Morgan Stanley: The AI compute gap is widening, and five major directions are emerging

Deep Tide TechFlow message, citing tide research, a global thematic strategy report from Morgan Stanley dated July 27 states that the recent pullback in the AI infrastructure sector is driven by sentiment rather than a fundamental turning point. The AI compute capacity supply-and-demand gap will continue to widen, and this round of adjustment is an allocation window.

The report lays out five key investment themes: AI infrastructure bottlenecks (Talen Energy, Vistra, Bloom Energy, Hut 8, and other power-supply enclosure providers), the compute manufacturing ecosystem (NVIDIA, Broadcom, Micron), China’s AI solutions (ByteDance, Alibaba, Tencent accelerating capital expenditures; the expected AI chip self-sufficiency rate is projected to rise from 42% to 70%), energy security (Korean power equipment manufacturers benefiting from the US shortage of transformers, including HD Hyundai Electric and LS Electric), and hyperscale vendors (Meta, Google, Microsoft, Amazon, with the ROI of AI capital expenditures being underestimated).

Google plans to expand compute capacity by 1,000 times over 4 to 5 years; over the same period, NVIDIA’s shipment CAGR is only 140%, making it difficult to close the “scissors gap” in the short term. The average monthly consumption per enterprise Token is only $11, while AI upgrade savings per project are $55, with ROI exceeding 20 times. The budget ceiling adjustments are phased and not a demand inflection point. The US datacenter power shortfall from 2026 to 2028 will reach 38GW, and 14 states across the US are developing control bills. Morgan Stanley recommends adding positions in batches when the thematic pullback is at 5% and 15%.