Invest1 publisher2 min readPublished
TrendForce's 2030 data center power forecast exceeds its own grid estimate by 268 GW
TrendForce projects 490.7 GW of data center power demand by 2030 against 222.6 GW of available grid capacity, a shortfall of about 268 GW. On the forecaster's own numbers, grids would carry about 45% of the load cloud providers are spending to build.
The Investor · Invest desk
What happened
- TrendForce projects global data center power demand of 161 GW in 2026, up 31% from the 122.9 GW it expects for 2025.
- TrendForce says the gap between data center demand and the power grids can deliver opens in 2026 and widens significantly after 2028.
- The US deficit exceeds 170 GW by 2030, concentrated in the PJM, ERCOT and MISO grid regions where data centers already cluster.
- Crypto Briefing, reporting the forecast, says power availability is becoming the main constraint on data center operators' growth.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- decision Cloud providers planning builds after 2027 have to budget for their own power supply, and every dollar spent on generation or transmission is a dollar not spent on chips.
- exposure Households and factories on the three US grids would pay if prices ration the shortfall; Crypto Briefing argues concentrated data center demand could push up wholesale power prices there.
- capability Long-distance HVDC lines lose less power than AC lines and could tie data centers to remote generation that is otherwise stranded, a route TrendForce names for adding supply.
Read year by year, TrendForce's forecast, as reported by Crypto Briefing, applies close to one growth rate five times over. Demand rises from 161 GW to 211 GW in 2027, another 31% [2][1], then to 490.7 GW by 2030, about 32.5% a year for three more years [3][2]. Measured from 2025's 122.9 GW, that is roughly four times the load, compounding near 32% a year [3]. TrendForce has capital spending at the major cloud providers growing above 30% a year into 2027 [5], close to the growth rate of the power forecast.
Most of the load added in 2026 is AI [5]. At about 25% of 122.9 GW, AI servers drew roughly 30.7 GW in 2025. At 33.4% of 161 GW they draw about 53.8 GW, an increase of 75% [4]. The rest of the data center load goes from about 92.2 GW to 107.2 GW, up 16% [5]. Of the 38.1 GW added, AI servers account for about 23 GW, or 60% [5].
Against that demand, TrendForce puts grid capacity available to data centers in 2030 at roughly 222.6 GW [7]. That is only about 11.6 GW more than the 211 GW it forecasts for 2027, so the 2027 load alone would fill about 95% of the capacity expected three years later [7]. The United States has most of the exposure [8]. Its deficit of more than 170 GW is over 63% of the global gap of about 268 GW [8][8].
The gap can close in one of three ways: cloud providers spend less than the forecast assumes, power arrives from sources outside the grid estimate, or prices ration what the grid can carry. I think the 222.6 GW, more than the 490.7 GW, sets how much AI spending can actually be connected after 2027 [7].
The counter-thesis is that capacity is the TrendForce figure most open to revision, and a higher grid estimate would shrink the gap without any cut in spending. Whether power binds before chips do is a separate question, because the forecast as reported does not weigh chip supply against power. The view is wrong if cloud capex keeps compounding above 30% after 2027 and connected load keeps pace with it [5].
What to watch
- Whether TrendForce revises its 222.6 GW estimate of 2030 grid capacity available to data centers, the figure that sets the size of the gap.
- Wholesale electricity prices in PJM, ERCOT and MISO, where TrendForce places most of the US deficit of more than 170 GW.