The power super-cycle is about more than just AI! Morgan Stanley: The "Three A" forces are pushing the global power shortage beyond expectations, and about half of market demand is not yet priced in.

Morgan Stanley recently published an in-depth global power industry report titled "Powering AI: Agents, Aircons, Allshoring: All Not Priced In," putting forward a judgment that differs from the mainstream market narrative: the power consumption upgrade dominated by data centers and semiconductor fabs over the past three years is spreading to the industrial and residential sectors, and the "Three A" forcesAI agents (Agentic), air conditioners (Aircons), and reshoring/supply chain restructuring (Allshoring)will push global power tightness beyond expectations. The bank raised its 20252030 global power consumption growth forecast by 10%, projecting global electricity consumption of about 36.3 trillion kWh by 2030, about 60 basis points above the IEA forecast, and said it is about 15% faster than market expectations. Morgan Stanley's core conclusion is that the power super-cycle is not driven by AI alone; industrial reshoring, manufacturing expansion, semiconductor localization, and space cooling demand are jointly raising baseload and peak load. The market is currently clearly underpricing generators, natural gas midstream pipelines, grid operators, and coal producers, while growth expectations for the upstream power supply chain and renewables have already been priced in to a greater extent. Morgan Stanley estimates that about half of the power demand brought by AI agents, the Asian industrial super-cycle, and new air conditioning/heat pumps is not yet reflected in the valuations of power generation companies, natural gas midstream, or coal producers. From "Powering AI" to "Powering Everything" Morgan Stanley views AI agents as the first incremental driver of power demand. Morgan Stanley's US internet team expects agents to form a US$30 trillion per year market by 2029, while the token volume consumed by AI agents is 5 to 30 times that of traditional chat Siasun Robot&Automation; complex enterprise workflows may deploy 5 to 7 agent instances, each consuming about 15 times the tokens of a traditional query, with end-to-end task token consumption reaching 375,000 to 525,000. The report says agent workloads already account for about 64% of enterprise output tokens, and falling token prices have not reduced total power demand; instead, broader adoption and more complex tasks have pushed up absolute computing power and electricity consumption. But the bank stresses that AI is not everything. Industrial sector power consumption will rise to about twice that of data centers by 2030; Asia's industrial sector will account for about 43% of the region's incremental power consumption, about 2.7 times the contribution of data centers, with China and India contributing 97% of global incremental industrial demand. Global industrial power demand is expected to rise from 11.1 trillion units in 2025 to 13.1 trillion units in 2030, with Asia accounting for 83%. The report regards steel, autos, semiconductors, defense, refining, chemicals, and related industries as the main vehicles of "Allshoring," expected to add about 2 trillion units of power consumption by 2030, a compound growth rate of about 3.43% in 20252030, about 1.5 times faster than the past decade. Space cooling is the third incremental driver. Global cooling power demand is expected to rise from 2.9 trillion units in 2025 to 4.1 trillion units in 2030, a compound growth rate of about 7%, higher than the 4.2% growth rate of global power demand; cooling accounts for about 10% of global power consumption but about 30% of peak power demand. In India, cooling accounts for as much as 40% of peak power demand under El Nio conditions in 2026. The bank expects Asia's installed air conditioner stock to rise from about 1.1 billion units to about 1.4 billion units by 2030, with Asia's cooling electricity use rising from about 1.5 trillion units to 2.1 trillion units; India's cooling demand is expected to nearly double to about 280 billion units by 2030. European heat pump electricity use is expected to rise from 113TWh in 2025 to 179TWh in 2030, with its share of European power demand rising from 3.8% to 5.4%. Morgan Stanley believes that incremental cooling demand excluding data centers may be as large as the incremental power demand from data processing centers. Pricing power shifts from equipment makers to infrastructure owners Morgan Stanley believes a key shift is taking place in pricing power in the power market. After equipment cost inflation over the past two years, generators and grid operators are beginning to gain stronger pricing power, while equipment suppliers face resistance to high prices from consumers and policymakers. Morgan Stanley expects commercial power spreads to widen by an average of about 15% by 2028; Asian wholesale power prices have already risen about 20% over the past year, with generator spreads widening 2 to 3 times in most regions; global power prices are expected to rise about 18% to 20% year on year in 2026. The report says generators can counter high equipment prices by refurbishing existing equipment and delaying new capacity deployment, while announcements of turbine and transformer capacity expansion are expected to ease supply chain concerns. The key basis for this judgment is that AI hyperscalers are not sensitive to power prices. Morgan Stanley's analysis of a 100MW fully integrated AI hyperscale project (with its own power shell, GPUs, and LLMs) shows that when the power price rises from US$120/MWh to US$140/MWh, an increase of about 17%, project returns are affected by only about 60 basis points, with ROE still well above 30%; even if the power price rises to about US$160/MWh, a 1.5-fold increase, ROE remains high. In the 100MW project calculation, revenue is US$1.93 billion, power cost is US$116 million, EBITDA is US$1.802 billion, net profit is US$714 million, net margin is 37%, and ROE is 33.6%. On this basis, the bank believes power utilities have the ability to implement differentiated pricing for data centers to fund grid and generation capacity expansion and enhance returns over the long term. However, Morgan Stanley also warns that the power system is facing a steeper load curve. India's duck curve is being reshaped into a double-hump curve, Thailand's peak has shifted to the evening due to EV charging, Australia's deep midday trough highlights the value of storage and flexible generation, and Japan's nuclear restarts have reduced the grid's flexibility to absorb renewable fluctuations. Asia is expected to need about US$1 trillion of grid investment by 2030, with data centers bearing part of the grid upgrade cost through higher power prices, while tiered pricing is used to cross-subsidize residential consumers. Asia is the center of gravity, US shortfall spills over At the regional level, Morgan Stanley sees Asia as the center of the global power super-cycle. Asia (including China) is expected to account for about 64% of global incremental power demand in 20252030; Asia will consume about 20 trillion units of electricity by 2030, more than half of the global total; Asia's incremental power consumption is expected to be 4 times that of the US, with consumption growth 2.5 times that of Europe plus the UK and 1.2 times that of the US. The report expects Asia's power consumption compound growth rate of about 5% in 20252030, about 140 basis points faster than the current average of most Asian economies. By country, India's power demand compound growth rate is expected at 6.5% in 20252030, China 5.2%, ASEAN 4.9%, the US 4.7%, Australia 3.5%, Latin America 3.4%, South Korea 3.0%, Europe plus the UK 2.7%, and Japan 0.9%. Japan's power demand will reverse a two-decade decline, but inflexible nuclear baseload led the Tokyo metropolitan area to curtail as much as 4.41GW of local CECEP Solar Energy in MarchApril 2026 while importing power from the Tohoku region. South Korea's data center power consumption is expected to rise from about 8TWh in 2025 to about 30TWh in 2038, a compound growth rate of about 10%, far above the roughly 2% growth rate of overall power demand; however, Korea Electric Power Corporation Sponsored ADR's (KEPCO) debt reached KRW21 trillion in the second quarter of 2026, and Samsung Electronics and SK Hynix (SKHY.US) reportedly rejected KEPCO's proposal for five-year prepayment totaling KRW25 trillion in electricity fees to fund grid investment, and grid financing constraints may limit the speed at which South Korea can accommodate AI load. For the US, the bank estimates that US data centers will face a 43GW power shortfall by 2028 and a cumulative shortfall of 80GW by 2029; even including "time-to-power" solutions such as behind-the-meter generation, fuel cells, nuclear power, and site conversions, the net shortfall in 2028 would still be about 43GW, equivalent to 38% of US data center power demand. The American Electric Power Company, Inc. team expects US data centers to need about 116GW of additional power in 20262028; after considering projects under construction, available grid capacity, and additional behind-the-meter turbines, Bloom Energy (BE.US) fuel cells, nuclear site data centers, and other solutions, the gap still reaches 70GW, and after including about 6GW of Asia-Pacific spillover, the net gap remains 56 to 38GW. Morgan Stanley expects about 15GW of US data center shortfall to spill over to Asia, including about 4GW in Malaysia, about 1.5GW in Japan, about 4GW in the rest of ASEAN, about 2GW in Australia, and about 3.6GW in India and elsewhere; Chinese hyperscalers plan to add 2 to 3GW of capacity outside China each year. Thailand has received about US$25 billion in data center investment commitments, and Malaysia has approved about US$55 billion of data center projects since 2025, with 57 of 84 projects requiring about 6.2GW of power, all supplied by Tenaga. Natural gas is seen as a key bridge connecting AI and power security. The bank expects Asia (excluding China) to add about 19GW of natural gas generation capacity in 20252030, underwriting about 20 million tons/year of LNG imports, equivalent to one-fifth of new US shale exports; globally, 20252030 will see 252GW of new coal power and 234GW of gas power; Asia will add about 303GW of coal power, 61GW of gas power, 51GW of nuclear power, more than 2TW of renewables, and about 3TWh of storage. The report believes natural gas, renewables, and storage are increasingly cooperative rather than competitive, and earnings for natural gas midstream pipelines, LNG importers, and related infrastructure should rise, but global conflicts have kept LNG costs high and the market is not yet ready to price this change. For Europe, Morgan Stanley expects European power demand to grow 1.7% in 2026 and at a compound rate of 2.2% in 20262030, the first sustained multi-year growth since 2014; data centers will contribute about 90TWh of incremental demand in 20252030, accounting for 25% to 30% of European power demand growth, with data centers' share of European power consumption rising from nearly 3% in 2026 to about 5% in 2030 and about 8% in 2035. The report estimates that data center demand could push European power prices up by about EUR2/MWh by 2030, by more than EUR3/MWh in the UK, and by more than EUR4/MWh across Europe in a bull market. Europe's electrification rate was only 21% in 2024, rising only 0.1 percentage point per year on average since 2010; if it rises to 28% to 30% by 2035 under the IEA path, the European power market would need about 1,400TWh more over the next decade and about 900GW of additional generation capacity, much of it from CECEP Solar Energy. Latin America's power demand is expected to grow at a compound rate of 3% to 4% in 20252030, with data centers adding about 1TGW of demand by 2030 in a bull scenario (about 7.9GW in Brazil), equivalent to about 4% of total power demand and contributing about 80 basis points to overall growth, but the traditional economy, industry, and electrification remain the main drivers. Data center electricity use in Australia's National Electricity Market is expected to rise from about 5TWh in 20252026 to 34TWh in 20352036, with its share of grid supply rising from about 3% to about 13%; Australian battery deployment is nearing saturation in the commercial market, with BESS commissioning of 3.7GW/9.8GWh and household batteries of 6GW/1GWh year-to-date in 2026, and battery power capacity accounting for about 14% of total system capacity of 117GW, which may suppress future wholesale power prices. Investment strategy: 36 winners and an avoid list Morgan Stanley screened 36 "Powering AI" winners from the global power value chain, covering reliable generators, natural gas midstream pipelines, grid companies, SES AI Corporation Class A, and storage suppliers, but remains selective on equipment suppliers, citing the shift in pricing power toward infrastructure owners. The bank believes equity valuations currently price in only one-third of its base-case power consumption growth, close to mid-cycle consumption levels of the past decade; its top picks have about 14% upside to 2028 EPS, corresponding to about US$0.5 trillion of market value creation. Morgan Stanley's top winners include: Europe's Energias de Portugal (EDP), Germany's RWE; Asia's Gulf Development, Keppel, Adani Power, China Yangtze Power (Yangtze Power), Samsung SDI, Hokkaido Electric; US's Bloom Energy, Solaris Energy (SEI.US). The report also lists names to avoid: Asia's KEPCO, Tata Power, Global Power Synergy, Huaneng Power (Huangen Power), Electric Power Development; Europe's Enagas, SNAM. On the specific logic, about 35% to 40% of EDP's 2027 EBITDA is exposed to Portugal, one of Europe's fastest-growing power demand markets, with demand expected to grow at a compound rate of 4.5% in 20252030, and EDP has signed memoranda of understanding with Start Campus and Merlin; RWE is positioned to supply European data centers through a combination of renewables and flexible capacity, and has about 3GW of available capacity (10 sites of 300MW) for powered land monetization. In the US, Bloom Energy is seen as a key beneficiary of the "time-to-power" shortage, with Morgan Stanley expecting 6GW of fuel cell deployment in 20262028, but recent site expansion suggests upside; Solaris Energy is also listed as a US top pick. In Asia, Gulf Development, Adani Power, Sembcorp, and others are providing behind-the-meter power to data centers, China Yangtze Power is viewed as a high-quality defensive hydro utility, and Samsung SDI benefits from demand for ESS, BBU, and UPS from data center backup power and grid-scale storage.