What data centres consume is seats. Among the ways to create new seats, batteries are the fastest.
A data centre is not a resource that makes frequency wobble. It is demand that draws a near-constant volume all day long. So what grows scarce as they multiply is not balancing power but the volume of generation that can "definitely deliver" on hot and cold evenings — that is, seats (capacity).
Japanese demand has turned upward
Japan's electricity demand fell for a long time with population decline and efficiency gains. The 7th Strategic Energy Plan reverses this, putting FY2040 generation at 1.1–1.2 trillion kWh (10–20% above FY2022). The reasons are data centres, semiconductor fabs and electrification. In OCCTO's FY2026 demand projection, data-centre ramp-up is running behind expectations, but from FY2034 onward it exceeds the previous projection.
← scroll sideways →
What happened in the US: seat prices up elevenfold
PJM, the grid operator in the eastern United States, buys seats three years ahead by auction. Its price jumped from $28.92 for 2024/25 to $269.92 for 2025/26, $329.17 for 2026/27 and $333.44 for 2027/28 (per MW per day), and cleared at $325 for 2028/29 as well (14 July 2026). The three auctions from 2026/27 onward all cleared pinned to the price cap. PJM attributes roughly 5,100 MW of the 5,250 MW demand increase for 2027/28 to data centres, and its market executive Stu Bresler has said that data-centre demand is far outrunning new supply. The market monitor estimated that including the data-centre demand forecast raised total 2027/28 auction costs by around 60% ($6.2bn). At ERCOT in Texas, roughly 410 GW of large loads were waiting to connect as of March 2026, 87% of them data centres.
← scroll sideways →
Batteries create seats fastest
There are three ways to add seats: build thermal, keep thermal, or install batteries. New thermal takes years from plan to operation, and keeping it requires payment (48-6). A battery goes on flat ground, runs in one to two years in Britain (in Japan some projects take three to five, waiting on grid works — see 48-13), and can commit to "definitely delivering" for several evening hours. In PJM's 2028/29 auction, a four-hour battery counts 59% of its rating as a seat. In Japan's capacity market, batteries likewise count as seats through a coefficient based on how long they can deliver continuously (the calculation method differs, but the thinking is the same).
The transmission outlook changed with data centres too
On 1 September 2026, OCCTO published its outlook for transmission congestion in FY2031. From this edition, large loads such as data centres are treated as flat round-the-clock demand rather than demand that rises and falls with the area's demand curve, with the curve applied only to general demand. This raises assumed demand in the places where data centres sit. The effect shows in the figures: in the transition scenario, Tokyo's curtailed volume fell by 1,263 GWh against the previous year's projection. The FY2031 assumptions include 24.09 GW of batteries (up 11.79 GW on the previous projection) and 17 reactors. Write in the large loads, and the assumed congestion and curtailment in that location shrink. Where data centres sit is starting to bear on where batteries should go.
- 48-2The grid used to run fine without batteries, didn't it?
- 48-3What disappears from a grid with fewer spinning machines?
- 48-4Why "ten seconds"?
- 48-5Does more renewables mean more balancing power is needed?
- 48-6Can't thermal just do the balancing?
- 48-7If nuclear grows, do we stop needing batteries?
- 48-8Wouldn't more pumped storage be enough?
- 48-9Where does the balancing power that the market failed to buy come from?
- 48-10Why do data centres point to batteries? (this article)
- 48-11What happens to solar from here?
- 48-12Thermal volume, nuclear operation, demand — what next?
- 48-13How much battery capacity is actually coming?
- 48-14Does the government actually want more batteries?
- 48-15What does battery storage resemble as an infrastructure investment?
- 48-16They say the earnings will thin out. Is that true?
- 48-17Can you copy the overseas playbook and make money in Japan?
- 48-18How much battery storage will Japan ultimately need?
- 48-19Is a battery the same wherever you put it?
Sources
- Agency for Natural Resources and Energy, "Outlook for Energy Supply and Demand in FY2040 (supporting materials)," Cabinet decision of 18 February 2025 (generation 1.1–1.2 trillion kWh, 10–20% above FY2022)
- OCCTO, "National and Area Demand Projections (FY2026)," 21 January 2026 (data centres pushed back; exceeding previous projections from FY2034)
- PJM, "Base Residual Auction Report," various editions (2024/25 $28.92; 2025/26 $269.92; 2026/27 $329.17 [cap]; 2027/28 $333.44 [cap]); PJM news releases, 17 December 2025 (data centres approx. 5,100 MW; remarks by Stu Bresler) and 14 July 2026 (2028/29 $325 [cap]); Monitoring Analytics, "Analysis of the 2027/2028 RPM BRA Part A," 5 January 2026 (increase in total auction cost from including the data-centre demand forecast: $6.24bn, 61.4%, scenario 3)
- ERCOT, "Large Load Update," House Committee on State Affairs, 9 April 2026 (approx. 410 GW as of 26 March 2026; approx. 87% data centres)
- PJM ELCC (four-hour battery, 2028/29: 59%); OCCTO, "Capacity Market Main Auction," adjustment coefficients for batteries
- OCCTO, 103rd Wide-Area Grid Development Committee, Material 3, "Medium- to Long-Term Outlook for Grid Congestion in FY2031," 1 September 2026 (p.14 treatment of large loads, p.26 Tokyo curtailed volume, p.56 assumed installed capacity)