Inside Japan’s underground AI experiment to save a shrinking workforce

Deep beneath Tokyo’s Otemachi business district, a quiet experiment is underway. Marunouchi Heat Supply Company, which runs a 30-kilometer network of heating and cooling pipes serving offices and transit hubs, has begun using artificial intelligence to manage its infrastructure. The AI system, called PlantPilot, was built with Tokyo startup Preferred Networks and trained on decades of operational data and the know-how of veteran engineers. The company aims to move toward fully automated operations by 2027.

For Japan, the world’s oldest society, this isn’t just a tech upgrade—it’s a survival strategy. With a shrinking workforce and experienced engineers retiring, companies face a looming loss of institutional knowledge. “Our biggest challenge is securing skilled personnel,” said Hidetaka Yagi, assistant general manager at Marunouchi. “AI can help preserve that expertise and pass it on to the next generation.” Unlike fears elsewhere that automation will replace jobs, Japan is treating AI as a partner—a way to extend human capabilities, not eliminate them.

But Japan’s AI ambitions extend beyond fixing old systems. The country has fallen behind the U.S. and China in the global tech race, with only about eight unicorn startups compared to America’s 690 and China’s 160. Tokyo-based Sakana AI is trying to change that. Chairman Ren Ito argues Japan shouldn’t copy Silicon Valley. “If we try to compete with American companies in exactly the same way, we will not succeed,” he said. Instead, he believes Japan’s edge lies in applying AI to practical problems in energy, manufacturing, and healthcare. The government has pledged 370 trillion yen ($2.3 trillion) by 2040 for AI, semiconductors, and research. Yet experts warn that cash alone won’t close the gap—Japan also needs to rebuild its universities and attract top talent. The real test is whether a nation built on precision manufacturing can reinvent itself as an AI problem-solver.