xAI's Memphis battery runs two hours at its top rated output

At 1.4 GW, the top of the output range cryptobriefing.com gives, 2.8 GWh empties in two hours; at 720 MW, the bottom, it runs about 3.9 hours ^[3]. So this is a store that moves power around inside a day. It absorbs grid power during off-peak hours and discharges it during demand spikes ^[9]. The temporary turbines were filling the same gaps in grid supply ^[8].

The capacity is an estimate, and the 3.3 GWh high end sits 0.5 GWh above the 2.8 GWh central figure, a spread of about 18% ^[3][4]. The output range spans nearly a factor of two because the specific Megapack model in the field is not identified ^[4].

Now the money. Over the roughly two years cryptobriefing counts, the near-$1bn works out at about $500m a year, and the single April 2026 order of $269m is about 27% of the entire two-year total ^[1][2]. Divide $1bn by 720 packs and you get roughly $1.4m each ^[5]. Treat that as a ceiling: the figure covers related infrastructure and purchases by two companies, not 720 units at list ^[5]. Tesla's energy segment revenue sits outside this report, so this customer's share of it can't be worked out from what is here.

SpaceX acquired xAI in February 2026 ^[7]. cryptobriefing.com wrote that "SpaceX acquired xAI, xAI buys Tesla batteries, and the capital flows in a loop that benefits the broader Musk ecosystem" ^[13]. An order placed inside one ownership group tells you less about third-party demand than the same order won competitively. I'd discount it for that reason when reading Tesla's storage backlog. The more interesting version of the counter-argument is that the hardware still has to be built and shipped. Tesla started production of the next-generation Megapack 3 at its Texas Megafactory in August 2026 ^[11], and 720 self-contained utility-scale units standing at one address is manufacturing throughput a utility buyer can go and count ^[15][1].

If the direct grid connection the Tennessee Valley Authority board approved in August 2026 carries the firm load of a campus training on hundreds of thousands of GPUs ^[10][12], the battery cycles less often. Then the installation says little about storage economics at AI sites in general. And if the $1bn is weighted toward SpaceX infrastructure away from Memphis, the per-site figure drops well below $500m a year ^[5][1]. The turbines are the observable check. They were on site to fill supply gaps that the battery now absorbs off-peak, and Memphis Light, Gas and Water officials have acknowledged the scale of the system publicly ^[8][9][14].