Morgan Stanley released a sweeping report on March 13 warning that a transformative AI leap will arrive in the first half of 2026, driven by an unprecedented accumulation of compute at America's top AI labs. The bank cited OpenAI's GPT-5.4 "Thinking" model, which scored 83.0% on the GDPVal benchmark — matching or exceeding human experts on economically valuable tasks — as evidence the explosion is arriving "faster than almost anyone is prepared for."
The report identifies a looming U.S. power shortfall of 9–18 gigawatts through 2028, a 12–25% deficit that is already forcing developers to convert Bitcoin mining operations into AI computing centers and deploy fuel cells. A new "15-15-15" dynamic is emerging: 15-year data center leases at 15% yields generating $15 per watt in value, reshaping energy infrastructure economics.
Perhaps most striking, Morgan Stanley warns that executives are already executing large-scale workforce reductions due to AI efficiencies, with "transformative AI" acting as a powerful deflationary force. The bank even flags the possibility of recursive AI self-improvement loops by the first half of 2027, which could accelerate disruption far beyond current projections.
