Solar stops at night and a data center does not, and that mismatch has turned the hyperscalers into some of the largest buyers of grid-scale battery storage in the world, according to SpaceDaily, reported by Lachlan Brown. The IEA figures anchoring it: global data center electricity consumption reached about 485 terawatt-hours in 2025 and is projected near 950 terawatt-hours by 2030, with 20 to 25 gigawatts of batteries potentially installed at data centers by the end of the decade. Google's Steel River Energy Center in Arkansas pairs 1.6 gigawatts of solar with 1.9 gigawatt-hours of storage across its first two phases, reaching 2.5 gigawatts and 2.9 gigawatt-hours when the third completes in 2029. Its Pine Island site in Minnesota combines 1.4 gigawatts of wind, 200 megawatts of solar, and a 300 megawatt iron-air battery rated at 30 gigawatt-hours for roughly 100 hours of discharge. Amazon had 15 paired solar and storage projects by the end of 2025 totaling 2.3 gigawatts, including 600 megawatts of solar, 600 megawatts of battery, and 100 megawatts of geothermal with NV Energy in Nevada.
Lithium iron phosphate holds its cycle life best in roughly the 15C to 35C band, and pushing cell temperature above that accelerates calendar aging in a way no warranty absorbs quietly. That is a far tighter requirement than the racks, which are moving in the opposite direction toward 45C supply water precisely so the heat becomes easy to reject. A gigawatt-hour of storage on a data center site is therefore a second cooling system with its own setpoint, its own fire code, its own containment, and its own failure mode, and it cannot share the warm loop the IT side is standardizing on. Iron-air chemistry behaves differently and still needs thermal management sustained across a hundred-hour discharge, which is a duty cycle no data center mechanical system has ever been asked to hold. Google also reported 100 million lithium-ion cells sitting inside racks in 2025, which puts a third thermal regime inside the cabinet itself. Nickel-zinc is being pitched into this gap, and SoftBank is building the same stack in Sakai.
Peak solar output and peak dry bulb arrive within a couple of hours of each other. That is the worst moment for a dry cooler, whose capacity falls as ambient rises, and simultaneously the cheapest power on the site. Overnight the ambient drops, the dry cooler recovers its approach, and every kilowatt the mechanical plant draws is coming out of a battery that had to be bought, sited, and cooled. An operator that shifts deferrable work into the cool hours wins twice, once on cooling efficiency and once on storage depth, and almost nobody schedules compute that way today. Load flexibility is worth hundreds of millions to a western grid operator, and the heat waves that break this pattern are getting more frequent.
The NV Energy line is the one to watch. That utility spent last week in court arguing about who funds two gigawatts of data center interconnection in Reno while signing 600 megawatts of paired solar and storage on the other side of the building. Both stories are the same story. The cooling scheme determines how much generation and how much storage a campus needs, and right now that number is being set by whoever specifies the chiller plant, months before anyone prices a battery. Ask for the annualized PUE against the local weather file before you size the storage, not after.