Sam Altman-backed Exowatt aims to run AI data centers on hot rocks
Exowatt, a startup backed by OpenAI CEO Sam Altman, this month outlined an audacious plan to power high-density AI data centers using modular thermal energy storage built around heated rocks. The company says its approach — essentially a giant, networked thermal battery — could cut costs, buffer grids and provide dispatchable power for compute-heavy workloads that are driving explosive demand for electricity across hyperscale campuses worldwide.
Why hot rocks? The tech and the market context
Data centers running large-scale AI models are among the fastest-growing electricity consumers. Providers from Microsoft and Google to Amazon Web Services and dedicated AI colocation operators are racing to add capacity, while utilities grappling with renewables integration worry about peak demand. Exowatt is pitching what it calls a low-cost, long-duration thermal energy storage (TES) alternative to lithium batteries and conventional generation.
Rather than batteries or molten salt systems, Exowatt’s design uses inexpensive, inert rocks as the storage medium. The rocks are heated electrically — at times of excess renewable output or low wholesale prices — to high temperatures and held in insulated modules. When compute demand spikes, that stored heat is converted back into electricity or used to drive heat engines and turbines paired with waste-heat integration for servers. Exowatt’s public materials emphasize scale: the startup says the approach is modular and manufacturable, able to be deployed by the thousands to meet multi-gigawatt requirements.
How it compares to other storage technologies
Thermal storage has been used in concentrated solar power and industrial settings for years. Exowatt is trying to adapt that model for data centers’ unique profile: extremely high and predictable power draw, local heat reuse opportunities, and the ability to colocate storage physically near compute loads. Compared with lithium-ion, thermal systems promise cheaper long-duration capacity but lower round-trip efficiency. Compared to molten salt, rock-based systems avoid corrosion and expensive heat-transfer fluids but raise engineering questions about thermal cycling, packing density and materials handling.
Business model and deployment plans
Exowatt says its target customers are hyperscale cloud providers, AI-specialized colocation firms, and operators looking to reduce grid dependence and energy costs. The company plans a phased rollout with demonstration projects adjacent to existing data centers to validate integration with server cooling and power electronics. Exowatt’s pitch centers on three selling points: lower levelized cost of energy (LCOE) for long-duration needs, resilience during grid outages, and the ability to time-shift renewable power for AI workloads.
Company materials describe a vision of “billions of hot rocks” in industrial-scale repositories acting as a distributed thermal battery network that can be scaled by adding more modules. That language speaks to the startup’s ambition but also raises practical questions about land use, permitting and supply chains for insulated containment systems, heat exchangers and associated power conversion systems.
Expert perspectives and industry implications
Industry analysts say thermal storage for data centers is a logical extension of existing efforts to capture waste heat and improve PUE (power usage effectiveness). An energy-storage consultant noted that long-duration, low-cost storage can be transformative for facilities with predictable load shapes — like AI training clusters — but cautioned that round-trip efficiency and conversion-capacity constraints will determine which workloads are suitable for running off thermal storage.
From a utility and policy perspective, technologies that can shift demand away from peak hours help smooth grid integration of renewables. For operators, the main trade-offs are capital costs, operational complexity, and integration with power electronics and cooling systems. If Exowatt’s modules can deliver multi-hour discharge at competitive economics, they could become an attractive option alongside hydrogen, molten salt and battery systems for specific use cases.
Risks, challenges and unanswered questions
Key technical questions remain: how hot will the rocks need to be, what conversion technology will be used at scale, and how will thermal cycling affect longevity and maintenance costs? Environmental and regulatory considerations — dust, fire safety, and local zoning for large thermal stores — will also shape rollouts. Finally, interoperability with evolving AI hardware and the trading of wholesale power markets will determine the commercial viability of Exowatt’s deployments.
Conclusion: a potential piece of the AI power puzzle
Exowatt’s proposal to power AI data centers with “billions of hot rocks” is emblematic of the creative infrastructure thinking happening around the rapid growth of AI compute. Whether rock-based thermal batteries become a mainstream solution will depend on demonstrable economics and operational reliability. For cloud operators and utilities, the bigger picture is clear: solving AI’s energy challenge will require a diverse portfolio of storage and generation technologies, and Exowatt is positioning itself as one more option in that evolving stack.
Related topics for further reading: AI infrastructure, data center cooling, thermal energy storage, grid-scale batteries, PUE and data center resilience.
About the Author
