"200GW Hiding in Grid, Sodium Batteries 10x Cheaper, Wave-Powered Datacenters w/ Ramez Naam | EP #280" — Peter Diamandis (Moonshots)

Why this is in the vault

Ramez Naam is a tracked energy-cost-curve forecaster whose framing of the AI power bottleneck (distribution grid, not generation) bears directly on RDCO's chip-fab/memory capital-cycle investing thesis and the broader question of what actually gates AI-infrastructure buildout.

Episode summary

Diamandis and Ismail interview energy investor Ramez Naam across three linked segments: why the US grid's distribution bottleneck (not generation) is capping AI data center growth, how sodium-ion battery chemistry could extend the ~14x lithium-ion cost decline another 10x, and a wave-powered ocean datacenter startup targeting ~2 cents/kWh in the Southern Ocean. A long middle segment also covers fusion's shift from "always 50 years away" to a credible multi-approach venture-funded race (Commonwealth Fusion, NIF, Helion).

Key arguments / segments

• The real AI power constraint is the distribution grid (substations, transformers, poles/wires) — generation capacity for solar/wind/batteries/gas can be built fast; interconnection cannot.
• A Texas ERCOT data center requesting power today won't get grid-connected before 2031-2032, even on the most deregulated, fastest-moving grid in the US; ERCOT peaks near 80GW against 200GW+ of mostly speculative demand-side queue requests.
• Naam frames the core gap as ~230GW of projected 2030 AI power demand against only ~100GW of projected US grid buildout — a roughly 200GW / $10 trillion shortfall — and argues most of it closes through demand flexibility (batteries, interruptible loads) rather than new generation.
• Lithium-ion battery costs have already fallen ~14x since 2010; sodium's far greater geological abundance could push sodium-ion chemistry another ~10x cheaper, governed by Wright's Law (cost falls with cumulative production, not elapsed time) rather than a fixed timeline.
• Batteries solve daily day/night load-shifting economically but not yet seasonal storage — Naam argues that gap is why nuclear, fusion, and geothermal still matter.
• Fusion has shifted from perpetual punchline to a real multi-approach race: Commonwealth Fusion Systems (MIT spinout tokamak, shrunk via high-temperature-superconducting magnets), NIF (laser inertial confinement, hit net gain but not commercializable), and Helion (pulsed magneto-inertial, captures energy directly as electricity).
• Helion holds a Microsoft power-purchase agreement for 50MW of fusion power by 2028 — the sector's most aggressive timeline — though Naam is candidly skeptical any startup hits its date exactly.
• Adjacent to the three named topics, Elon Musk's orbital-datacenter math is teased: putting 1GW of compute in orbit needs roughly 6x SpaceX's best-ever annual launch year, implying a scale-up to 5-6 Starship launches per day for 10GW/year of orbital compute.
• A wave-powered datacenter startup (Naam passed on its seed round five years ago, has since invested twice at higher valuations; Peter Thiel led its most recent round) deploys ~80-meter submerged, wave-bobbing cone units in the Southern Ocean near Antarctica, targeting ~2 cents/kWh with GPU heat sinks passively cooled by ~40°F seawater; three units are already deployed, a fourth launching soon.

Notable claims

• Grid interconnection queue times have grown from ~15 months (20 years ago) to ~45 months today.
• New Texas regulation (enacted June 2026) fast-tracks "interruptible load" interconnection from 5-7 years down to 12-18 months; FERC has directed six other regional grids to adopt similar rules.
• A cited paper (Tyler Norris) finds that being flexible just ~100 hours/year (~1% downtime) unlocks ~100GW of grid capacity, worth roughly $5 trillion in data center capex.
• UAE 24/7 firm solar+battery plant: 1GW guaranteed output from 5GW solar + 19GWh batteries at $6/watt capex, versus ~$15/watt for the last US nuclear plant built and ~$4/watt for the cheapest Chinese nuclear builds.
• Solar panel cost fell from ~$100/watt (1975) to ~$0.08/watt (current Chinese panels), a >1,000x decline over 50 years; Naam self-describes as one of the top five solar-cost forecasters.
• Wave-datacenter target cost: ~2 cents/kWh, competitive with or cheaper than nearly everything on land except solar (and wind in some locations).

Guests

Ramez Naam — energy investor and author (founder/managing partner, Planetary VC); former Microsoft computer scientist and founding faculty at Singularity University leading its energy track; author of The Infinite Resource and the Nexus sci-fi trilogy; tracked by RDCO as a Wright's-Law-style forecaster of solar, battery, fission, and fusion cost curves. Salim Ismail co-hosts alongside Peter Diamandis, as usual for the Moonshots format.

Sponsorship

Three distinct third-party ad-reads appear in addition to Diamandis's own Moonshots/Abundance360 self-promotion (a plug for the Sept. 25 Moonshots Live event): Google for Startups (a generative-media technical guide for deploying Google DeepMind models), Blitzy (an AI coding-agent platform pitch: "autonomous software development with infinite code context"), and Fountain Life (a health/longevity segment on cancer screening with its CMO, explicitly framed as the show's sponsored health section).

Mapping against Ray Data Co

Naam's core claim — that distribution-grid interconnection, not generation buildout, is the binding constraint on AI datacenter capacity through at least 2031 — is a direct input to RDCO's power-cycle and memory/chip-fab capital-cycle investing theses (see 01-projects/investing/theses/2026-05-18-power-cycle-v1.1.md): if power delivery lags chip supply, the AI capex cycle's pacing item shifts from fabs to grids and batteries, which reshapes which phase of the cycle we think we're in. The sodium-battery and wave-datacenter threads are also directly relevant AI-infrastructure signal for the phData bet and L5 north star — power availability is a real, dated (2031-2032) bottleneck the org should track alongside chip supply, not treat as a solved background assumption.

Related

[[2026-05-18-power-cycle-v1.1]] [[2026-05-15-jane-street-dwarkesh-tour-ai-datacenter]] [[2026-01-08-stratechery-interview-power-for-ai]] [[2024-11-08-moonshots-ep129-elon-predictions-ai-energy]]