Morgan Stanley expects market-based power generation spreads to widen by an average of about 15% by 2028, with its preferred companies' 2028 earnings per share having approximately 14% upside versus consensus estimates.
Over the past two years, AI power investment focus has centered primarily on data centers: more chips, higher server power, and whoever can supply transformers, gas turbines, and cooling equipment stands to capture new orders.
In its global power research published on October 26, Morgan Stanley further broadened this framework. The expansion of AI agents, air conditioning, and manufacturing across different regions is simultaneously increasing electricity demand and reshaping intraday consumption patterns. Power market constraints thus extend to stable supply, peak-shaving capacity, and grid interconnection speed.
The bank estimates that since 2024, "powering AI" has generated approximately $1 trillion in market capitalization growth for related supply chains. In the next phase, broader electrification demand could create another approximately $500 billion in market capitalization by 2027, with beneficiaries increasingly distributed across power generation, natural gas pipelines, power grids, and integrated energy sectors.
36 trillion kWh of electricity demand, with increments beyond AI expanding
Morgan Stanley has further raised its global electricity demand growth forecast by about 10%, projecting a compound annual growth rate of 4.2% from 2025 to 2030, approximately 0.6 percentage points higher than the International Energy Agency's forecast. On a basis that includes behind-the-meter consumption, global electricity consumption will reach approximately 36 trillion kWh by 2030.
"Behind-the-meter" refers to power generation or consumption facilities located on the user's side of the electricity meter, such as on-site generation systems configured for factories and data centers. This demand may not be fully reflected in traditional public grid supply statistics, yet it equally requires fuel, equipment, and construction investment.
This round of growth has a broader geographical base. China and India continue to expand from a larger electricity consumption base, while the United States, Europe, Japan, and Australia are seeing more pronounced demand recovery compared to the past decade, driven by industrial investment, manufacturing reshoring, and electrification. The bank estimates that from 2025 to 2030, Asia will contribute approximately 64% of global incremental electricity demand.
Industry remains the largest pillar. Morgan Stanley estimates that sectors including steel, automotive, semiconductors, defense, refining, and chemicals will add approximately 2 trillion kWh of annual electricity demand by 2030 compared to 2025. Asian industrial electricity consumption is expected to grow at a compound annual growth rate of about 4%, contributing approximately 43% of the region's incremental electricity demand—about 2.7 times the contribution from data centers.
The report calls this force "Allshoring," encompassing supply chain reshoring, localization, and cross-regional relocation. Wherever factories land, the associated electricity load forms there; new semiconductor, chemical, or metal processing projects, even if they do not significantly change national electricity consumption scale, may still place considerable pressure on local grids.
Air conditioning is another easily underestimated increment. The bank estimates that global space cooling electricity consumption will rise from approximately 2.9 trillion kWh in 2025 to approximately 4.1 trillion kWh in 2030, with a compound annual growth rate of about 7%—faster than overall electricity demand. The scale of incremental cooling electricity consumption is comparable to that of data centers.
Asia's air conditioning stock is expected to increase from approximately 1.1 billion units currently to 1.4 billion units by 2030. Combined with extended usage hours, cooling electricity consumption will rise from approximately 1.5 trillion kWh in 2025 to approximately 2.1 trillion kWh, contributing approximately 15% of Asia's incremental electricity demand over the same period.
Pressure from cooling is also concentrated during peak hours. Under hot weather conditions, large numbers of air conditioners running simultaneously require the grid to prepare sufficient peak supply capacity. Sufficient annual generation does not mean stable supply during the hottest few hours, which simultaneously increases investment demand for peaking power sources, energy storage, and transmission and distribution.
The more complex the agent, the more valuable power access becomes
Improved AI model efficiency does not necessarily lead to lower total electricity consumption.
Traditional chatbots may complete a response through a single primary model call, while agents may repeatedly reason, retrieve information, invoke tools, execute code, and then verify results. Even as the price per Token—the basic unit of text processed by models—continues to decline, the number of Tokens consumed and model calls required to complete a task may still rise.
Morgan Stanley uses a hypothetical scenario to illustrate this shift: if a complex enterprise workflow deploys 5–7 agent instances, each consuming about 15 times the tokens of a typical 5,000-token query, the total consumption to complete an end-to-end task could reach 375,000–525,000 tokens. This is an example of task complexity, not the average consumption of all agent applications.
As a result, model price cuts, rising adoption, and increasing task complexity may occur simultaneously. Cheaper inference expands the scope of applications, while more complex workflows in turn drive up compute consumption. Power demand depends on the combined effect of these factors.
The speed of power supply has thus become an important variable in compute expansion. Morgan Stanley estimates that by 2028, new U.S. data center power demand will be about 116GW. After accounting for projects under construction, available grid capacity, and accelerated supply solutions such as on-site generation, fuel cells, and nuclear power, there may still be a gap of about 43GW, equivalent to about 38% of related demand.
This figure uses a basis that has not yet deducted demand migration to Asia-Pacific. The U.S. team separately estimates that after about 6GW of demand shifts to Asia-Pacific in 2026–2028, the gap would still be about 38GW. The approximately 15GW hosting opportunity proposed by the Asia team in the report corresponds to a longer-term window through 2030, and the two cannot be directly subtracted.
The logic of demand migration is to deploy some compute in regions with better power supply conditions and then provide services through the network. Markets such as Thailand, Malaysia, India, and Japan are therefore gaining new data center investment opportunities. Power resources, grid access, and delivery timelines are jointly influencing compute site selection.
However, overseas expansion is also constrained by local grids, construction approvals, delivery capabilities, and customer requirements, making it difficult to fully absorb the U.S. gap. Morgan Stanley remains bullish on on-site power companies that can provide available power in 2027–2028, including suppliers of fuel cell and gas-fired generation solutions.
"Bring Your Own Power" is also spreading in Asia. The bank expects that by 2030, Asia behind-the-meter power demand will reach about 8GW. For data centers, on-site generation can reduce dependence on public grid interconnection timelines; for power producers, directly serving large customers creates new contracts and revenue models.
Pricing power shifts to stable power sources, grids, and gas supply
Who ultimately profits from electricity growth depends on the scarcest resources in the power system.
Wind and solar will remain important sources of new generation, but their output varies with weather and time of day. Nuclear power can provide stable electricity but typically lacks the flexibility to quickly adjust output. As data centers running around the clock, industrial loads, and evening cooling demand grow together, the system's requirements for dispatchable power sources, energy storage, and grids will also increase.
Morgan Stanley expects Asia will need approximately $1 trillion in grid investment by 2030 to meet new electricity demand and improve system flexibility. Grid access rights, land already equipped for power supply, and stable power sources that can be brought online quickly are therefore rising in value.
Natural gas plays a bridging role. The bank estimates that from 2025 to 2030, Asia excluding China will add approximately 19GW of gas-fired power generation capacity, supporting annual LNG import demand of about 20 million tonnes. Related opportunities will extend from power stations to gas pipelines, LNG terminals, and other supporting facilities.
Energy storage will also expand in application, but its business models need to be differentiated. Batteries can charge during low-price periods and discharge during high-price periods, alleviating curtailment of wind and solar power and evening peak pressure. However, as deployment scales up, peak-valley price spreads may narrow, and returns relying solely on electricity price arbitrage will decline. Capacity compensation and ancillary service revenue will therefore become more important—namely, batteries being paid for providing backup supply capacity, frequency regulation, and other services.
Potential gains for power generators come from the spread between electricity selling prices and fuel costs. Morgan Stanley expects that by 2028, market-based generation spreads will widen by an average of about 15%. If increased LNG supply drives fuel costs lower while electricity supply and demand remain tight, some generators' profit margins could expand.
Data centers' emphasis on power supply timing may also enhance power companies' bargaining power. In a calculation for a 100MW integrated AI operator with data center facilities, GPUs, and large model businesses, the bank estimates that electricity prices rising from $120 per megawatt-hour to $140 per megawatt-hour—an increase of about 17%—would impact return on equity by only about 0.6 percentage points.
This result depends on model assumptions and cannot be directly generalized to all data centers, but it illustrates that for high-investment AI projects, securing power early and getting chips running may have more economic value than negotiating lower unit electricity prices.
At the same time, rising residential electricity bills will increase policy resistance. Setting differentiated electricity pricing for data centers and having new large customers bear more grid expansion costs may become ways to balance investment and residential burden. Whether power companies can convert new demand into revenue still depends on whether regulators allow them to recover costs.
$500 billion opportunity hinges on whether profits materialize
Morgan Stanley expects that its preferred companies' 2028 earnings per share have approximately 14% upside versus market consensus, and related earnings growth and valuation changes could support approximately $500 billion in new market capitalization. This is an institutional scenario assessment for a group of companies, not a unified earnings forecast for the entire power industry.
The bank's preferred directions include stable power sources, power grids, natural gas infrastructure, and on-site power generation and energy storage companies that can directly address power supply bottlenecks. Companies highlighted in the report include EDP and RWE in Europe, Gulf Development, Keppel, Adani Power, China Yangtze Power, Samsung SDI, and Hokkaido Electric Power in Asia, as well as Bloom Energy and Solaris Energy Infrastructure in the United States.
Greater divergence may emerge within equipment suppliers. Over the past two years, equipment shortages brought price increases and order expansion; as capacity for gas turbines, transformers, and other equipment increases, power generation and grid companies may control spending by refurbishing old equipment and adjusting construction schedules. Morgan Stanley therefore places greater emphasis on companies with infrastructure and customer resources, and adopts a selective approach to the equipment segment.
Power demand growth does not guarantee synchronized profit growth for utility companies. The report maintains a cautious view on Korea Electric Power, citing fuel cost pressures, limited room for electricity price adjustments, and high leverage and grid investment needs. New load can bring revenue, but may also require companies to bear substantial capital expenditure first.
The key verification points for the next stage are whether data centers and industrial projects can commence operations on schedule, whether on-site power generation can truly be delivered, whether fuel costs decline as expected, and whether electricity pricing mechanisms can support investment returns. Morgan Stanley's core judgment thus rests on a specific shift: as electricity demand expands, infrastructure that can deliver on time, maintain continuous power supply, and convert costs into reasonable returns will gain stronger pricing power in the power industry chain.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
