Morgan Stanley warns of near-term AI shock

Morgan Stanley’s internal “Intelligence Factory” model projects a net U.S. power shortfall of 9 to 18 gigawatts through 2028 — a gap Morgan Stanley quantifies as roughly 12%–25% of the electricity needed for the projected AI buildout. finance.yahoo.com The report says operators are already repurposing infrastructure to bridge that gap, converting Bitcoin-mining sites into high-performance compute centers and adding on-site generation like natural-gas turbines and fuel cells. finance.yahoo.com Morgan Stanley highlights an emerging “15‑15‑15” economics for data centers — 15‑year leases, ~15% yields and an implied ~$15 per watt of net value creation at the facility level. finance.yahoo.com On the demand side, Morgan Stanley estimates AI could deliver about $920 billion in annual net benefits to S&P 500 companies, a figure it released in earlier thematic research modeling corporate capture of AI-driven productivity gains. blog.tmcnet.com That upside translates to Morgan Stanley’s long‑run valuation lift estimate of roughly $13–$16 trillion in additional market value across the S&P 500 if firms capture those gains. finance.yahoo.com The bank points to concrete capability progress — citing OpenAI’s GPT‑5.4 “Thinking” model scoring an 83.0% on the GDPVal benchmark as evidence that frontier models are reaching expert‑level performance on economically valuable tasks. digit.in Morgan Stanley explicitly references Elon Musk’s public claim that applying ~10× more compute can roughly double a model’s “intelligence,” and the report argues observed scaling laws make that pathway plausible while warning of rapid, non‑linear capability jumps. finance.yahoo.com The research also flags the risk of recursive self‑improvement becoming material sooner than expected, quoting xAI co‑founder Jimmy Ba’s view that such loops could begin appearing as early as the first half of 2027. digit.in