Over the past two years, governments have begun using export controls and semiconductor tariffs to shape where innovation happens, who controls it, and how it is financed. These measures are reshaping the architecture of the AI economy, from chip design to data-center siting. The ripple effects of these changes are moving through global supply chains, energy and credit markets, and the investment strategies that emerge from these market shifts.

In aggregate, restrictions on GPU exports and on the advanced manufacturing tools used to produce them are altering the economics of data-center construction and accelerating the localization of power-intensive computing hubs.

Analysts estimate that these changes will add between $75 billion and $100 billion in additional AI-infrastructure costs over the next five years, with total build-out costs as high as $3 trillion.ⁱ This capital will largely come from corporate debt issuance and infrastructure finance, creating investible assets. A wave of policy-driven investment is already being reflected in fixed-income markets, with utilities, chip foundries, and equipment manufacturers emerging as significant borrowers.

Tariff-driven cost increases are also deepening the dominance of the tech sector by a few firms. Large, well-capitalized large cloud providers (known as hyperscalers) can absorb higher prices for chips, servers, and power. They also have far greater access to capital markets to finance expansion than smaller firms, which face steeper borrowing costs and tighter liquidity conditions.

As a result, control over computational capacity is concentrating in the hands of the few companies that can finance it. Earlier periods of industrial history saw control of scarce inputs, such as steel or oil, determine competitive advantage. Today, we see a similar potential. Whoever controls processing power and energy may come to dominate the global AI race.