Forward-looking commentary for a general audience, not personalised advice. Prices as-of US close 4 Sep 2026; research as-of 7 Sep 2026. Horizon: 2–5 years. Voice refresh 17 Sep 2026 (view unchanged).
The market is pricing “AI needs nuclear” as if every press release were a power plant. Hyperscalers will buy firm carbon-free megawatts — that part is real. The catch is simpler: only a few issuers can deliver those megawatts this decade on assets that already exist. Everything else is a licensing and financing option with a customer logo attached.
The demand is real. The plants are not interchangeable.
IEA’s Energy and AI Base Case has global data-centre electricity around 415 TWh in 2024 (~1.5% of world power), heading toward ~945 TWh by 2030 (~3%), with the US taking the largest share of the increase.
Near-term AI load still mostly meets gas, renewables, and the existing grid. Nuclear shows up as firm, clean attributes — either through restart / life-extension / uprate PPAs on operating LWRs, or through advanced reactors and SMRs whose first commercial Western fleets are still described as late-2020s / ~2030+.
Policy has been moving: ADVANCE Act, EO 14300 (May 2025), NRC Part 53 finalized Mar 2026, proposed Part 57 for microreactors. HALEU remains a disclosed bottleneck for many non-LWR designs (Reuters, Mar 2025). Secondary mid-2026 trackers put hyperscaler nuclear commitments around ~9.8 GW across disclosed deals — useful as a map, useless as a substitute for reading whether each line is a binding PPA or an MOU.
Three very different businesses
Constellation (CEG) is the largest US competitive nuclear fleet plus retail, ~22 GW of nuclear and ~55 GW total capacity post-Calpine. It sells firm MWh and attributes under long PPAs — cash earnings today.
Oklo (OKLO) designs Aurora sodium fast reactors (~15–75 MWe), aims to build-own-operate, and is adding fuel recycle and isotopes. First-of-a-kind is under construction at INL; Meta is backing an Ohio campus plan. You are buying an option on FOAK plus a customer-funded campus.
Nano Nuclear (NNE) is developing KRONOS MMR (HTGR / TRISO) plus other concepts, with STS fuel logistics bolted on. The NRC has docketed a construction permit application for a University of Illinois demo. Commercially, most of the GW talk is still MOUs.
They do not compete with the same people. Oklo and NNE sit with NuScale, Kairos, X-energy, TerraPower. CEG sits with Vistra, Talen, NRG — and with gas peakers for near-term data-center load.
What the numbers say (and what they don’t)
OKLO and NNE are loss-making with near-zero reactor revenue — P/S is not an economic multiple. CEG is an operating generator; use earnings and EV/EBITDA. Prices below are 4 Sep 2026 close.
Oklo closed at $41.27 (~$7.7B mkt cap, ~$5.0B EV, 186M shares). Q2 brought a first $1.21M of product/services revenue against a TTM net loss of ~$153M and deeply negative free cash flow. Liquidity is the story that keeps the option alive: roughly $2.5–3.0B of cash and marketable securities, almost no debt. The cost of that liquidity is dilution — share count marched from ~70M in 2022–23 toward 186M, including a H1’26 ATM of 23.1M shares for ~$1.85B, plus a fresh $1.0B ATM in May. Aurora-INL broke ground 22 Sep 2025 with a 2028 target on a DOE pilot path. Meta Ohio (announced 9 Jan 2026) is up to 1.2 GW, with Meta prepaid / early funding, Phase 1 aimed around 2030 and full buildout 2034. That is not a classic operating-plant $/MWh PPA like CEG–Meta Clinton. Yahoo’s average 1y target sat near $80 with a $14–$130 range; consensus EPS still deeply negative.
Nano Nuclear closed at $17.72 (~$0.95B mkt cap). TTM revenue was a rounding error ($214k on Yahoo); STS logistics revenue exists separately and is not reactor MWh. Cash ~$580M, debt tiny, TTM loss ~$34M. Dilution has been heavy: an Oct 2025 $400M PIPE (8.5M shares), plus ATM usage and ~$373M of capacity still hanging. Short interest ~34% of float. The real regulatory step is the May 2026 CPA docketing for KRONOS at Illinois, with construction hoped for H2’27. The Tillman “~6 GW by 2040” framework (24 Aug 2026) is non-binding and contemplates warrants — dilutive if executed, not a PPA. No disclosed binding hyperscaler PPA comparable to CEG’s Microsoft or Meta deals.
Constellation closed at $298.96 (~$106B mkt cap). Q2 revenue $7.5B (+23% YoY), adjusted operating EPS $2.55 (+33.5%), TTM net income ~$3.5B, FY26 adj. EPS guide $11.50–$12.50. Nuclear ran 44,160 GWh in the quarter at a 93% capacity factor (ex-Salem/STP). The Microsoft Crane restart (~835 MW, 20-year PPA) still targets 2027 with remaining approvals. Meta Clinton is 1,121 MW from Jun 2027. Q2 also added 920 MW of long clean PPAs for later years. Trailing P/E ~28–29x, forward ~21x, EV/EBITDA ~15x. And instead of ATMs, the company has been buying back stock (~$2.2B YTD cited).
Contract quality is the whole game
Same headline — “nuclear for AI.” Different cash-flow dates.
CEG’s Microsoft and Meta deals are long PPAs on restart and operating LWR megawatts. Oklo–Meta is a customer-backed development plus prepay toward a 2030–34 campus. NNE’s GW figures are mostly non-binding frameworks.
Cash on the balance sheet is not the same as a fleet. OKLO and NNE are liquid because they raised equity — that cash is the cost of FOAK CapEx and dilution, not proof the power business is finished.
NuScale’s US460 Standard Design Approval (May 2025) is the regulatory benchmark the market underweights when it treats every SMR ticker as interchangeable. Oklo’s DOE pilot build and NNE’s CPA are real steps. Neither is an SDA. Oklo’s HALEU/recycle path adds fuel-chain risk NuScale’s LEU pitch tries to dodge.
Inside a 2–5 year window, CEG can compound contracted MWh. Disclosed schedules put material Aurora and KRONOS fleet revenue outside the early part of that window.
What seems priced in — and what would change our mind
Base case: hyperscalers keep signing firm clean PPAs with operators; SMR names raise equity and hit intermediate NRC milestones without commercial fleet revenue this side of 2028–30; gas and renewables still clear most incremental AI load.
Bull case needs Crane and Clinton on time, more GW-scale PPAs migrating to operators, Oklo INL hitting 2028 with Meta Ohio converting into a priced offtake, NNE converting CPA into paying load, Part 53/57 actually shortening calendars, and HALEU easing.
The steelmanned bear: interconnection and gas peakers absorb AI load cheaper and faster; Crane slips; FOAK costs blow out; HALEU stays scarce; OKLO and NNE keep diluting into ATMs while LOIs never become PPAs; NNE’s short interest amplifies volatility instead of validating a thesis.
Thesis breaks on a hyperscaler power CapEx pause, Crane/Clinton schedule breaks, Oklo INL slipping past 2028 without a clear path, NNE frameworks staying non-binding past CPA milestones, or serial dilution without offtake conversion. Share rotating from developers to operators (or the reverse) only changes which name — not whether AI needs firm power.
By durability of cash flows inside 2–5 years: CEG, then OKLO, then NNE. Asymmetric upside if FOAK works can invert two and three — that is a different question.
We are watching Crane approvals, Clinton’s Jun 2027 start, Oklo INL milestones and any priced Meta Ohio offtake, NNE’s 2027 safety evals and any binding offtake, HALEU headlines, dilution versus CEG buybacks, and NuScale as the SMR yardstick.
Bottom line
AI power scarcity is real. Nuclear that already spins turbines is scarce in a different way than nuclear that still needs a license. Inside a 2–5 year horizon, CEG is the cash-flowing expression of firm clean megawatts under binding PPAs; OKLO is a well-funded FOAK option with a Meta logo and heavy dilution; NNE is smaller, CPA-progressing, and still mostly non-binding on commercial GW. We would rather underwrite contracted operating megawatts than LOI slideware — and we would size developer names as options that can expire against NRC calendars and equity raises.
Primary sources
- IEA Energy and AI; Yahoo Finance
OKLO / NNE / CEG (close 4 Sep 2026)
- Oklo IR / Meta Ohio PR (9 Jan 2026); Oklo Fact Sheet (Jul 2026); NRC Oklo pages
- Nano Nuclear Q3 FY26 IR (12 Aug 2026); PIPE PR; Tillman framework (24 Aug 2026)
- Constellation Q2’26 (6 Aug 2026); Microsoft Crane PR (20 Sep 2024); Meta Clinton PR (2025); Calpine close (7 Jan 2026)
- NRC Part 53; NuScale US460 SDA (May 2025); Reuters HALEU (Mar 2025)