Second-mover advantage in the age of AI

The internet bubble of 1995–2001 produced a classic mismatch between investor returns and eventual economic impact. Early backers of Amazon watched the stock collapse and then languish; the net present value of those investments looked terrible for decades. Eventually, the company’s extraordinary effects on retail, cloud computing, and logistics become fully visible. Capital destroyed on the way up created the conditions for patient capital to buy undervalued assets on the way down. The second movers won.

Eerily similar events are unfolding in AI. Trillions of dollars are flooding frontier model companies and the surrounding ecosystem. Hype cycles are reliable: AI will reach the plateau of productivity only after the bubble deflates and the trough of disappointment has been crossed. When those events occur, valuations will reset. Clever investors will again find underpriced assets whose economic viability is clearer than their current market prices suggest.

The geography of those assets will matter. SemiAnalysis’s recent mapping of Chinese AI infrastructure shows a market already larger than EMEA and the rest of Asia combined, with more than 24 GW of capacity across a thousand facilities, with ByteDance alone occupying roughly a fifth of delivered space. Build-out is rapid: 100 MW facilities routinely appear in under twelve months. Policy (“Eastern Data, Western Compute”), state carriers, and hyperscalers are driving wholesale AI demand while legacy retail racks sit underutilized. Chip constraints and high vacancy rates exist, yet the physical capacity keeps expanding.

Dan Wang’s Breakneck supplies the cultural contrast. China is an engineering state whose elite, trained in civil and chemical engineering, prefers to build first and manage consequences later. The United States has become a lawyerly society whose elite, steeped in process and compliance, excels at obstruction. The difference shows up in high-speed rail timelines, power permitting, and the speed of modular data center construction. After the AI bubble pops, government influence will shape which second movers capture durable economic value. In China, engineering-oriented leadership is already directing capital toward infrastructure that can be filled once model economics stabilize. In the United States, the same leadership style that slows housing and transmission lines may also slow the reallocation of capital toward the firms that survive the trough.

The frontier labs may still produce the most advanced models. The second-mover advantage, however, is likely to accrue to those who own the undervalued capacity, the power contracts, and the policy alignment that turn models into sustained economic activity. History suggests the largest returns will go not to those who paid peak prices for the most hyped names, but to those who bought the right assets after the bubble popped.