AI reliability gap could trigger a sharper tech stock correction, Levin warns

John Clark Levin said the gap between AI’s peak capabilities and its reliability in business could trigger a deeper, sharper technology stock correction than the recent pullback.

AI demonstrations have helped drive market expectations, but companies’ ability to deploy the technology reliably remains a constraint. If adoption fails to keep pace with those expectations, investors could face a more pronounced sell-off.

In a David Lin interview recorded on November 5, Levin, head of research at Kurzweil Technologies, said the recent selling did not appear to be the correction he had previously anticipated. He expected that eventual decline to be deeper and more marked by panic, describing the current retreat as closer to temporary market jitters than a change in the underlying progress of AI.

Levin said speculation responds to the best performance AI can demonstrate, while adoption across the economy depends on what it can do consistently enough to operate autonomously. By some measures, reliability has improved roughly 1,000-fold over the past two and a half years, he said, but it is still catching up with peak capabilities.

He expected that gap to persist for roughly another year and questioned how long markets would tolerate it. Before AI reaches the reliability needed to work autonomously near the threshold of artificial general intelligence, or AGI, companies face substantial fixed costs to redesign workflows. They must also retrain employees to supervise AI and check for fabricated outputs, he said.

Levin expected demand for data centers to remain resilient to political changes. He said the midterm elections were unlikely to have a major effect on construction across the United States because demand for AI training and inference capacity rests on the value of intelligence itself. The locations of projects could shift without materially changing that underlying demand, he said.

Requirements to protect local communities from higher electricity bills could increase, Levin said. Developers might have to pay for guarantees that utility prices would remain at least flat. He argued that such measures could benefit residents while accounting for only a small share of next-generation AI model development budgets, leaving the competitiveness of the US AI industry intact.

For Levin, the central risk behind a sharper correction is the time companies need to turn AI’s demonstrated abilities into dependable business performance. Continued technical progress could coexist with investor disappointment if reliability takes too long to catch up.