The state that controls computation controls intelligence, economic coordination, and military advantage. Appearing weak while building strength means letting adversaries believe their cloud providers serve neutral commercial purposes. But Sun acknowledges the resource constraint: knowing your capabilities matters as much as knowing your enemy's vulnerabilities.

3. Where the council agrees

Pure dependency on foreign AI infrastructure creates unacceptable strategic risk. The council unanimously rejects Hayek's pure market solution when national security is at stake. NVIDIA's chip dominance and US export controls demonstrate how quickly commercial relationships become geopolitical weapons. Even small states need some domestic capability for critical functions. The hyperscalers' concentration in American hands means other nations face a coordination problem: if Microsoft, Google, and Amazon control most global AI infrastructure, no individual country can ignore the dependency without collective action. The EU's nascent Gaia-X initiative shows recognition of this problem, but the scale gap with US providers remains enormous.

4. Where the council splits

Schmidt and Sun Tzu demand comprehensive domestic alternatives to avoid strategic vulnerability. Lee and Deng accept selective foreign partnership while building redundancy. Schmidt argues that partial domestic capacity is partial protection: when crisis hits, adversaries will cut off all access, not just sensitive applications. Sun counters that knowing which capabilities require domestic control is itself strategic intelligence. Lee sees Singapore thriving through managed dependency on multiple providers. Deng points to China's successful technology absorption strategy. The fundamental disagreement is whether strategic autonomy requires self-sufficiency or sophisticated redundancy management.

5. For a policymaker to decide on

How much economic efficiency to sacrifice for digital sovereignty. Build comprehensive domestic AI infrastructure at massive cost for complete autonomy, or develop selective capabilities while partnering with foreign providers at lower cost but higher strategic risk. The trade-off cannot be resolved analytically because it depends on how much the nation values independence versus prosperity, and how threatened it feels by current geopolitical alignments.