Can xAI’s giant battery help solve AI’s power problem?
Subscribe to our free newsletter today to keep up to date with the latest renewable energy news.
A large Tesla Megapack installation at xAI’s Memphis computing hub could offer an early view of how battery storage will support the next generation of AI data centers.
Its exact size remains unclear.
Canary Media counted 720 Tesla Megapack containers in satellite imagery of the Colossus 2 site taken July 11. Based on Tesla’s available Megapack models, the publication estimated that the installation visible at the time could provide about 2.8 GWh of energy storage and between 720 MW and 1,400 MW of power.
Other figures have since emerged. A SpaceXAI executive told the Tennessee Valley Authority board in August that the system had 3.3 GWh of storage. Memphis Light, Gas and Water CEO Doug McGowan later referred to roughly 2,000 MW of batteries behind the meter.
Neither the utility, TVA nor SpaceXAI has publicly reconciled those figures. That makes it too early to describe the installation as the largest battery project in the US.
The distinction between the figures also matters. MW measures how much power a battery can deliver at one time. MWh or GWh measures how much energy it can store. The figures describe different parts of a battery system and cannot be directly compared.
What is clearer is the scale of xAI’s investment. Even the lower estimates would place the project among the largest battery installations in the US.
More importantly, the project shows how storage could become part of the infrastructure needed to connect large AI facilities to power grids.
Batteries could become part of the data center connection model
AI data centers require large amounts of electricity around the clock. Their power needs can also grow faster than utilities can build substations, transmission lines and new generation.
Battery storage gives utilities and data center operators another way to manage that gap.
Batteries can charge when grid capacity is available and discharge when electricity demand is high. For a data center, this can reduce peak demand from the grid without shutting down computing equipment. Storage can also provide short-term backup during outages and help manage sudden changes in load.
That appears to be part of the approach in Memphis.
MLGW previously said xAI had agreed to reduce its grid consumption when system demand was high. Utility documents from 2025 showed that xAI was receiving 150 MW through two substations at its Paul Lowery Road site. Work was also underway on infrastructure intended to support another 150 MW.
McGowan has since said xAI has enough on-site generation and battery storage to meet its load for up to four hours when the company is asked to reduce grid consumption. He said that capability was a condition of connecting the data center.
This model could become more common as AI companies compete for grid connections.
For an AI developer, delays in securing electricity can leave costly computing equipment unused. Utilities face a different problem. They must connect new customers without reducing reliability for homes and existing businesses.
A large battery gives both sides more flexibility during periods when electricity supply or network capacity is limited.
Storage could therefore become part of the cost of connecting very large data center loads, especially in regions where electricity demand is rising faster than grid infrastructure can be built.
AI demand is meeting a rapidly expanding battery market
The Memphis project is taking shape as US battery deployment grows at a rapid rate.
Utility-scale battery capacity reached nearly 52 GW by the end of June 2026, according to the US Energy Information Administration. That followed average annual growth of 70% during the previous three years. Operators have also reported plans for another 54 GW of battery capacity through 2028.
At the same time, data centers are becoming a much larger source of electricity demand.
The International Energy Agency expects US electricity consumption to rise by more than 420 TWh between 2025 and 2030. Data centers are expected to account for about half of that increase.
A separate Department of Energy report estimated that US data centers consumed about 4.4% of national electricity in 2023. Their share could rise to between 6.7% and 12% by 2028, with annual electricity use reaching between 325 TWh and 580 TWh.
The two trends could become increasingly connected.
Large batteries have often been built alongside renewable energy projects or installed by utilities to support grid operations. AI developers could become another major source of demand, particularly in areas where large new grid connections are difficult to secure.
The economics may also differ from traditional battery projects.
A utility-scale battery often earns money by selling power or grid services. A data center operator can place additional value on storage if it helps bring expensive computing infrastructure online sooner or avoids the need to reduce computing loads during periods of high grid demand.
That could support battery projects that would be harder to justify through electricity-market revenue alone.
Storage can ease grid pressure, but it cannot replace generation
The Memphis project also shows an important limit of battery storage.
Batteries store electricity. They do not generate it.
A data center operating continuously still needs enough energy from the grid or on-site generation to run its servers and recharge its batteries. Short-duration lithium-ion storage can cover periods of high demand for several hours, but it cannot supply a large computing campus indefinitely.
The environmental impact also depends on the source of the electricity used to charge the batteries.
If batteries charge during periods of high renewable output or when lower-carbon generation is readily available, they can shift that electricity to times of higher demand. They may also reduce the need for fossil-fuel power plants that operate mainly during system peaks.
However, a battery charged from gas-fired generators still depends on fossil fuel. Reporting on the Memphis site indicates that xAI’s batteries could potentially charge from its on-site generation when enough grid power is not available.
That means the source of the electricity matters alongside the size of the battery system.
There could also be benefits beyond the data center itself.
If AI developers install large batteries to secure grid access, some of that capacity could eventually support the wider electricity system. xAI has indicated an interest in making its storage available as a contracted resource for the TVA network, although no such arrangement has been confirmed.
Such agreements could give utilities access to large amounts of flexible capacity financed partly by private data center investment. In return, AI developers could gain faster access to electricity while helping utilities manage periods of high system demand.
The Memphis project remains an early test of this approach.
Its final battery capacity still needs to be confirmed, and storage cannot remove the need for new generation and grid infrastructure. Even so, the installation points to a wider change in data center planning.
As AI companies seek larger power connections, batteries may become a standard part of how those connections are designed. Their role will be less about powering AI independently and more about making large, continuous loads easier for the grid to accommodate.
Source
