Fiscal commitments can also stimulate overinvestment in AI-related sectors, contributing to macroeconomic instability through boom-bust cycles. Episodes from history suggest that booms in technology are highly susceptible to such cycles, even in advanced economies (refer to box 6.1). Unanticipated application of state power deters investment, while cozy arrangements between governments and firms could lead to rent seeking and weakened institutions When firms cannot anticipate how state power will be applied, they factor this uncertainty into their choices by seeking higher returns, shortening investment commitments, or opting out entirely. Crucially, the fear is not regulation itself, given that firms routinely invest in heavily regulated environments, but rather sudden rule changes. The risk of unanticipated policy is compounded because much of the AI infrastructure, such as data centers and chip fabrication plants, requires large investments tied to specific locations. Once the investment is completed and the capital becomes sunk, the firm’s ability to relocate declines. This shifts greater bargaining power to the host government, in a dynamic known as the “obsolescing bargain.”35 Governments can now raise preferential tax rates agreed upon before the investment was made, renegotiate access to land or power, introduce data localization requirements, and in extreme cases, expropriate the assets. Thus, reflecting the political risk, firms may make investments that have shorter commitments, limit transfers of capabilities, and create less local value. Box 6.1 Boom, bang, bust in technology investments: Does history repeat itself? Developments in artificial intelligence (AI) in the United States share certain characteristics with historical episodes like the railroads boom in the 1800s and the telecommunications boom of the late 1990s, in which the early promises of technological breakthroughs generated waves of exuberant expectations, massive leveraged investment, and eventual financial distress. Importantly, these episodes did not reflect technological failure but overinvestment based on overly optimistic forecasts. Although demand eventually caught up, the adjustment path was costly. Understanding how AI developments play out among firms in the United States matters because they are one of the dominant builders of the global AI infrastructure, including in developing countries. Whether a crash happens depends crucially on whether demand matches the high levels of investment. High expectations of rapid growth. Past technology booms, like today’s AI wave, have been marked by expectations of rapid growth in demand. A 1998 report by the US Department of Commerce claimed that internet traffic was doubling every 100 days—a ­figure later credited to WorldCom, a leading telecommunications company at the time.a This expectation underpinned a massive build-out of network infrastructure, even as growth rates slowed in subsequent years.b Similar expectations can be observed today and have been used to justify (Box continues next page) AI’s Political Impact: Reshaping Power Within and Across Countries 209