AI Data Center Energy Demand Pushes Battery Storage to Record

Battery storage has become an AI infrastructure business

Lists of the "energy storage giants powering AI" usually come down to the same names: Tesla, Fluence, CATL, LG Energy Solution, Samsung SDI, and power-equipment vendors such as Schneider Electric, Vertiv and Eaton. This autumn's news shows why those companies are getting attention. The most important story is not which stock might rise. Grid electricity is running short, and batteries have become the fastest way to fill part of the gap. The coverage points to a second story too: the companies best placed to profit are struggling to build fast enough.

The demand numbers frame the problem. TrendForce estimates that global data center power demand capacity will reach 161 GW in 2026, up about 31% from 122.9 GW in 2025. AI servers are expected to make up roughly a third of that load.26 The firm forecasts total demand of 490.7 GW by 2030. It estimates that grids will be able to supply only about 222.6 GW of that, leaving a gap of roughly 268 GW.22 For the United States alone, TrendForce expects the gap to pass 170 GW by 2030.26

The forecasts agree on direction but not on size

The forecasts all point the same way, but they measure different things, and that affects how alarming the numbers look. TrendForce counts power capacity in gigawatts. Gartner, as summarized in a 2026 industry roundup, counts consumption. It projects global data center electricity use will rise 26%, from 447 TWh in 2025 to 565 TWh in 2026, with AI-optimized servers taking 31% of that power.24 Goldman Sachs is the most focused on the US. It expects US data center power demand to grow 38% in both 2026 and 2027, adding 12 GW and then 17 GW.25 Goldman also raised its end-2026 US capacity estimate by 5 GW to 64 GW and cut its end-2027 estimate by 5 GW to 90 GW. In other words, the bank sees a faster near-term buildout but a somewhat lower peak.27

TrendForce is unusually open about the limits of its own headline number. It says the 268 GW figure leaves out on-site "behind-the-meter" power that operators build for themselves, so the real shortfall may be smaller. It also says delays in connecting to the grid and in building transmission could cause genuine shortages.22 The firm estimates that about 49 GW of US demand will not get grid power on schedule. Three regional grid operators account for most of it: PJM with 19.5 GW, ERCOT with 11.2 GW and MISO with 7.6 GW.

The UK shows the same problem in miniature. In February, regulator Ofgem warned that about 140 proposed data center projects had asked for roughly 50 GW of grid connections. That is more than the country's peak demand of about 45 GW. Many of those requests are thought to be speculative attempts to hold a place in the queue.24

The forecasts differ widely on size, but they agree on one point: grid connections, not chips, now set the pace of AI buildouts. Batteries fit into that gap.

Why batteries, and why now

The most useful framing comes from Wood Mackenzie. Its analysts argue that storage can add capacity faster, more cheaply and more reliably than gas alone. As large cloud operators run into grid-connection bottlenecks, they say, utility-scale batteries will deliver "speed to power" for data centers.18 The change in outlook is striking. A year ago, Wood Mackenzie expected a four-year slump in the US storage market starting in 2026. It now expects more than 130 GW / 534 GWh of new utility-scale storage between 2026 and 2031, and says data centers are the main reason.17

Installation data supports the more optimistic view, with one caveat. The US installed a record 18.9 GWh of storage in the second quarter of 2026.18 But utility-scale power capacity, measured in gigawatts, actually fell about 8% from a year earlier, to just under 4.7 GW.18 Energy grew while power fell because systems are getting longer-lasting: average duration rose from 2.8 hours to 3.5 hours.17 Our reading is that buyers increasingly want batteries that move energy across hours, not just smooth out short power fluctuations. That shift favors suppliers that can deliver large volumes of cells.

AI workloads also change what batteries inside data centers are expected to do. Industry analysis describes AI training as a steady, schedulable load. AI inference, by contrast, produces sharp demand spikes that traditional backup systems, sized to run for minutes, were never designed for.16 One market study expects "AI load management" to be the fastest-growing use of data center storage, driven by split-second power swings from large GPU clusters.12 A survey of 150 data center professionals found that 57% see higher power density and smaller footprints as AI's main effect on their storage needs.11

The companies in the spotlight

Tesla has the clearest showcase. Satellite images show about 720 Megapacks at xAI's Colossus 2 facility in Memphis, with an estimated 2.8 GWh of storage. Reports call it the largest grid-connected battery in the US.7 xAI and SpaceX are reported to have spent about $1 billion on Tesla batteries from 2024 to mid-2026.7 Reports also say the batteries reduced the site's reliance on temporary gas turbines.7 Much of this revenue stays within companies controlled by Elon Musk, which makes the demand signal less independent than it first looks.

Fluence shows both the opportunity and the risk. It had about $850 million in data center business, including roughly $550 million from one large cloud customer, and its backlog reached a record $6.4 billion.49 But it has now cut its fiscal 2026 guidance twice, mostly because of delays at its new Houston factory, which is expected to reduce revenue by about $1 billion.4 CEO Julian Nebreda said the delays "create problems" for data center customers, who go from first contact to signed contract in under three months, compared with about 18 months for other buyers.4 Jefferies analysts warned that the pitch for batteries depends on fast, cheap capacity, and that buyers who lose confidence will look elsewhere.4

Cell makers are where the money ultimately flows. EVE Energy signed a framework deal to supply Fluence with 206 GWh from 2027 to 2031. Only 16 GWh is a firm commitment for 2027; the other 190 GWh is reserved capacity that depends on Fluence winning orders and on US policy toward Chinese batteries.9 Samsung SDI is converting production lines in the US to make lithium iron phosphate (LFP) cells for storage.17 The head of a Turkish battery industry association says AI data centers have become the battery industry's second-largest growth driver after electric vehicles, and notes that China makes 86% of the world's batteries.15

Automakers are also moving in, using factory capacity left idle by weaker-than-expected EV demand. Ford Energy is converting its Kentucky plant to make stationary storage. GM's joint venture with LG Energy Solution, Ultium Cells, has started producing storage batteries.19 At the same time, at least 75 data center projects worth about $130 billion were delayed or disrupted by local opposition in early 2026. That raises the question of whether the political backlash against data centers could reach their battery suppliers.19

Power-equipment vendors compete for the batteries inside the building. One market study lists Schneider Electric, Vertiv and Eaton as having the largest shares of the data center battery market.20 Vertiv has partnered with CPower to let data centers use on-site batteries for demand response and grid services.11

Market forecasts vary wildly

Estimates of the data center storage market's size are a reminder to be skeptical. One firm puts it at $952 million in 2025, growing to $4.27 billion by 2035.20 Another says $4.96 billion in 2026, rising to $18.79 billion by 2036.12 A third projects $10.48 billion by 2032.13 Differences this large mostly come from different definitions, especially whether grid-scale batteries serving data centers are counted alongside backup systems inside the building. The precise totals matter less than the direction, which all three agree on.

The bottom line

Batteries do not generate electricity. They move power from one time to another and buy time while new generation and transmission catch up. That limits how much they can do against a gap of the size TrendForce describes. Google's 30 GWh iron-air battery project with Xcel Energy and Form Energy in Minnesota points to longer-duration systems that could do more.14 For now, though, storage is a bridge while the grid catches up.

The coverage suggests that demand for that bridge is real and growing fast. The constraint has shifted to manufacturing: factory ramp-ups, cell supply deals and dependence on Chinese producers. Fluence's troubles show that supplying fast-moving data center customers takes manufacturing discipline. The storage companies that benefit most from the AI boom will be the ones that deliver equipment on schedule, not the ones announcing the biggest pipelines.