Behind the AI Computing Frenzy: Looming Risk of a Millionfold Demand Collapse

The artificial intelligence boom has placed computing power on a pedestal, but beneath the surface glamour, astute observers are spotting fissures. A recent perspective from Professor Wang Jiange, Chief Researcher at Japan's NTT Data Group, has sent shockwaves through the global computing race. He posits that the industry ecosystem built around cutting-edge AI chips is experiencing a massive bubble that could burst within the next three years.

The Root of the Bubble: Lost in the "Black Box" of Data Scale

Professor Wang pinpointed a fundamental contradiction in current AI development. Today's dominant large language models operate like complex "black boxes," lacking efficient, elegant mathematical tools to describe their internal logic. This has led the industry into a cognitive trap: equating a model's competitiveness directly with its parameter count and training data volume.

"This completely disregards the real-world constraints of physics," he illustrated. "Describing a falling apple requires just three basic parameters in Newtonian mechanics. Yet, for today's large models to understand 'an apple,' they might need to be fed billions of apple images." This development path, reliant on brute-force data scaling, is inherently an inefficient form of "compute brute force."

The Disruptive Moment: New Math Could Trigger a Cliff-like Drop in Demand

The real inflection point lies in theoretical breakthrough. Wang predicts that when a revolutionary AI mathematical theory or algorithmic framework emerges—one capable of understanding and generating knowledge in a more elegant and efficient way—the current colossal demand for computing power could evaporate almost overnight.

"A millionfold reduction in computing demand is not fantasy," he stressed. Should that day arrive, the entire business model built on the current "computing is everything" logic, including those vendors relying on a single hardware architecture for outsized profits, would see its foundation shaken. The bottleneck in power supply is merely an external catalyst accelerating this process.

Navigating the Cycle: Why Memory Chips Are Seen as the "Ballast"

In contrast to the potential volatility of the computing sector, Wang directs attention to a more foundational field: data storage. Regardless of whether AI's future hardware is electronic or photonic chips, and regardless of how algorithms evolve, the need to store data remains an immutable constant.

The industrial logic of memory chips demonstrates distinct resilience:

  • Stable Demand: Data generation and preservation are continuous processes, not subject to violent disruption from a single algorithmic paradigm shift.
  • Milder Cyclicality: Unlike computing demand, which could change abruptly due to theoretical breakthroughs, the memory market exhibits a smoother, more predictable growth curve.
  • Long-Term Growth: In an era of data explosion, the long-term growth trend for storage demand is clear, allowing related manufacturers to potentially weather sharp computing cycles.

This viewpoint offers a fresh lens for investors and industry watchers: while chasing computing peaks, it may be wiser to also focus on the more stable, foundational technologies that underpin the entire digital world.