• The Nuclear Renaissance: Atoms for Algorithms
  • The Scale of the Problem
  • The Dirty Bridge: Gas Now, Nuclear Later
  • The Ratepayer Rebellion
  • The Nuclear Portfolio
    • Existing Plant Restarts
    • Advanced Reactors (Gen IV)
    • SMR Landscape
    • Fusion (Longer Horizon)
  • Bitcoin Miners: The Original Nuclear-Compute Pioneers
  • The Timeline Gap
  • Global Race
  • My Take
  • Connections
  • Sources

§The Nuclear Renaissance: Atoms for Algorithms

#technology #energy #AI #nuclear #bitcoin

[!abstract] Summary AI’s insatiable power hunger is triggering a genuine nuclear renaissance — but the timeline gap between demand (now) and nuclear supply (2030+) is being filled with natural gas, creating an environmental justice crisis. Meanwhile, Bitcoin miners quietly pioneered the nuclear-compute colocation model years before Big Tech caught on.

§The Scale of the Problem

The numbers are staggering and deserve to be stated plainly:

• $600+ billion in capital expenditure announced by Amazon, Microsoft, Meta, and Google — much of it for data centers
• U.S. power demand will hit record 4,260 billion kWh in 2026 and 4,388 billion kWh in 2027 (EIA, March 2026)
• By 2028, data centers could consume 580 TWh — 12% of total U.S. electricity demand (DOE estimate)
• By 2030, U.S. data center consumption will exceed all heavy industry combined — cement, steel, chemicals, cars (IEA)
• PJM capacity prices hit a record ceiling of $333.44/MW-day — a 12x increase in two years
• Residential electricity prices up 36% since 2020: 12.76¢ → 17.44¢/kWh, headed to 19.01¢ by September 2027

OpenAI alone has announced plans for facilities requiring 30+ GW of power — more than all of New England’s peak demand. xAI’s Colossus complex in Memphis would consume as much electricity annually as twice the city of Seattle. These are, as Princeton climate modeler Jesse Jenkins put it, “the largest single points of consumption of electricity in history.”

§The Dirty Bridge: Gas Now, Nuclear Later

Here’s the uncomfortable truth the industry doesn’t want to headline: the market has converged on “add gas now, add nuclear later.”

Sam Altman has said it explicitly: “Short-term: natural gas.” And that’s exactly what’s happening:

• xAI’s Colossus in Memphis installed up to 35 unpermitted natural gas turbines — railcar-sized engines that are major smog sources. The Southern Environmental Law Center obtained thermal imagery showing the illegal operation. The NAACP sued on behalf of the largely African American community. Residents reported noise, air quality issues, and visible soot. The facility could emit 6 million tons of greenhouse gases and 1,300 tons of health-harming pollutants per year.
• Louisiana is building three new natural-gas plants specifically for a Meta data center
• Coal plant lifespans are being extended to power new data centers
• The IEA estimates data center emissions could more than double by 2030

The xAI story is particularly damning: a company promising to build god is literally poisoning a Black neighborhood in Memphis to do it. Mississippi eventually granted permits (after the fact), but the EPA revised its rules to require permits even for “temporary” large turbines — a direct response to xAI’s regulatory arbitrage.

[!warning] Environmental Justice The pattern is clear: energy-intensive compute infrastructure gets sited in low-income, often minority communities that lack the political power to resist. This is the same playbook as heavy industry, just with “AI” branding.

§The Ratepayer Rebellion

The backlash has reached the White House. During his February 2026 State of the Union, Trump announced the Ratepayer Protection Pledge — requiring Big Tech AI developers to fund their own energy usage and cover costs of new power generation so they don’t get passed to households.

Signatories include Amazon, Google, Microsoft, Meta, and Anthropic. Key commitments:

• Cover all costs of new power generation required for their data centers
• Protect consumers from electricity price hikes
• Lower electricity costs in the long term

But as Counterpoint Research’s Marc Einstein noted: “The problem is, the industry’s not making money, so that puts even more pressure on them.” These pledges are essentially unfunded promises from companies burning cash on AI infrastructure that hasn’t proven its ROI.

§The Nuclear Portfolio

Despite the gas bridge, the nuclear buildout is real — just slower than headlines suggest:

§Existing Plant Restarts

• Three Mile Island Unit 1 → Microsoft/Constellation: NRC review underway, 2027 target. Ironic symbolism: the site of America’s worst nuclear accident becoming its AI power source.
• Duane Arnold → Google/NextEra: Iowa restart planned
• Beaver Valley → Meta/Vistra: Pennsylvania PPA signed January 2026
• Palisades → Michigan restart with DOE loan (first-ever U.S. reactor un-decommissioning)

§Advanced Reactors (Gen IV)

• TerraPower Natrium (345 MW, sodium-cooled): NRC construction permit granted March 4, 2026 — first commercial nuclear construction permit in nearly a decade. Kemmerer, Wyoming, on the site of a retiring coal plant. $4B project, DOE + Gates ($1B) + Nvidia-backed. Molten sodium thermal storage enables load-following. Backed by GE Vernova Hitachi.
• Kairos Power Hermes (fluoride salt-cooled): Demonstration reactor at Oak Ridge. Hermes 2 (50 MW) targeted for 2030 — first PPA signed between a U.S. utility (TVA) and a Gen IV SMR, for Google data centers. ORNL partnership for TRISO fuel manufacturing. DOE committed $303M via milestone-based funding (COTS model from NASA).
• Oklo Aurora (microreactor): Idaho National Laboratory demonstration
• X-energy Xe-100 (80 MW pebble-bed HTGR): Amazon-backed, Dow Chemical partnership for industrial heat

§SMR Landscape

• NuScale (77 MW PWR per module): Only NRC-certified SMR design, but the UAMPS Carbon Free Power Project was canceled in 2023 due to cost overruns. Pivoting to data center clients.
• Rolls-Royce SMR (470 MW): UK-designed, exploring Czech deployment (3 GW potential)
• EDF Nuward (~400 MW): Conceptual design completion by mid-2026, commercialization in 2030s
• EU committed €200M through InvestEU to accelerate SMR deployment through 2028

§Fusion (Longer Horizon)

• Helion Energy Polaris: Switched to deuterium-tritium shots in January 2026, achieved 150 million °C D-T fusion on February 13, 2026. Has a PPA with Microsoft. Still pre-net-energy.
• Commonwealth Fusion Systems: SPARC tokamak under construction, targeting 2027 first plasma
• TAE Technologies: Merging with TMTG, expected mid-2026 — first publicly tradeable fusion pure-play

§Bitcoin Miners: The Original Nuclear-Compute Pioneers

Here’s a detail that deserves more attention: Bitcoin miners figured out nuclear colocation years before Big Tech.

• TeraWulf formed a joint venture with Talen Energy in 2021 to build the Nautilus Cryptomine next to the Susquehanna nuclear plant in Pennsylvania — drawing electricity directly from the reactor.
• Nuclear’s share of Bitcoin mining rose from 4% (2021) → 9% (2022) → ~10% (2025). Sustainable sources overall now account for 52.4% of Bitcoin mining electricity (Cambridge Centre for Alternative Finance, 2025).
• Bitcoin miners recognized that nuclear power — once viewed as declining — could become critical infrastructure for high-performance computing. They were right, and now AI hyperscalers are following the same playbook at 100x scale.

The convergence is natural: both Bitcoin mining and AI inference are location-flexible, energy-intensive, 24/7 loads — exactly what nuclear baseload is designed for. The difference is that Bitcoin miners can ramp down instantly (acting as grid balancing), while AI inference demands uninterruptible power.

This connects directly to the agentic economy thesis: as agents proliferate and inference demand scales, the energy infrastructure underneath becomes the binding constraint. Whoever controls reliable baseload power controls the compute layer.

§The Timeline Gap

Here’s what concerns me most: the decade-long mismatch between demand and supply.

That’s 5-10 years of natural gas filling the gap. The “nuclear renaissance” is real, but it’s a 2030s story being sold as a 2026 solution. In the meantime, communities bear the cost of dirty bridge fuel while hyperscalers make clean energy pledges for the future.

The LA Times nailed it: “A global nuclear renaissance will be heavily dependent on government support and the willingness of technology giants to pay a premium to ensure continuous supply to their data centers.” Solar and wind are now the cheapest new generation in most markets. Nuclear has to compete not on cost but on reliability, density, and 24/7 availability.

§Global Race

• U.S.: $900M allocated for SMR projects, TerraPower permit, aggressive deregulation under Trump
• EU: €200M InvestEU commitment, but no SMR construction licenses granted yet
• UK: Rolls-Royce SMR with £385M government fund
• Canada: $3B CAD investment in SMR development
• China & Russia: Already operating SMRs (HTR-PM, RITM-200)
• France: Rassemblement National backing Bitcoin mining with nuclear energy (the politics of this are fascinating)

§My Take

The nuclear renaissance is real, necessary, and happening too slowly.

The fundamental tension: AI companies need power NOW but nuclear takes a DECADE. This means the actual near-term outcome is more fossil fuels, higher electricity prices for everyone, and environmental justice violations — all dressed up in nuclear optimism.

TerraPower’s construction permit is genuinely historic. A sodium-cooled reactor with thermal storage, built on a coal plant site, backed by Gates and Nvidia — this is the right template. But it’s one reactor, and it won’t produce power until ~2030.

The Bitcoin mining angle is underappreciated. Miners proved that nuclear-compute colocation works at scale. They’re now pivoting to AI hosting (Core Scientific, etc.), creating a bridge between the two worlds. The 52.4% sustainable energy share in Bitcoin mining undercuts the “Bitcoin wastes energy” narrative and positions mining infrastructure as pre-built clean compute capacity.

What worries me: the Ratepayer Protection Pledge is performative. Hyperscalers haven’t proven AI profitability, and unfunded promises to cover energy costs are just PR until the revenue materializes. If the SaaSpocalypse is real and enterprise software revenue collapses, who pays for the reactors?

The most likely outcome: a messy, decades-long transition where nuclear grows meaningfully but never fast enough, gas fills the gaps, communities fight back, and the compute layer concentrates around whoever secured power contracts earliest. Energy sovereignty — whether for nations, companies, or individuals — becomes the defining constraint of the AI era.

§Connections

• The Agentic Economy - SaaSpocalypse and the Rise of Micro-Firms — energy as the binding constraint on agent proliferation
• Bitcoin eCash - Cashu and Fedimint — Bitcoin mining’s energy infrastructure as AI compute substrate
• The Sovereign Stack - Self-Hosting in 2026 — energy independence as sovereignty prerequisite
• RISC-V - The Open Silicon Revolution — hardware efficiency as the other side of the energy equation
• The Local AI Inflection - Sovereign Inference in 2026 — local inference as energy-efficient alternative to hyperscaler data centers

§Sources

• The Atlantic, “Inside the Dirty, Dystopian World of AI Data Centers,” April 2026 issue
• CNBC, “Who is really footing the AI energy bill?,” March 13, 2026
• Reuters, “US power use to beat record highs in 2026 and 2027,” March 10, 2026
• TechCrunch, “Bill Gates’ TerraPower gets approval to build new nuclear reactor,” March 6, 2026
• White House, “Ratepayer Protection Pledge,” March 2026
• Data Center Dynamics, “TVA signs 50MW PPA with Kairos for Google,” March 2026
• Southern Environmental Law Center, “xAI built an illegal power plant,” March 2026
• Cambridge Centre for Alternative Finance, Bitcoin Mining Industry Report, 2025
• EU Commission, SMR Strategy, March 2026

Research note by Fromack — 2026-03-17