Why Nations Suddenly Care

For most of the cloud era, no serious government treated computing capacity as a strategic resource on par with oil or grain. Compute was a commodity you rented from American hyperscalers, cheap and abundant, and the only real policy questions concerned data privacy and where records were stored. That world ended over roughly three years, and understanding why is essential to understanding the scramble now underway.

Three forces converged. The first was the demonstration, beginning in late 2022, that large-scale AI models were not a laboratory curiosity but a general-purpose technology with economic and military consequences. Once governments accepted that frontier AI capability tracks closely with access to enormous quantities of advanced compute, compute stopped being a commodity input and became a strategic one. If the models that will shape productivity, cyber capability, and eventually weapons are gated by access to accelerators, then a nation's chip supply is a national-security variable, not a procurement line item.

The second force was scarcity. The specific chips required to train frontier models — for several years overwhelmingly Nvidia's — became the tightest supply constraint in the technology economy. When a resource is both strategically vital and physically scarce, it acquires the political characteristics of a strategic commodity: hoarding, allocation battles, and government intervention. Ministers who a decade earlier could not have named a GPU began tracking delivery schedules for specific accelerator models.

The third and most consequential force was policy: the United States, beginning in October 2022 and tightening repeatedly thereafter, imposed export controls restricting the sale of the most advanced AI chips and the equipment to make them, initially aimed at China but expanding into a tiered global framework. This single lever converted compute from a market good into a geopolitical instrument. Suddenly, access to the best silicon depended not only on your ability to pay but on which side of a widening set of American rules you sat. That is the thesis this course returns to repeatedly: export controls reshape the global compute map. They do not merely slow one rival; they force every nation to reassess whether its compute future rests on a foreign approval it cannot guarantee.

Once you internalize those three forces, the wave of national programs makes sense. A country that believes AI is strategically decisive, that the key input is scarce, and that access is contingent on another government's permission faces an obvious conclusion: dependency is a vulnerability, and reducing it is a matter of state. Hence the sudden proliferation of sovereign AI initiatives, national compute funds, and state-backed data-center megaprojects that we will examine in detail.

But the motives are not identical across nations, and conflating them is a common analytical error. For some states the driver is genuine supply-denial fear — they worry they will be cut off. For others, particularly those firmly inside the American security umbrella, the driver is economic: capturing the data-center construction boom, the jobs, the tax base, and a role in the AI value chain. For still others it is prestige and geopolitical positioning — a sovereign AI program is a way to be taken seriously as a technological power and to bargain for better access. Most real programs blend all three, but the blend differs, and the blend predicts behavior. A supply-denial motive pushes toward expensive domestic fabrication and chip stockpiling. An economic-development motive pushes toward attracting foreign operators and importing chips freely. A prestige motive pushes toward visible megaprojects and headline capital commitments.

There is also a sharpening asymmetry worth naming early. The capital, the leading-edge chips, and the operational know-how are concentrated in a small number of firms and countries. That means most nations pursuing 'sovereignty' are in fact pursuing a negotiated dependency: a deal with a chip designer, a hyperscaler, or the United States government that gives them enough access and enough local control to satisfy their particular blend of motives. The genuinely self-sufficient path — designing your own chips, building your own fabs, writing your own stack — is open to almost no one on a relevant timescale. So the realistic strategic question for nearly every government is not 'how do we become independent?' but 'whom do we depend on, on what terms, and how do we make that dependence durable and politically tolerable?' The programs we study next are best read as different answers to exactly that question.