Morgan Stanley: Nearly Half of Global Power Super-Cycle Demand Remains Unpriced
nashnova research
Morgan Stanley raised its 2025–2030 global power demand forecast by 10%, arguing that roughly half the demand driven by AI agents, industrial reshoring, and air conditioning has yet to show up in generator and gas-midstream valuations.
Why does Morgan Stanley say the market has only priced half?
The bank projects global power consumption at roughly 36.3 trillion kWh by 2030 — about 60 basis points above the IEA forecast and roughly 15% faster than the market consensus.
This means → Morgan Stanley sees a wide gap between where power demand is heading and where generator, gas-pipeline, and coal-producer stocks are priced — current valuations reflect only about one-third of the bank's base case.
In plain terms = not enough power plants are being built, prices haven't risen enough, yet the companies that benefit most directly haven't re-rated.
What are the "Three As" — and why do they push power demand higher?
Morgan Stanley groups three forces under the label "Three As": AI Agents (Agentic), Air Conditioning (Aircons), and Industrial Reshoring (Allshoring).
AI agents consume 5–30× more tokens — the basic unit of AI computation — than a standard chatbot query. A complex enterprise workflow may chain 5–7 agent instances, pushing end-to-end token use to 375,000–525,000 per task. This means → falling token prices do not cut total power use; wider adoption and harder tasks keep pushing absolute compute — and electricity — higher.
Industrial power is forecast to reach 13.1 trillion kWh by 2030, roughly twice the data-center total. Space cooling is projected to rise from 2.9 to 4.1 trillion kWh at a ~7% CAGR — the cooling increment alone, excluding data centers, may match the data-center increment itself.
Why is pricing power shifting from equipment makers to plant and grid owners?
After two years of equipment-cost inflation, consumers and policymakers are pushing back on high prices; equipment makers' pricing leverage is peaking. Generators and grid operators, by contrast, are gaining room to raise rates.
Morgan Stanley expects commercial power spreads to widen roughly 15% by 2028. Asian wholesale prices have already risen about 20% over the past year, and the bank forecasts ~18–20% global power-price growth year-on-year in 2026.
This means → the real sustained earnings power in the supercycle is moving from "those who sell the shovels" to "those who own the mine" — generators and pipelines have more profit leverage than equipment vendors.
Do AI hyperscalers really not care about power prices?
Morgan Stanley modeled a 100 MW fully integrated AI hyperscale project: when power costs rise from $120/MWh to $140/MWh (about a 17% increase), project returns fall by only about 60 basis points, with ROE still well above 30%.
Key project numbers: revenue $1.93 billion, power cost just $116 million, net margin 37%, ROE 33.6%.
In plain terms = electricity is a tiny fraction of an AI hyperscale project's cost structure. A 20% jump in power prices barely dents profits — so hyperscalers can absorb higher rates, which in turn supports generators' pricing power.
Why is Asia the epicenter of this supercycle?
Asia (including China) is expected to account for roughly 64% of incremental global power demand from 2025 to 2030. Its increment is 4× the U.S. figure; its growth rate is 2.5× Europe's. Country-level CAGRs: India 6.5%, China 5.2%, ASEAN 4.9%, U.S. 4.7%.
The U.S. faces a 43 GW data-center power shortfall by 2028 — about 38% of data-center demand. Morgan Stanley expects roughly 15 GW of that gap to spill over into Asia: Malaysia ~4 GW, Japan ~1.5 GW, other ASEAN ~4 GW.
This reflects a structural dynamic: the U.S. cannot build power fast enough, so demand migrates to Asia — making Asian power companies passive beneficiaries of America's supply bottleneck.
Who did Morgan Stanley pick — and who did it avoid?
The bank identified 36 "Powering AI" winners across generators, gas midstream, grids, and storage. Its top picks carry roughly 14% EPS upside to 2028, implying about $500 billion in market-cap creation.
Top picks include: in Europe, EDP and RWE; in Asia, Adani Power, Yangtze Power, Samsung SDI, and Hokkaido Electric; in the U.S., Bloom Energy and Solaris Energy.
Names to avoid: KEPCO (debt reaching KRW 21 trillion by Q2 2026; grid-financing constraints), Tata Power, Huaneng Power, and others. This means → Morgan Stanley's framework is clear: pick companies with pricing power and capacity to expand; avoid those weighed down by debt or policy headwinds.
市场有风险，内容仅供研究参考，不构成投资建议。