Library/200GW Hiding in Grid, Sodium Batteries 10x Cheaper, Wave-Powered Datacenters w/ Ramez Naam | EP #280

Peter H. Diamandis

200GW Hiding in Grid, Sodium Batteries 10x Cheaper, Wave-Powered Datacenters w/ Ramez Naam | EP #280

2h 06mAug 15, 2026
Key Themes
AI power demandgrid bottlenecksdata centerssolar and storagenuclear powerenergy flexibilityseasonal storageenergy geography
Summary

AI’s growth is colliding with the limits of electricity, grids, and long-duration clean power

This episode explores one big idea from several angles: the next phase of AI is likely to be constrained less by chips alone than by the ability to deliver electricity at scale. The conversation moves from today’s data-center bottlenecks and interconnection delays to near-term fixes like flexible loads, batteries, behind-the-meter generation, and solar-plus-storage, then widens out to the longer-term roles of nuclear, geothermal, and even space-based energy concepts. Throughout, the guests emphasize that geography, regulation, and infrastructure build speed may matter as much as raw technology.

1
Infrastructure can be the real bottleneck behind a fast-moving technology wave

The episode repeatedly shows that breakthrough technologies do not scale only because demand exists; they also depend on physical delivery systems, permitting, and build speed. In this conversation, AI’s growth is tied directly to electricity access, grid interconnection, and data-center infrastructure.

2
Flexibility is becoming as valuable as raw supply

A major thread in the episode is that the fastest way to unlock more capacity is not necessarily to build everything from scratch, but to use storage, interruptible loads, and software coordination to make existing grids behave more efficiently. That idea applies to data centers, EV charging, and broader power systems.

3
Cheap generation still needs firm delivery and storage

Solar is presented as an increasingly low-cost source of electricity, but the episode makes clear that a cheap resource is not automatically a usable one. Without batteries, transmission, seasonal storage, and land/permitting solutions, low-cost generation cannot reliably support round-the-clock demand.

4
Long-duration storage is a harder problem than short-term smoothing

The speakers distinguish between shaving peaks or shifting daily demand and solving winter-to-summer or season-to-season mismatch. That makes seasonal storage one of the most important unresolved issues for a solar-heavy power system.

5
Nuclear is being reframed as a manufacturing and learning-curve story

Rather than treating nuclear as a single megaproject category, the episode emphasizes repetition, standardization, and factory-style construction. That framing suggests that the economics of future nuclear power may depend as much on industrial process improvements as on reactor design alone.

6
Where energy is abundant may matter more than where demand is highest

The discussion repeatedly returns to geography: sunny regions, warm regions, places with strong regulation, and countries with easier land assembly all emerge as better candidates for energy-intensive compute. That makes siting a strategic decision, not just a logistical one.

Select any chapter text to Deep Dive with AI
01AI power demand, grid bottlenecks, and large-scale solar costs

The opening chapter argues that AI’s rapid expansion is constrained more by electricity delivery than by chip prices. It covers rising data-center demand, slow interconnection timelines, grid bottlenecks, and the idea that compute may need to move closer to abundant energy rather than waiting for the grid to catch up.

Electricity access is presented as the central constraint on AI growth.
Interconnection queues and permitting delays are slowing new generation and large-load hookups.
Data-center costs are so large that power availability can matter more than energy price.
Recursive self-improvement is treated as unlikely to produce a sudden AI takeoff because of diminishing returns.
Solar, batteries, and local generation are discussed as practical near-term responses.
The discussion frames grid expansion as a distribution problem, not just a generation problem.
02Nuclear power, grid flexibility, and powering AI data centers

The conversation shifts to the immediate infrastructure problem: how to power new AI data centers before the grid can be upgraded. The speakers discuss behind-the-meter generation, gas turbines, regulatory reform, and flexible-load strategies that can unlock more usable capacity from the existing system.

The immediate bottleneck is described as power delivery and available data-center infrastructure.
Behind-the-meter generation is presented as a major workaround.
Large gas turbines are in tight supply, creating multi-year lead times.
Texas is highlighted as a leader in flexible-load and interconnection reform.
Utility incentives are criticized as favoring slow spending over fast deployment.
Battery storage and software orchestration are positioned as practical near-term tools.
03Interruptible loads, grid storage, and the economics of solar-powered AI

This chapter explains how batteries and flexible workloads can make data centers easier to serve by turning them into interruptible loads. It then explores how cheaper solar and storage, plus faster build times, could support gigawatt-scale AI infrastructure, while also noting that chip costs are only part of the total system economics.

Batteries are framed as a way to shift demand to off-peak hours.
Interruptible-load rules could unlock significant extra grid capacity.
Solar-plus-storage is increasingly viable for large data-center projects.
Permitting and land assembly remain major deployment hurdles.
Sodium-ion batteries are presented as a potential cost breakthrough.
GPU pricing is discussed as a mix of supply-chain markup and ecosystem moat.
04Long-term prospects for solar, nuclear, and space-based energy

The final chapter broadens the lens to long-run energy systems, emphasizing solar’s decline in cost alongside its geographic and seasonal limits. It then examines nuclear fission, SMRs, geothermal, and storage as candidate firm-power solutions, and closes by tying these energy choices to AI siting, geopolitics, and future energy competition.

Solar remains cheap but is constrained by winter and latitude.
Seasonal storage is still an unsolved challenge.
Electrifying heat could materially increase winter electricity demand.
Nuclear is framed as a factory-scale learning problem.
Small modular reactors may benefit from standardization and factory production.
AI data centers are portrayed as a potentially major source of new nuclear demand.
Geography and geopolitics shape where energy-intensive compute can be deployed.