Issue 069 - AI infrastructure - Battery energy and power density
Can batteries keep an AI campus running?
Reuters has reported multiple proposed AI data-center campuses at gigawatt scale, including projects pairing large compute loads with dedicated power and battery capacity. Consider a hypothetical AI campus drawing 1 gigawatt continuously whose operators want enough battery backup for a four-hour outage.
Energy and infrastructureAbout 1 minute
Sources checked October 6, 2026. Figures and circumstances may have changed.
The problem
If an AI campus draws 1 gigawatt continuously, about how much usable battery energy is needed to keep it running for four hours?
Then estimate the battery mass, land area, and installed cost.
Because Fermi problems target an order of magnitude, I normally use no more than two significant digits and write most calculations in scientific notation; the Fermi reference explains both conventions.
Grounding facts
The key distinction is power versus energy. A 1-gigawatt load is an instantaneous power draw. A four-hour outage requires energy equal to power multiplied by time.
energy needed = power x timeTesla lists its four-hour Megapack at about 979 kW of power and 3,916 kWh of energy, with dimensions around 347 by 65 inches and a maximum weight of 84,000 lb.
For comparison, California ISO described the Moss Landing battery system as 400 MW / 1,600 MWh, enough to serve about 300,000 homes for four hours. EIA's 2024 construction-cost data put U.S. battery storage around $1,469 per kW of installed nameplate capacity.
After checking sources
Checked answer and calculation
Start with the required stored energy:
energy
= 1 GW x 4 hours
= 4 GWh
= 4,000 MWh
= 4.0 x 10^6 kWhUsing Tesla's four-hour Megapack as a scale peg:
Megapacks needed for power
~= 1,000 MW / 0.979 MW
~= 1.0 x 10^3 MegapacksMegapacks needed for energy
~= 4,000 MWh / 3.916 MWh
~= 1.0 x 10^3 MegapacksSo a 1-GW, four-hour backup system is roughly 1,000 utility-scale battery containers. The battery equipment alone would weigh about:
mass
~= 1,000 x 84,000 lb
~= 8.4 x 10^7 lb
~= 3.8 x 10^7 kg
~= 40,000 metric tonsThe raw Megapack footprint is only about 15,000 square meters, but with fire lanes, spacing, inverters, transformers, access roads, and support equipment, a practical site might be on the order of 0.1 to 0.3 square kilometers. That is much smaller than a square mile, but it is still a major industrial installation.
The cost check lands near the billion-dollar scale:
cost
~= 1,000,000 kW x $1,469/kW
~= $1.5 x 10^9Using an energy-cost shortcut of roughly $300 to $400 per kWh gives a similar answer: 4.0 x 10^6 kWh times a few hundred dollars per kWh is roughly $1 billion to $2 billion. Four-hour backup would therefore be a major infrastructure project in its own right, not a modest accessory to the campus.
Before checking sources
Matt's original estimate
This is the unverified estimate Matt wrote before checking sources, not the checked answer.
I don't have great anchors for this one. If it needs to draw 1 GW continuously, that's 10^9 W. If it needs to cover that usage for 4 hours, that's about 1.4 x 10^4 seconds at 10^9 J/s, or about 1.4 x 10^13 J total.
I tried to estimate what a home battery could cover, and I guessed that a home battery could produce about 1000 W continuously for 8 hours, and if oriented next to each other they could be arranged so 10 could fit in a square meter, and each one might have a mass of about 60 kg.
If those guesses are correct, it would take about 10^7 home batteries to cover the continuous demand, and about 6 x 10^6 home batteries to cover the total energy need for that 4 hour period, so the rate is the limiter that needs to be satisfied.
Assuming we need the equivalent of 10 million home batteries, if 10 fit per square meter without any further vertical stacking, it would cover a million square meters, a square kilometer, and the battery mass would be about 6 x 10^8 kg or 600,000 tons of battery.
Reasoning score
Matt's reasoning score: 40 / 100
Higher is better: earn points for useful facts, a sound reasoning approach, correct math, and a final estimate close to the sourced answer. The owl meter shows percent full of it: 100 minus the reasoning score.
Useful facts: 10/30. The home-battery energy anchor was in the right neighborhood, but the mass, power, and cost anchors were weak or missing.
Reasoning approach: 15/30. Power times time was exactly the right starting model, and scaling from modular batteries is reasonable. The rate-versus-energy comparison became muddled, though, and the cost part was left out.
Math: 5/10. The joule conversion was right, but the battery-count division introduced about a one-order-of-magnitude error.
Final estimate: 10/30. The 4-GWh energy requirement was right, but the mass and land-area estimates were high by more than an order of magnitude and the cost estimate was not attempted.
Post-check reflection
Matt's reflection
Well, I was off by a power of ten in terms of the battery count. Joule count for 4 hours was correct, but when I divided by assumed battery stats into that, I came up with one order of magnitude high counts.
I'm a bit surprised that my home battery assumptions were so close, I was a bit low on the rate and very low on the weight, about half the real-world. Ultimately the mass and area errors sort of cancelled out so I was almost within an order of magnitude, but still high.
Also, I didn't even attempt the cost calculation, I should really remember the cost per kWh and the specs on the Tesla home battery.
Recommended memory peg
A 1-GW load for 4 hours needs 4 GWh of storage. For today's grid batteries, that is roughly a thousand large battery containers and about a billion-dollar-scale project.
Reader results
Bars show how submitted estimates sort into the answer choices from the gut-check prompt.