Across 18,500 reactor-years of civilian operation, nuclear power has had 2 major accidents and kills ~0.03 people per TWh. Air pollution from fossil fuels kills ~8.7 million people every year - the equivalent of 2 Chernobyl disasters every day, 365 days a year, decade after decade. But that death is slow, invisible, with no blue Cherenkov glow to put on camera. This is the physics, economics, and psychology of an industry now staging a comeback.
Scope: This piece covers nuclear energy across four layers - nuclear physics (U-235, fission, energy density), the supply chain (mining → conversion → enrichment → fabrication), power-source comparisons (CO2/kWh, deaths/TWh, capacity factor), and the SMR wave (OKLO, NuScale, TerraPower, AI power demand).
Data: IPCC AR5, WNA, IEA, IAEA "Red Book" 2024, Our World in Data (Markandya & Wilkinson), Lancet 2021 (air pollution). Spot uranium prices from UxC/TradeTech. Chernobyl death estimates use the WHO/IAEA median (~4,000 long-term).
The Arithmetic Paradox - Who Actually Kills More?
Nuclear has the blue Cherenkov glow, a disaster you can film, a 30km exclusion zone. Coal has chronic lung disease, strokes, and deaths scattered across 40 years. The human brain reacts to drama, not statistics.
After Fukushima, Germany shut down every one of its nuclear reactors. The result: it had to burn more lignite (brown coal, the dirtiest fuel in the world), and a 2022 study (Jarvis, Deschenes & Jha) estimates that decision caused an additional ~1,100 deaths per year from air pollution. Irrational fear of nuclear has killed more people than nuclear itself - even counting Chernobyl.
This is a case of "policy driven by imagery, not by integral." When you pick an energy source, you're not choosing between risk and no risk - you're choosing which risk is visible and which one isn't. The death toll doesn't discriminate.
Fuel - 1kg of Uranium = 3,500 Tonnes of Coal
Uranium's energy density isn't "a few times higher" - it's higher by a factor of millions. That's why a 1 GW nuclear plant needs only ~27 tonnes of uranium a year, while a coal plant of the same capacity needs ~3 million tonnes of coal and emits ~6 million tonnes of CO2.
U-235 vs U-238 - Two Isotopes, Two Fates
Natural uranium is only 0.7% U-235 (fissile) and 99.3% U-238 (not fissile with slow neutrons). The difference comes down to the fission barrier - something only nuclear physics can explain.
A rule of nuclear physics: isotopes with an odd neutron count fission more easily than those with an even neutron count. U-235 has 143 neutrons (odd) → fissile. U-238 has 146 neutrons (even) → not. Pu-239 has 145 neutrons (odd) → fissile. Pu-240 has 146 neutrons (even) → not. This is the "hidden constant" that decides which elements can serve as nuclear fuel.
Reserves - Enough to Last, or Running Out?
The fear that "uranium will run out like oil" is misplaced. Uranium is more abundant than silver and mercury, with ~14 million tonnes of land-based reserves, and the ocean holds an additional 4.5 billion tonnes dissolved. The question isn't "is there enough" but "which technology gets used."
Only conventional LWRs + confirmed reserves (6.1M tonnes), demand doubling to 120K tonnes/year.
LWR + MOX recycling + inferred reserves (~14M tonnes). Demand rising gradually per COP28.
Breeders extracting 99% of U-238 + seawater extraction. Total resource base: billions of tonnes.
Supply Chain - Four Stages, Extremely High Barriers
The entire commercial nuclear industry flows through 4 sequential stages, each controlled by a small group of specially licensed companies. This is the highest barrier-to-entry industry on the planet after semiconductors.
Dig ore or pump acid down wells (ISR). Grade 0.1–15%. Cost ~$10–50/lb.
KAP (Kazatomprom) · CCJ (Cameco) · UEC · Paladin · CGNTurns yellowcake into Uranium Hexafluoride gas to feed into centrifuges. Only 4 plants exist in the Western world.
Cameco (Port Hope) · Orano · ConverDyn · RosatomCentrifuges separate U-235 from U-238 (a 1.3% mass difference). Unit: SWU. SWU price went from $55 (2021) → >$180 (2025).
Centrus/LEU (US) · URENCO · Orano · Rosatom TENEXPressed into ceramic pellets, loaded into ~4m zircaloy tubes. Each rod lasts 3–6 years.
Westinghouse · Framatome · GNF · TVELNew-generation SMRs (OKLO, X-energy, TerraPower, Kairos) require HALEU (High-Assay LEU, 5–19.75%). Before 2024, the US depended on Russia for ~24% of enrichment services. The Prohibiting Russian Uranium Imports Act (08/2024) banned imports from Russia - suddenly making Centrus/LEU the only US company with an NRC license to produce HALEU. DOE has signed a $2.7B contract to expand the Piketon, Ohio plant.
The "Kingpins" Carving Up the Market
Formerly USEC. America's only HALEU plant, at Piketon. Once ran the Megatons to Megawatts program (turning 500 tonnes of Soviet warhead HEU into civilian fuel).
World's #2 uranium producer. McArthur River & Cigar Lake mines (Saskatchewan) - the richest on the planet. In 2023, with Brookfield, acquired Westinghouse ($7.9B) → vertically integrating mining + conversion + fabrication.
#1 producer (~43% globally). Uses ISR at <$15/lb. But depends on rail logistics through Russia → the war in Ukraine turned the route to the port of St. Petersburg into a geopolitical risk.
The only vertically integrated player in the West. Its La Hague reprocessing plant (Normandy) - the largest in the world - extracts plutonium and uranium from spent fuel to make MOX. France has run a closed fuel cycle since the 1960s.
A joint venture of 3 governments. The most advanced Zippe-type centrifuge technology in the West. Applying for NRC approval to produce HALEU at Eunice, New Mexico - a direct rival to Centrus, but needs 3–5 years.
Doesn't mine or enrich - just buys and stockpiles U₃O₈ (currently >60M lbs). When the fund buys on spot, it withdraws supply → pushes up price. A mechanism similar to SPDR Gold for gold.
CO2 / kWh - Nuclear Is Cleaner Than Solar
Counting the entire lifecycle (mining, construction, operation, decommissioning, waste), nuclear emits CO2 on par with or lower than wind, and cleaner than solar, thanks to its high energy density. IPCC AR5 data:
Capacity Factor - "Actual Uptime"
Case Study - France vs Germany, Two Strategies, Two Outcomes
Two neighboring countries, both in the EU, both committed to Net Zero. The only difference: France chose nuclear, Germany chose Energiewende (shut down nuclear, expand renewables). Two decades later:
- 56 reactors, ~70% of electricity from nuclear since the 1974 "Plan Messmer"
- CO2/kWh: 55g - lowest of the EU-5
- The largest net electricity exporter in the EU - selling to Germany, Italy, the UK
- Stable 24/7 baseload - no worries when the wind stops or the sky is overcast
- Household electricity ~€0.21/kWh, industrial ~€0.13/kWh
- Shut down its last reactors 04/2023 (Isar 2, Neckarwestheim 2, Emsland)
- CO2/kWh: 385g - 7× France
- Has to buy nuclear power from France when the wind dies down
- Spent >€500B on Energiewende over 20 years
- Result: burning more lignite → +1,100 deaths/year from air pollution (Jarvis et al. 2022)
Germany spent hundreds of billions of euros replacing nuclear with wind and solar. But when Dunkelflaute (windless nights) strikes for 1–2 weeks a year, Germany has to burn lignite. Result: CO2 emissions still 7× France's. Shutting down nuclear reactors doesn't mean emissions disappear - it means emissions move elsewhere, usually somewhere dirtier.
Two Disasters - Causes, Consequences, Fixes
Across 18,500 reactor-years of civilian operation, 2 incidents caused significant radioactive release: Chernobyl 1986 and Fukushima 2011. Both were tragic. But under close analysis, they reveal old design flaws and human error - not something inherent to the technology. And both flaws have since been fixed.
'86 Chernobyl · Reactor #4, Pripyat, Ukrainian SSR 04/26/1986 · 01:23 AM · INES Level 7 31 direct · ~4,000 long-term
A (ironically) safety test during an inexperienced night shift. Power dropped too low to 200 MW (plan called for 700), and operators withdrew nearly all control rods to bring it back up. When the AZ-5 emergency shutdown button was pressed, the graphite tips of the RBMK reactor's control rods actually increased the reaction in the first few seconds → power spiked to ~30,000 MW (100× design rating) → explosion. No containment structure - the exposed core burned for 10 days.
Two root causes
- Positive void coefficient - steam bubbles in water → reaction INCREASES (opposite of a PWR)
- Graphite-tipped control rods - the SCRAM button briefly boosted the reaction before shutting it down
- No containment structure - the reactor was too large to enclose in a concrete shell
- RBMK was a Cold War-era Soviet design, prioritizing military plutonium production
- ECCS (emergency core cooling system) disabled before the test
- Only 6–8 rods withdrawn instead of the minimum 30
- The RBMK's flaw had been known since 1983 but was classified - the operating engineers didn't know
- The test was moved from the experienced day shift to the night shift due to political pressure ahead of May 1st
No new RBMKs will ever be built. The 10 remaining RBMKs in Russia have been retrofitted (graphite tips removed, additional control rods added, void coefficient reversed). Containment structures became mandatory worldwide. The IAEA established the INES Scale and the OSART program. Gen III+ reactors (AP1000, EPR, APR1400, VVER-1200) have passive safety - they shut themselves down even with no power and no operators present.
'11 Fukushima Daiichi · Reactors #1, #2, #3, Japan 03/11/2011 · 14:46 JST · INES Level 7 0 direct · 1 related
The Tōhoku 9.0 earthquake - the most powerful in Japan's recorded history. The Daiichi plant (6 BWR Mark I reactors from the 1960s) SCRAMed successfully within 15 seconds. Emergency diesel generators ran normally. The plant survived the earthquake.
49 minutes later, a 14–15m tsunami (the seawall was designed for 5.7m) struck, flooding the basement level → destroying all diesel generators and backup batteries → station blackout. Decay heat (1.5% of full power) kept being generated → cooling water boiled off → zircaloy cladding exceeded 1,200°C and reacted with steam to produce hydrogen → explosions in the reactor buildings of Units #1, #3, #4. The steel containment vessels remained intact at #1 and #3. Unit #2 leaked through a pressure relief valve.
Two root causes
- The 5.7m seawall was based on the largest historically recorded tsunami
- Paleo-geological samples showed a 14m tsunami had occurred around 869 CE (Jōgan) but were disregarded
- Backup diesel generators were located in the basement → easily flooded by seawater
- TEPCO had rejected a 2008 proposal to raise the seawall as "too expensive"
- No passive cooling - required electricity to pump cooling water
- Loss of power + loss of backup = unable to cool the decay heat
- BWR Mark I containment was small - not enough volume to safely handle hydrogen
- No mandatory hydrogen recombiner (became standard only after 2011)
Gen III+ reactors (AP1000, EPR, APR1400) have 72+ hour passive cooling - a gravity-fed water tank above the core, flowing down automatically to cool the reactor when power is lost. Hydrogen recombiners are now mandatory. Japan established an independent NRA to replace NISA. Filtered containment venting removes >99.9% of radioactivity if forced venting is required. Gen IV reactors (OKLO, Natrium) use liquid sodium - no water needed, no hydrogen risk.
| Criterion | Chernobyl (1986) | Fukushima (2011) | Possible with new reactors? |
|---|---|---|---|
| Reactor type | RBMK (built only in the USSR) | BWR Mark I (GE, 1960s) | No - RBMK retired, BWR Mark I retrofitted |
| Trigger | Operational + design error | 14m tsunami (natural disaster) | Gen III+ has 72h passive cooling |
| Core | Exploded + burned 10 days | Meltdown, no core explosion | Sealed containment + filtered venting |
| Containment | NONE | Present, retained most radioactivity | Mandatory worldwide since 1986 |
| Direct radiation deaths | 31 | 0 | - |
Layered Safety - Defence-in-Depth
Every Gen III+ reactor has 5 independent radioactive barrier layers. For radioactivity to escape into the environment, all 5 layers must fail simultaneously - a probability close to zero with modern designs.
The SMR Wave - From "Build Once" to "Mass-Produce"
The biggest problem with traditional nuclear was never safety - it was economics: $10–25 billion per plant, 10–15 years to build. The Small Modular Reactor (SMR) flips the paradigm: cast in a factory, shipped to site - like a shipping container instead of a brick-by-brick build.
| Criterion | Traditional Reactor | SMR |
|---|---|---|
| Capacity | 1,000–1,600 MWe | 20–300 MWe |
| Construction cost | $10–25 billion | $1–3 billion |
| Build time | 10–15 years | 3–5 years |
| Manufacturing | Built on-site (one-off) | Prefabricated at a factory → shipped to site |
| Safety | Active cooling (requires pumps + power) | Passive cooling (self-cools via physics) |
| Refueling | Every 18–24 months | 10–20+ years (fast reactor) |
Five Players Defining the Next Decade
Sam Altman-backed · NYSE: OKLO · IPO 05/2024. A 15 MWe fast reactor cooled by liquid sodium metal (no water, no pressure). HALEU fuel, refueled roughly every 10 years. Can use waste from other reactors as fuel.
NYSE: SMR · First SMR to receive NRC standard design approval (01/2023). A scaled-down PWR at 77 MWe/module, scalable to 12 modules = 924 MWe. Each module is 23×4.6m, small enough for a truck. Self-circulates via natural convection.
Bill Gates-backed. A 345 MWe sodium reactor plus a molten-salt thermal storage system that allows power output to flex from 0–500 MWe - compensating for wind/solar in a mixed grid. Under construction at Kemmerer, Wyoming.
An 80 MWe HTGR using TRISO fuel - uranium pellets coated in 4 ceramic layers, physically incapable of melting down. Has signed a contract with Dow Chemical to supply power and heat to its Seadrift, Texas chemical plant.
Google-backed. A fluoride molten-salt reactor (FLiBe) - low pressure, no hydrogen risk. Building a demonstration reactor at Oak Ridge, Tennessee. NRC has already issued a Construction Permit.
All of the SMRs above require HALEU (5–19.75%). Aside from Centrus/LEU, no US company holds an NRC license to produce it. URENCO is applying - that takes 3–5 years. No HALEU means no SMR.
AI's Power Hunger - Why Big Tech Is Betting on Nuclear
A single ChatGPT query consumes ~10 times more electricity than a Google search. As AI scales up, data center power demand is expected to double or triple by 2030. Wind and solar aren't stable enough for data centers that need 99.999% uptime. The only answer that can scale fast enough while staying clean: nuclear.
Signed a 20-year power purchase agreement from Three Mile Island Unit 1 - a nuclear plant being restarted. Constellation Energy will restart it by 2028. Microsoft's target: carbon-negative by 2030.
Signed a power purchase agreement with Kairos Power (fluoride molten-salt SMR). Goal: 24/7 carbon-free energy for every data center - achievable only with nuclear baseload, not solar plus batteries.
Bought a data center campus adjacent to the Susquehanna plant (Pennsylvania) - connected directly to ~960 MW of nuclear power. Invested $500M in X-energy plus a deployment agreement for an SMR in Washington State.
Altman has personally invested in OKLO (fission) and Helion (fusion). Slogan: "AI needs nuclear - period." He served as OKLO's chairman before the company's IPO.
Renaissance - 22 Countries Pledge to Triple Capacity by 2050
At COP28 (Dubai, 12/2023), 22 countries including the US, France, the UK, Japan, South Korea, Canada, the UAE, and Poland signed a pledge to triple global nuclear capacity by 2050 - the first time in COP history that nuclear was formally recognized as a climate solution.
- Japan: after 10 years of shutdowns, restarting reactors - targeting 20–22% nuclear power by 2030
- Sweden: reversed its "phase-out" policy - planning new construction for the first time in 40 years
- Netherlands: 2 new Borssele-II reactors - dropped plans to shut down
- Italy: Meloni reopened debate on building nuclear plants (banned since 1987)
- Poland: building for the first time - Westinghouse's AP1000, aiming to escape Russian coal
- UAE: Barakah (South Korea's APR1400) - 4 reactors already operating
- US: the IRA allocates $30B+ for nuclear. Restarting Three Mile Island, Palisades. DOE ARDP funding for SMRs
- UK: Hinkley Point C + Sizewell C - despite delays and cost overruns
The Macro View - Three Legs, No Fourth Option
A clean power source must satisfy 4 conditions simultaneously: low CO2, runs 24/7, high output per unit of land, low death toll. IPCC/IEA/WNA data show only one source satisfies all 4 at gigawatt scale:
| Source | CO2 < 50g | 24/7 | High density | Low deaths |
|---|---|---|---|---|
| Coal | ✗ 820g | ✓ | ✓ | ✗ 24.6 |
| Natural gas | ✗ 490g | ✓ | ✓ | ~ 2.8 |
| Solar | ✓ 41g | ✗ | ✗ 36 km²/GW | ✓ 0.05 |
| Wind | ✓ 11g | ✗ | ✗ 72 km²/GW | ✓ 0.04 |
| Hydropower | ✓ 24g | ~ dry season | ✗ geographically limited | ~ 1.3 |
| Nuclear | ✓ 12g | ✓ 92% | ✓ 1 km²/GW | ✓ 0.03 |
Solar and wind excel on the "clean" leg but fail on "24/7" and "density." Coal and gas win "24/7" and "density" but fail on "clean" and "safe." Nuclear is the only choice that wins all 4 legs simultaneously - and that's why Bill Gates, Sam Altman, Google, Microsoft, France, China, South Korea, and the UAE are all betting on it, Chernobyl and Fukushima notwithstanding.
The question was never "is nuclear safe" - it's "compared to what." Against the invisible deaths of 8.7 million people a year from dirty air, against climate change threatening billions, one Chernobyl every 40 years is a very small price, quantitatively speaking. Math doesn't care what you're afraid of.
03 Discussion
Leave a note
A considered space for questions, counterpoints, and useful additions. Civil, on-topic, signed.
Reader notes
...Loading notes...