TerraPower

Phase A — Understand the business

Company Overview

TerraPower is a privately held advanced-nuclear developer founded in 2008 by Bill Gates and Nathan Myhrvold (with Dr. John Gilleland) out of Intellectual Ventures, headquartered in Bellevue, WA. It is best understood as three businesses stacked on one nuclear-physics platform, not a single reactor bet:

1. Natrium (the flagship). A 345 MWe sodium-cooled fast reactor (SFR) bolted to a molten-salt thermal-energy-storage (TES) tank. The reactor runs flat at 345 MWe; the salt battery lets grid output flex up to 500 MWe for 5.5+ hours on demand — a hybrid of baseload nuclear and dispatchable storage that is genuinely differentiated from every light-water SMR. First unit: Kemmerer Unit 1, Wyoming, on the site of PacifiCorp's retiring Naughton coal plant (existing grid interconnection + a coal workforce to re-skill).
2. Molten Chloride Fast Reactor (MCFR). A distinct second reactor line, run as the Molten Chloride Reactor Experiment (MCRE) with Southern Company, CORE POWER, and DOE at Idaho National Lab. INL produced the world's first molten-chloride fast-reactor fuel in Dec 2025; operations target ~2030 under a 5-year, $170M cost-share; a lead application is maritime propulsion (via CORE POWER). This is real optionality most TerraPower coverage ignores.
3. TerraPower Isotopes (TPI). The only unit shipping product today: commercial-scale Actinium-225 (a scarce alpha-emitter for targeted radiopharmaceuticals — prostate, breast, neuroendocrine cancers). Commercial supply since ~Oct 2024, weekly production runs, a June 2025 supply agreement with Ratio Therapeutics, and a $450M, 250,000-sq-ft Philadelphia production facility announced Mar 2026.

Contract structure. On the Natrium side, revenue is not yet commercial — it is cost-share and pre-payment: DOE's Advanced Reactor Demonstration Program (ARDP) funds up to $2B on a 50/50 match, and Meta pre-funds deployment of its units (Jan 2026 deal, below). The economics of a repeat Natrium — a take-or-pay PPA at a defensible LCOE — do not yet exist; they are the thing Kemmerer must prove.

Supply Chain

Upstream inputs → TerraPower → end customer, named at every link (the lens fails if it stays generic):

• Enriched fuel (the binding chokepoint). Natrium runs on HALEU (high-assay low-enriched uranium, 5–20% U-235). The first core alone needs ~15–20 metric tons. Suppliers being assembled: Centrus Energy (Piketon, OH — delivered 900 kg to DOE Jun 2025; DOE awarded ACO a $900M task order Jan 2026 for commercial HALEU), Framatome (a HALEU metallization pilot line with TerraPower in Richland, WA), and ASP Isotopes (planned enrichment in South Africa). Single-source dependency, historic: TerraPower's original fuel came from Russia's Tenex; cutting that after the 2022 invasion cost the project ~2 years (first power 2028 → 2030).
• Reactor design & nuclear island. Natrium is co-developed with GE Vernova Hitachi Nuclear Energy.
• EPC / construction. Bechtel is the engineering-procurement-construction contractor; it is mobilizing field teams at Kemmerer and building a full-scope sodium Test-and-Fill Facility and operator training center.
• Heavy manufacturing. HD Hyundai (also an equity investor) has a manufacturing collaboration to fabricate key reactor components and scale the global supply chain; KBR signed a global deployment alliance for PM/EPC support.
• Molten-salt / storage. The chloride-salt system for MCRE is built with Southern Company Services and INL — described by DOE as the world's largest chloride-salt system.
• Site / grid / utility. PacifiCorp (the Naughton site owner) and the State of Wyoming; expansion siting under study with the Utah Office of Energy Development + Flagship Companies.
• End customers. Hyperscalers (Meta, Sabey Data Centers, Prometheus Hyperscale) and, ultimately, regulated utilities.

Verdict on the chain: the fabrication and EPC links are strengthening (Bechtel + HD Hyundai + GE Vernova is a credible industrial spine). The fuel link is the whole ballgame — one Natrium first core (~15–20 MT) is on the order of the entire near-term US HALEU pool DOE has committed to make available (21 MT by Jun 2026). HALEU is not a line item; it is the gate.

Competitive Advantages (moats)

• Regulatory first-mover — real and now banked. On 2026-03-04 the NRC issued Kemmerer Unit 1 the first construction permit for a commercial-scale advanced (non-light-water) reactor in US history; the safety review finished Dec 2025, ahead of schedule and 11% under budget, and full construction began 2026-04-23. In a field where NuScale has a certified design but no built plant and Oklo/X-energy are still pre-permit on power reactors, TerraPower is the only one pouring structural concrete on a utility-scale unit. That is a licensing moat measured in years.
• Product differentiation (dispatchability). The molten-salt TES makes Natrium a load-following nuclear plant — it can sell both firm baseload and peaking capacity from one asset. For a data-center buyer that needs 24/7 firm power plus headroom for demand spikes, that is a materially better product than a flat-output SMR.
• Backing / balance-sheet moat. Bill Gates as chairman and >$1B personal committed, plus tier-1 strategics (NVIDIA/NVentures, HD Hyundai, SK) and DOE cost-share, buys the one thing first-of-a-kind nuclear needs most: patient, deep capital to survive schedule slips.
• Under-priced moat — isotopes. TPI's Ac-225 is a scarce, high-margin, already-commercial medical isotope with a structural supply shortage; it is a genuine cash-and-strategic asset unrelated to the reactor timeline.

Bargaining power: weak vs. its fuel suppliers (it needs HALEU more than Centrus needs any single customer), improving vs. its offtakers (Meta/Sabey chose Natrium's dispatchability), and strong vs. its site/utility partners (Kemmerer is bespoke to it). Durable moats: process/licensing know-how (the NRC docket is a 2-year head start no rival can copy) and capital access. Fragile moats: cost — none of the above matters if the LCOE doesn't pencil (Lens 12/13).

Segments

No our figures exists (private, pre-revenue on power) — so this is a structural breakdown, not a financial one; every figure ``, unaudited:

One widely repeated third-party figure claims "TerraPower's revenue is $369.1M" — treat with suspicion: that same source lists a stale 2028 first-power date, and on a pre-commercial reactor developer such a figure can only be DOE cost-share reimbursement plus isotope sales, not power revenue. Flag as not sourced to a primary disclosure. The honest read: power revenue is zero and years away; the only recurring commercial cash is isotopes.

---

Phase B — Measure performance

(+private overlay: Lens 5 → Funding & valuation trajectory; Lens 7 → Cap table & comps; Lens 8 → funding/product events; plus a Traction sub-lens.)

Funding & Valuation Trajectory

• 2020 — DOE ARDP award: up to $2B, 50/50 cost-share on Natrium.
• Late 2022 — $830M private round; valuation ~$3.8B. Same window: the HALEU-driven 2-year delay announced.
• 2025-06-18 — $650M raise closed, new investor NVentures (NVIDIA's VC arm) alongside Gates and HD Hyundai; valuation undisclosed (UBS placement agent). Cumulative private capital >$1.4B; with DOE cost-share, >$3.4B committed to Natrium.
• Gates personally: >$1B.

Read: the syndicate is a bullish tell — an NVIDIA strategic entry in 2025 signals the AI-power thesis is pulling nuclear capital, and HD Hyundai is a supply-chain-plus-balance-sheet investor. But the undisclosed 2025 valuation is a caution flag: a step-up from $3.8B would normally be trumpeted; silence is consistent with a flat or lightly-up round in a capital-hungry FOAK build. Unverifiable — labeled ``.

Founder / Management Communication (sentiment trend)

No earnings calls (private). Substituting founder/exec communication:

• Bill Gates has moved from pure advocacy to candid cost admission — the single most important tonal shift: he has conceded Natrium's cost estimate has risen sharply (Lens 11).
• CEO Chris Levesque frames 2026 as the "scaling up" phase — permit-in-hand, construction started, pivoting the story from "can it be licensed?" to "can it be replicated?".
• Recurring phrases: "first commercial-scale," "dispatchable," "on schedule for 2030." Stopped saying: "2028," and the earlier implication that HALEU was solved.

Tone trajectory: more confident on licensing/construction, notably more guarded on cost and fuel. That divergence is the honest signal.

Cap Table & Comparables

No public multiples exist for TerraPower (private). The useful comp set is the public/near-public SMR cohort — the market's read on what this space is worth:

EV/Sales, EV/EBIT, P/E, dividend yield, 5-yr ROE: n/a for the pre-revenue names (and meaningless — the whole cohort trades on option value, not earnings). The signal in the table: the market pays ~$11B for Oklo with no permit while TerraPower — the only one actually building — last priced at $3.8B private. If SMR sentiment holds, a TerraPower IPO would likely price well above its 2022 mark; if it cracks (Lens 13), the whole cohort re-rates together.

Milestone Catalysts (the events that move a private nuclear name)

The "price-move" analogue for a private is the milestone that re-rates the next round / IPO. Five-year pattern:

• 2020 — ARDP $2B award (thesis validated by DOE).
• 2022 — $830M @ $3.8B and the HALEU delay (the market learned fuel is the constraint).
• 2024 (Jun) — non-nuclear groundbreaking at Kemmerer.
• 2025-01-21 / 2025-03 — Sabey Data Centers MOU (Rocky Mountain + Texas) and Prometheus Hyperscale — the data-center offtake thesis begins.
• 2025-06 — $650M raise + NVIDIA entry (AI-power capital arrives).
• 2025-12 — NRC safety review complete (11% under budget) + first-ever molten-chloride fuel at INL.
• 2026-01-09 — Meta deal (below) — the marquee offtake.
• 2026-03-04 — NRC construction permit granted — the single biggest de-risking event in the company's history.
• 2026-04-23 — construction officially begins.

Pattern: this name re-rates on regulatory milestones and marquee offtakes, and de-rates on fuel/cost news. The next re-raters: NRC operating-license progress, a secured HALEU contract for the first core, a second confirmed site (Utah), and any move toward an IPO/S-1.

Traction & Unit Economics (Phase B, +private add)

• Offtake pipeline (all, non-binding to varying degrees): Meta — up to 8 Natrium plants, 2.8 GW baseload / 4 GW with storage; 2 units (~690 MW firm) as early as 2032, 6 more (2.1 GW) by 2035, with Meta funding deployment. Sabey and Prometheus Hyperscale MOUs. Meta's TerraPower slice sits inside a broader up-to-6.6 GW Meta nuclear push also spanning Vistra and Oklo.
• Unit economics: the LCOE is unproven and the FOAK $/kW has ballooned (Lens 11). The isotopes unit is the only line with real gross margin today (undisclosed).

---

Phase C — Judge people & books

Management

• Chris Levesque (President & CEO) — joined 2015 (president), 30+ years in nuclear, previously led new-reactor build efforts at Westinghouse and AREVA. Directly relevant experience; also carries the scar tissue of an industry infamous for overruns. Track record at TerraPower: shepherded the first US commercial advanced-reactor construction permit — a real, hard-won delivery.
• Bill Gates (Founder & Chairman) — >$1B personal capital, and the reason the company can absorb delays that would kill a normal startup. Skin in the game is unambiguous.
• Nathan Myhrvold (Vice Chairman) and Dr. John Gilleland (co-founder).
• Capital allocation: disciplined enough to finish the NRC review 11% under budget — a genuinely good data point in a field defined by overruns. Against that: the top-line cost estimate has tripled on their watch (Lens 11).
• Founder-vs-professional-manager: a founder-chaired, professional-CEO structure — arguably ideal for a capital-intensive, decade-long build (patient founder capital + operational nuclear hand).
• Red flag — governance/reputational. A 2026 report surfaced "tough questions about Gates, Myhrvold and Epstein ties" at the leadership level. This is a reputational/governance overhang, not a legal-enforcement finding (see Lens 10); for a company that will eventually seek public-market capital and federal support, it is a non-zero headline risk worth tracking, not dismissing.

Forensic Red Flags

No income statement, balance sheet, or cash-flow statement is public — so classic forensic accounting analysis is not applicable; the analogous private-company risks:

• No audited financials. Everything here is management/press-sourced and unaudited. The undisclosed 2025 valuation is the closest thing to a "quality-of-disclosure" flag.
• Cost credibility. The gap between the legacy "~$4B project" framing and the implied ~$10B FOAK from Gates' own $/kW admission (Lens 11) is the single most important reconciliation item — surfaced explicitly, not smoothed over.
• Fuel-supply contingency. Whether a binding HALEU contract for the first core exists is not publicly confirmed; the supplier list (Centrus/Framatome/ASP) is a set of agreements and pilots, not a delivered core.

Regulatory findings (required sub-section):

• SEC (EDGAR LR + AAER): None — TerraPower has no CIK, is private, and is not required to file. No EDGAR enforcement search is possible.
• Non-SEC enforcement (FTC/DOJ/FDA/NRC/consent-decree web search): no material enforcement action, fine, or consent decree found against TerraPower. The NRC relationship is constructive — it just granted the construction permit — not adversarial.
• 10-K Item 3 (Legal Proceedings): n/a — no 10-KForm 10-KA company’s audited annual report to the US regulator. The most complete thing it publishes. exists (private).
• Net: No material regulatory or legal enforcement findings — verified via SEC EDGAR (no CIK), web search, and the regulatory-findings file as of 2026-07-10. The only open governance item is the reputational Gates/Myhrvold/Epstein-ties reporting (Lens 9), which is press scrutiny, not enforcement.

---

Phase D — Project & stress-test

IPO-Readiness & Path-to-Tradeable (+private Lens 11)

TerraPower is not in research/private-watch.json — so this is web-grounded judgment, and an open item to backfill into that ledger so our model shows it dossier-warm.

• Stage: late-stage / growth, not pre-IPO-imminent. There is no S-1, no secondary-active program, no stated IPO timeline — unlike X-energy (draft S-1 Mar 2026).
• IPO-readiness: ~3/5 (late-stage private with tier-1 crossover-adjacent backers — NVIDIA/NVentures, HD Hyundai, SK — but no filing and a pre-commercial P&L that is hard to take public before first power). The NVIDIA entry is the closest thing to a crossover tell.
• Milestones that would unlock an S-1 / tradeable event: (1) a secured HALEU supply for the first core; (2) construction cost tracking to plan through 2027–28; (3) a binding, priced PPA (vs. today's MOUs); (4) sector-window — a successful X-energy IPO would pull TerraPower's comp higher and could pull it public.
• Estimated window: an IPO is more plausibly a 2027–2029 event (aligned to construction de-risking and the SMR-IPO wave) than a 2026 one.
• Cost — the number that governs everything. Original DOE/TerraPower framing: ~$4B project. But per Gates' own admission, Natrium's cost has risen from a touted ~$11,594/kW to ~$28,986/kW — a ~150% increase. At $28,986/kW × 345 MWe that is ~$10B for the FOAK unit — a conflict with the legacy $4B that must be resolved in TerraPower's favor (via steep NOAK cost-down) for the utility/data-center thesis to hold. This is the whole investment question.

(No our model create in unattended/watchlist mode, and no EPS line exists for a pre-revenue private — the tracked binary, when this graduates to a call, is "Kemmerer Unit 1 reaches first power by YE2031" and "TerraPower discloses a Natrium NOAK target ≤ $X/kW.")

Bull vs Bear

Bull. TerraPower is the only advanced-reactor developer with a utility-scale unit actually under construction — a licensing lead measured in years over NuScale/Oklo/X-energy. Its dispatchable (reactor-plus-battery) product is the best-fit nuclear asset for AI data centers, which is exactly why Meta and NVIDIA showed up in 2025–26. It has >$3.4B committed, a patient billionaire chairman, an industrial spine (Bechtel + GE Vernova + HD Hyundai), a second reactor line (molten chloride, maritime) as free optionality, and a quietly valuable, already-commercial Ac-225 isotopes business the market isn't pricing. If Kemmerer proves even a credible path to NOAK cost-down, TerraPower re-rates hard above its $3.8B private mark.

Bear (permanent-impairment risks). (1) Cost. A tripled FOAK $/kW (~$10B implied) with an unproven NOAK curve is the Vogtle risk in a new package — $35B and 7 years late is the cautionary comp for how new US nuclear actually goes. (2) Fuel. One first core (~15–20 MT HALEU) approaches the entire near-term national HALEU supply; if domestic enrichment slips, first power slips again — as it already did once. (3) Offtakes are options, not contracts. Meta/Sabey are MOUs and pre-funding frameworks; a data-center power buyer can walk if the schedule or price moves.

Pre-mortem (18 months out, thesis broken): it's early 2028, Kemmerer construction has hit a first-of-a-kind cost/schedule wall, the HALEU contract for the first core still isn't delivered, Meta has quietly re-weighted toward its Vistra/Oklo tranches, and the undisclosed 2025 valuation turns out to have been flat — the IPO window closes and the next raise is a down round.

Are multiples too high? There is no TerraPower multiple. But the cohort (Oklo ~$11B, no permit) prices deep option value; TerraPower is the cheapest way to own the most-de-risked asset — which is either a mispricing or a warning that the private mark is stale.

Contrarian view (what the market refuses to see): the consensus story is "AI needs power → nuclear → buy the SMRs." The thing under-appreciated is that TerraPower's isotopes business may de-risk the equity independent of the reactor timeline, and that the permit — not the offtakes — is the asset: it is the one thing no competitor can replicate on any timeline.

Devil's Advocate (short-seller)

Dismantling the bull case: The permit is not a product. A construction permit means TerraPower is allowed to spend $10B finding out whether it can build this economically — Vogtle also had every permit. Revenue concentration is theoretical and lumpy: the "customers" are MOUs (Meta/Sabey) that convert only if the delivered price beats gas-plus-CCS and grid firm power in 2030–32 — a bar a tripled-cost FOAK may not clear. The moat may be weaker than bulls think: GE Vernova (a Natrium co-developer) also sells the light-water BWRX-300, and Kairos already has Google's money and a construction permit of its own — TerraPower's lead is in one technology lane, and sodium's "ignites on contact with air/water" handling burden is a permanent cost penalty. Worst capital-allocation/governance signal: the undisclosed 2025 valuation plus the Gates/Myhrvold/Epstein-ties reporting are exactly the disclosure-quality and headline-risk items a short would press. What must hold for today's ~$3.8B mark: that FOAK cost-down to a competitive NOAK is real, that HALEU shows up on time, and that hyperscaler MOUs become priced PPAs. If growth/timeline disappoints 20–30% (first power slips to 2033+, one more HALEU delay), the private mark is stale and the next raise is down. Single permanent-impairment scenario: a first-of-a-kind sodium-systems failure or cost blowout at Kemmerer that makes NOAK uneconomic — plausible enough (this is genuinely unproven at scale in the West) to keep this a watch, not a buy.

Fifteen Questions for the CEO (ordered by information value)

1. What is your committed NOAK cost target in $/kW, and what is the specific engineering path from the FOAK figure Bill Gates has cited (~$28,986/kW) down to it?
2. Do you have a binding, delivered-quantity HALEU contract for Kemmerer Unit 1's first core (~15–20 MT), or only framework agreements — and what is the latest date fuel can arrive without moving first power?
3. What is the firm, priced LCOE you can offer a data-center offtaker in 2030–32, and how does it compare to gas + firm grid power?
4. Which of the Meta/Sabey/Prometheus arrangements are binding PPAs vs. MOUs, and what conditions convert them?
5. What was the valuation of the June 2025 round, and why was it not disclosed?
6. Vogtle ran $35B and 7 years late — what specifically about Natrium's design/EPC model breaks that pattern, and what's your contingency if Unit 1 costs run 50% over?
7. How much of the Meta funding is at risk if the 2032 delivery window slips, and what are the penalty/exit terms?
8. What is the standalone value and capital plan for TerraPower Isotopes — could it be financed or spun out independently of the reactor business?
9. What is your path to positive project-level cash flow, and how many more equity rounds does the reactor business need before first power?
10. Is an IPO on the roadmap, and what milestones (permit, cost, PPA) gate an S-1?
11. How do you manage the sodium-handling safety/cost burden across a fleet, and what does it add per kW vs. a light-water SMR?
12. What is the realistic commercial timeline for the molten-chloride (MCFR) line, and is it a business or a science project?
13. How exposed is your fuel and component supply chain to policy/tariff shifts (HALEU import rules, HD Hyundai/Korea sourcing)?
14. How do you address the governance questions raised about the founders' ties, given future reliance on federal and public-market capital?
15. What is the second confirmed Natrium site after Kemmerer, and when does it start?

---