Episode Summary
Executive Summary: The episode examines how exploding data center demand is reshaping clean power strategy, especially the rise of co-located and behind-the-meter generation. Sheldon Kimber argues the grid is too slow and constrained for new gigawatt-scale loads, making solar, storage, and some gas at the site the most practical way to deliver speed, scale, and clean power. He also updates the solar/storage market, emphasizing policy, tariffs, and pricing realities.
Main Topics: Data centers as the new power-sector demand shock (Priority: 5/5): The conversation frames AI and broader electrification as creating unprecedented load growth, forcing developers and utilities to rethink where power is sourced and how quickly it can be delivered. Why co-located generation is gaining traction (Priority: 5/5): Kimber argues that putting generation next to data centers solves speed-to-power, transmission bottlenecks, community opposition, and grid capacity limits better than waiting for distant grid upgrades. Solar, storage, and gas as a hybrid baseload solution (Priority: 5/5): Rather than treating gas, renewables, and batteries as mutually exclusive, he describes a mixed portfolio that can serve data centers with high clean-energy uptime and competitive economics. Solar and storage market conditions in the U.S. (Priority: 4/5): Kimber describes a market in limbo: strong demand from AI and electrification alongside policy uncertainty, tariff risk, and shifting input costs that affect project viability and pricing. PPA pricing, tenor, and risk allocation (Priority: 4/5): The discussion covers how PPA terms moved from longer contracts to shorter structures and are now stabilizing around prices that better reflect project risk and capital costs. DeepSeek, Jevons paradox, and compute demand (Priority: 4/5): Kimber dismisses the idea that efficiency breakthroughs will reduce overall power demand, arguing that better models and inference will likely increase total compute and electricity use. Intersections with hydrogen, crypto, and prior development strategy (Priority: 3/5): Intersect’s earlier work on hydrogen and large load sites positioned it to pivot quickly toward AI data centers, similar to how Bitcoin-mining infrastructure is being repurposed.
Key Arguments: The grid is too slow and constrained to reliably serve the coming wave of multi-gigawatt data centers, so load must increasingly come to generation. Co-located renewables plus storage can provide 75%–80% carbon-free operation in favorable regions while still meeting industrial-scale power needs. A hybrid design with wind, solar, batteries, and limited gas can be cheaper and faster than building new combined-cycle gas plants from scratch. PPA prices have risen to a healthier level that better reflects risk and supports more sustainable developer economics; the issue is not windfall margins but project viability. Tariffs and supply-chain politics remain a major structural risk for the U.S. solar/storage industry, especially for Chinese-sourced equipment. Efficiency improvements in AI will not necessarily reduce electricity demand; they may increase it through Jevons-paradox effects and expanded inference usage. Data centers increasingly need grid optionality, but fully off-grid operation is becoming technically feasible for some sites. The most important constraint for data centers is now speed to power, not just access to low-cost electricity or clean-energy matching.
Data Points: Intersect Power/Google/TPG initial investment: about $800 million - Announced partnership to develop co-located solar, storage, wind, and some gas for data centers in the U.S. Energy Hub dispatchable capacity: 3.4 gigawatts - Virtual power plants made from 2.5 million customer devices across North America. Customer devices aggregated by Energy Hub: 2.5 million - Thermostats, batteries, and EVs shifted energy during peak periods. Time period for device shifting: May and June alone - Millions of devices shifted energy during peak periods across North America. Site scale discussed for Texas projects: 3 gigawatts and over 1 gigawatt - Two multi-gigawatt-scale data center sites in the Texas panhandle. Carbon-free operating window for the hybrid sites: 75% to 80% of hours of the year - Kimber described achievable uptime using wind, solar, and batteries on-site. Projected gas/grid reliance: about 20% of the time - Remaining hours where gas or grid power would be needed at the hybrid sites. First Google-linked data center timing: 2026 - Intersect said a sizable data center would come online then as part of the co-location plan. Gigawatt-scale phases timing: end of 2027 to early 2028 - First phases of the larger Texas projects could come online in that window. Project financing underway: about $9 billion of CapEx - Intersect was in the middle of financing a large project pipeline. Data center cost example: $10 billion total structure plus power assets - Kimber estimated a gigawatt data center with generation and powered shell, excluding chips. Power assets cost example: $5 to $6 billion - Estimated wind, solar, batteries, and some gas for a gigawatt-scale site.
Pivotal Quotes: "there are opportunities to actually make these data centers some of the most, you know, sort of clean-powered industrial loads in the country" — Shao Khan: Opening framing of the discussion on whether data centers must be fossil-heavy or can be clean-powered. "the grid is broken, right? Which is something... which means the load will have to come to generation." — Sheldon Kimber: Core rationale for co-locating generation with data centers instead of waiting on traditional grid expansion. "When machines find the internet, they will be its biggest user by order. Orders of magnitude." — Sheldon Kimber: Argument that AI inference and agentic systems could create recursive demand growth rather than efficiency-driven decline.
Implications: Expect faster growth in hybrid, co-located clean power for data centers, with solar/storage plus limited gas becoming a serious alternative to grid-only service. Policy uncertainty and tariffs remain major risks, but demand from AI and electrification appears durable.