US AI overinvestment risks a shakeout as China gains an edge on power costs
Steve Keen warned that US AI overinvestment could trigger a shakeout next year, while China’s lower power and infrastructure costs could give its providers a profitability advantage.

Steve Keen warned that heavy investment and high electricity costs could force a shakeout in the US AI industry, while cheaper infrastructure could give Chinese competitors an advantage in profitability.
The investment boom could leave US AI companies with excess capacity and services too expensive to generate sustainable returns. Keen said the resulting struggle for survival could begin next year, with Chinese providers better placed to compete on costs.
In a discussion shared on Mario Nawfal’s X account on November 16, Keen described AI investment as a technological bubble driven by innovation and excessive spending. He suggested that widespread failures among American AI companies could leave Chinese competitors as the profitable survivors.
Keen estimated that the AI boom was roughly 1.5–2 times the scale of the 19th-century railway bubble. He described a cycle in which transformative technology attracts financing, multiple businesses expand in pursuit of the same market, and capacity eventually exceeds demand. In his assessment, the questions are the timing and size of the collapse and how far its effects spread through the economy.
He estimated that spending on data centers and AI amounted to about 2.5–3% of GDP. With the major investment phase already advanced, he said a shakeout would probably follow next year. New technology can erode established businesses’ earnings while intense competition among new entrants leaves many of the investors unable to recover their spending, he argued.
Keen identified the scale of American data centers and their energy requirements as obstacles to profitability. Providers could struggle to cover those costs at prices customers consider reasonable, he said. Comparing the required charges to “Lamborghini prices,” he argued that most casual users would be unwilling to pay them.
China’s electricity network was central to his argument. Keen said its use of 800-kilovolt transmission systems, rather than 400-kilovolt systems, reduced losses and allowed electricity to reach data centers from thousands of kilometers away. He argued that the US faced higher costs and greater constraints in transmitting power over long distances, increasing pressure to locate facilities near populated areas.
Data center locations could widen that cost gap, Keen said, pointing to Chinese facilities in sparsely populated areas and underwater. He suggested that a lower cost base could let Chinese providers charge less while remaining profitable. His illustrative comparison was Chinese AI offering 90% of the functionality at 10% of the cost, with the additional performance of American products insufficient to justify the premium for many users.
Keen linked the infrastructure gap to different incentives for investment and maintenance. Private companies focused on quarterly earnings and share valuations have reasons to defer maintenance that reduces near-term profits, he argued. State enterprises face less pressure to generate immediate returns, which he said could favor the upkeep of long-lived electricity and water systems.
He described China’s approach as a combination of state-led infrastructure investment, competition among private businesses and experimentation across regions. In his account, public investment supplies transport, communications and power, while companies and provinces compete in innovative industries. Keen argued that this combination could translate into lower AI service costs and a stronger position for Chinese providers seeking profitability.