BasicsEvery SMR press release now arrives loaded with acronyms. NRC, GDA, HALEU, FOAK, LCOE, MWe. Skip past enough of them and you start nodding along without actually knowing what a company just promised, or how expensive that promise might get. 📚
That’s the problem this glossary solves. Ten terms, defined once, with the specific numbers and deals attached to each one so the definition sticks. Print this out. You will need it the next time a hyperscaler announces another gigawatt-scale nuclear deal.
The basics: SMR, MWe, and microreactor
Start with the term in the name. A small modular reactor is a nuclear plant that generates roughly 300 megawatts electric or less and is built largely from factory-fabricated modules, then shipped to site rather than poured and welded in place over a decade. 🏭 The modular part is the whole pitch: standardize the manufacturing, and the economics are supposed to improve with every unit built, the same way Boeing gets cheaper per plane after the first one. The International Atomic Energy Agency counts more than 90 distinct SMR designs in development worldwide, which tells you how crowded and unsettled this field still is.
MWe just means megawatts electric, the actual electricity output a plant delivers to the grid, as opposed to MWt (megawatts thermal), which measures heat output before conversion losses. Always ask which one a company is quoting. ⚡ The spread across current designs is wide:
NuScale’s uprated US460 module: 77 MWe per unit
GE Hitachi’s BWRX-300, now under construction at Darlington, Ontario: 300 MWe
TerraPower’s Natrium, breaking ground in Kemmerer, Wyoming: 345 MWe
Rolls-Royce SMR, headed toward deployment in Wales: 470 MWe
Microreactors, the smallest subcategory: typically 1 to 20 MWe
That last line introduces microreactor, a reactor so small it can be transported by truck, rail, or cargo plane. The Pentagon’s Project Pele, built with the Defense Department’s Strategic Capabilities Office, is a 1-to-5 MWe TRISO-fueled design meant to power forward operating bases. Microreactors trade raw output for portability, which is a very different value proposition than a 470 MWe grid plant.
The paperwork: design certification, GDA, and construction permits
Nuclear regulators do not license a company. They license a design, then separately license the site and the construction, and the vocabulary around each step trips up more readers than any technical term on this list. 📋
In the US, the NRC’s design certification (or the newer standard design approval, a faster variant) confirms a reactor design meets federal safety requirements and can be referenced in future license applications. NuScale’s 50 MWe module became the first SMR ever certified in January 2023, and the company’s uprated 77 MWe US460 followed with a standard design approval in May 2025. As of mid-2026, NuScale remains the only company holding either.
The UK runs a parallel process called Generic Design Assessment, or GDA, jointly overseen by the Office for Nuclear Regulation, the Environment Agency, and Natural Resources Wales. It’s a three-step, multi-year review, and Rolls-Royce SMR’s 470 MWe design entered Step 3 in July 2024, with the full 53-month process expected to wrap by the end of 2026. Passing GDA doesn’t guarantee a build gets approved either; it just clears the design itself.
Then there’s the construction permit (CP), which authorizes actually pouring concrete for a specific reactor at a specific site, distinct from the combined license (COL) that bundles construction and operating authority into one approval. Kairos Power holds the first NRC construction permit ever issued for an advanced reactor, covering its Hermes test unit in Oak Ridge, Tennessee, and TerraPower’s construction permit application for Natrium is under active NRC review right now. Three separate approvals, three separate timelines, and none of them guarantees the next one comes easily.
The fuel bottleneck: HALEU
HALEU, high-assay low-enriched uranium, is fuel enriched to between 5% and 20% U-235, well above the roughly 3% to 5% used in today’s commercial reactor fleet. 🇺🇸 More than half of the SMR designs currently in development need it, and until recently the only reliable commercial supplier was the Russian state firm Tenex.
That’s the bottleneck everyone in this sector talks about. Centrus Energy operates the only domestic HALEU production line, a demonstration cascade at the American Centrifuge Plant in Piketon, Ohio, and delivered its contracted 900 kilograms to the Department of Energy by mid-2025. The DOE followed with a $900 million HALEU enrichment award in January 2026 to scale that up. On the customer side, Centrus signed a letter of intent to supply HALEU to Oklo for up to five Aurora powerhouses at a planned 1.2 GW campus in southern Ohio, with deliveries starting in 2029. ⛓️ Every HALEU supply deal, DOE contract, and enrichment milestone we cover ends up structured, sourced, and searchable in SMRbrief Pro, which is genuinely the fastest way to track a fuel chain this fragmented.
A federal ban on Russian uranium imports takes full effect in 2028, so the fuel question isn’t theoretical. It’s the single biggest constraint standing between a certified design and an operating reactor. Which of these fuel deals do you think closes first, Oklo’s or one of the hyperscaler-backed developers’? 🔬
The money terms: FOAK, NOAK, LCOE, and offtake agreements
FOAK means first-of-a-kind, the initial build of any new reactor design, when engineers are still finalizing details during construction and the supply chain doesn’t exist yet. NOAK, nth-of-a-kind, is what happens once a design has been built enough times that manufacturing, workforce training, and permitting all get repeatable. 💰 The gap between the two is enormous and it’s the reason SMR economics are so contested.
The clearest case study is NuScale’s own Carbon Free Power Project in Idaho. Target pricing started at $58 per megawatt-hour in 2021. By late 2023, construction cost estimates had jumped 75%, from $5.3 billion to $9.3 billion, pushing the target price to $89/MWh. The project’s utility customers walked, and NuScale cancelled it. Analysts at Wood Mackenzie peg FOAK SMR costs around $180/MWh today, possibly falling closer to $100/MWh by 2030 as NOAK production kicks in, though that forecast depends entirely on someone actually building enough units to get there.
LCOE, levelized cost of electricity, is the standardized metric analysts use to compare that cost per MWh across technologies over a plant’s lifetime. Lazard’s most recent unsubsidized estimate puts conventional nuclear at $141 to $221/MWh, well above utility-scale solar or onshore wind. It’s the number every SMR developer is racing to beat, and so far only aspirational NOAK projections get there.
Which brings us to offtake agreements, sometimes structured as power purchase agreements (PPAs), the contracts that actually make a reactor bankable by locking in a buyer before a shovel hits dirt. The last eighteen months have produced more of these than the prior decade combined:
Meta and Constellation Energy: 1.1 GW, 20-year PPA, Clinton Clean Energy Center, Illinois (June 2025)
Amazon and Talen Energy: 1,920 MW through 2042, Susquehanna Steam Electric Station, Pennsylvania
Microsoft and Constellation: 835 MWe, 20-year PPA to restart Three Mile Island Unit 1 as the Crane Clean Energy Center, targeting H2 2027
Meta and TerraPower: agreement covering up to 8 Natrium reactor plants, announced January 2026
Amazon and X-energy: $50 billion partnership to deploy 960 MW of Xe-100 units in Texas
Big tech’s shift toward becoming anchor tenants for nuclear power, rather than just buying credits, is arguably the single biggest structural change in this industry’s financing since Vogtle. 📈 An offtake agreement from a company with Meta’s or Amazon’s balance sheet is exactly what lets a developer absorb FOAK cost risk that sank NuScale’s municipal utility customers.
So next time a headline drops with a number attached, you’ll know whether it’s describing thermal output or electrical, a design win or a construction win, and a first-of-its-kind gamble or a repeatable one. What’s the term you still see misused most often in nuclear coverage?