Radiant: Beyond Prosperity Mass-produced microreactors mean nuclear power anywhere. Behind the supermarket, on a military base, and eventually beyond the star we were born at. Radiant is building them. @DougBernauer @torishiv @RadiantNuclear

US ARMY PLANS UP TO $2.2 BILLION OVER FIVE YEARS FOR NUCLEAR MICROREACTORS AT FIVE MILITARY BASES TO BOOST ENERGY SECURITY. || ANTARES, BWXT, GENERAL ATOMICS, RADIANT AND WESTINGHOUSE SELECTED; PROGRAM COULD EVENTUALLY DEPLOY 20+ REACTORS, EACH GENERATING 1-20 MW.

BREAKING: @AntaresNuclear has secured a $470M Series C after achieving criticality with its Mark-0 reactor. The startup is developing nuclear microreactors for the U.S. Air Force. Paradigm and Caffeinated Capital co-led the round.

It took me a while even while working in the space to appreciate the power requirements of data centers and what it means to design power for them (in our case, nuclear power!). Easy to think that more energy, LCOE, etc. do all the work. I intuit the case on the ground as some “allowable” band of performance across a few dimensions: project risk, cost, speed, and power output. You have some leniency in the specifics of each (say, if marginal risk drops expected cost, or increases speed, you may make that tradeoff), but operators overall need a “goldilocks”: a happy medium across the four, without any “spike” (disqualifier) in risk, cost, timeline, or output. It is these spikes that defeat traditional nuclear categories as ideal (not passable in some situations, but ideal) power sources for data centers. Existing categories may be defined as microreactors, SMRs, and large plants. Microreactors “spike” in the output category: to power a 100 MW data center with 1 MW units (much less a 1 GW system!), you would need to build 100 plants. Having to build this volume of reactors cascades across the other categories (risk, timeline, cost) and invalidates the whole lot. Traditional SMRs are the best for data centers versus microreactors/large-scale plants (depending on your definition of "SMR"), but are not so small in practice: the core problem being that 10 MWe is the rough order of scale that can be manufactured in a factory and transported on normal trucks and road, without specialized (and extremely expensive) transport. So a single reactor producing 100 MW, much less 300 MW or 500 MWe, exists without the economies of scale provided by gigawatt-scale plants, but also without the speed and lack of capital risk provided by micro-scale reactors. They are still a little too hot for Goldilocks and not a “worst”, but mediocre-of-all-worlds situation. Large plants, finally, spike effectively in all categories. Even if we forget about the past two decades of American new-plant cost and timeline overruns and assume projects stay on schedule, they represent such a massive time horizon and capital investment that projects simply cannot match the rate at which data centers incrementally construct and turn on the plant, while presenting a degree of financial risk that freezes financing and insurance. Even if projects did not run over schedule and cost, they would delay data center developments by years. These being the disqualifiers for traditional nuclear, I've skipped over some important hard constraints, namely baseload and the thing actually being able to turn on. Baseload (intuitively, the variance of power output and whether it ever drops below some minimum) matters because you can't just have a data center turn off if it's, say, powered by a wind farm and the wind stops. It must be 24/7. This disqualifies or complicates lots of wind/solar projects for on-site data center developments. Implied in baseload is a high capacity factor: you can’t have the thing turn off or radically diminish its output for maintenance, refurbishing, or due to error. The other mentioned hard constraint that meshes with baseload in an interesting way is the fact that you do actually need the power. This sounds like a given, but alas, fails in the most common of solutions to just getting some power: connecting to the grid, evidenced by many a data center impeded for months (if not years) or canceled per the strain of permits, local pushback, and building grid interconnect and infrastructure. Policy like the Ratepayer Protection Act is formalizing this conclusion: data centers needing to bring their own power to sites. Combining these two hard constraints explains why fuel cells and combined cycle nat gas plants have grown incredibly valuable (love you Bloom!): one, you can readily deploy them, and two, they provide steady, 24/7 power to meet baseload. Few other generation sources fit these requirements as cleanly and nat gas has that happy medium of cost, risk, and size. Nuclear is one of the few other generation sources that meets these hard constraints, and maybe the sole source that fits them more cleanly than gas: 24/7 baseload availability, reactor and turbine redundancy, and no grid interconnect required. Its massive problem to date, going back way before data centers, has been that Goldilocks problem of risk, cost, and speed. This is functionally the approach we're taking: use the inherent advantages of nuclear power, then balance the scales. I’ll tackle them in order: Power Output. We’ve established our rough minimum and maximum bounds: anything that’s clearly a microreactor or near it is below the minimum, and anything SMR-scale (per reactor) or up is over the maximum. Instead, you want something that matches how a data center actually gets built and turned on, regardless of whether it requires 50 MW or 5 GW: incremental completion and operation of individual buildings or server halls over the course of months to years on the wider project campus. These buildings or server halls exist on a rough order of 10 MW to 100 MW. Building on the bottom end of that scale is still too inefficient with regard to capital and space, while the higher end takes too long to build. We also know the maximum power output for individual reactors to make them manufacturable in mass and transportable is roughly 10 MWe. This all produces a “Goldilocks” medium of a 50 MWe modular nuclear plant built on 10 MWe reactors. Having multiple reactors (beyond the 10 MWe constraint) solves the baseload reliability constraint, as they ensure the 50 MWe plant is never completely offline, while 50 MWe is a much better fit for the power needs of data centers. Shifts past 10 MWe per reactor or under 50 MWe as a modular unit to stack harms the economics of the happy medium. So, we get “lego blocks” of modular power units: 10 MWe reactors compiled in 50 MWe plants, with plants added in 12-month increments to match the pace of data center construction. Taking power output as the given for our first constraint, we may evaluate the others: Project Risk. With orders taking just 12 months to turn-on and incrementally scaling up to meet data centers as they turn on, project risk gets massively reduced, which eliminates those financing and insurance disqualifiers present for large-scale developments. So, our chosen output medium also stays in that qualifying bound for risk. Project Speed. The structure of the output – 10 MW reactors that can fit on trucks and be manufactured easily – is what unlocks the workable deployment speed. Going over this 10 MWe bound would induce a step-change decrease in speed and increase in cost. Power Cost. Speed is the unlock for cost: if you can build and deploy lots of modular, on-site power plants fast, the cost will be there, and schedule and cost overruns won’t. This all in mind, you get a fundamentally new category of reactor product that doesn’t fit into the microreactor, SMR, or large-scale plant categories: this is what we call the XMR, or extra modular reactor, and the 50 MWe product it unlocks in a manner that is designed from the ground-up for data centers. Reactors are sized for the factory, plants are sized for the data center. The XMR is the only nuclear product not disqualified in any way for data center usage. Now, I previously touched why natural gas is such a match for data center power needs, as it’s basically the same list as nuclear: baseload, onsite without requiring grid interconnect, and affordable, fast, and low-risk. So why use nuclear at all, and not just natural gas? A few reasons here: to point out the obvious, the supply of turbines is backed up for years. It's less of a this-or-that question because there is simply so much demand that all supply which meets customer criteria will be brought on at scale – it's the supply problem for all possible solutions. Nuclear has the direct benefit versus natural gas of being clean (both zero-carbon and no airborne pollutants), having minimal fuel price exposure, reaching a high (90%+) capacity factor, and not requiring continuous fuel lines. Net: having one big advantage as a power source is nice, but not having a disqualifying attribute is the baseline. Nuclear fits within that model, but requires a new product category: XMRs.

Congratulations to our customer, @valaratomics, on its announcement of new funding. Following its recent achievement of criticality with the Ward 250 #microreactor#, this latest milestone reflects the growing momentum behind the company's vision for scalable, resilient clean energy solutions. Valar was one of Urenco's earliest #AdvancedFuels# customers as we expanded our capabilities to support the emerging reactor technologies sector. We're proud to be strengthening our partnership and continuing to provide the fuel cycle services that will help enable its ambitious growth plans. More here: #nuclear# #energy# #AdvancedReactors# #innovation#

The next generation of nuclear energy must deliver dependable power where continuity matters most. That’s why we’re excited to announce that the @ENERGY has conditionally selected Radiant to receive a second allocation of high-assay low-enriched uranium, or HALEU. The allocation is intended to fuel Radiant’s microreactor for the Department of the Air Force’s Buckley Space Force Base in Aurora, Colorado, which will be delivered by 2028. Radiant was the sole private-sector recipient selected in DOE’s third round of HALEU allocations, alongside NASA. Read more below.

.@valaratomics just made history by becoming the first startup to power an NVIDIA Blackwell with a nuclear reactor. Watch the full conversation as @saranormous joins founder and CEO @isaiah_p_taylor at the Utah San Rafael Energy Lab, home to the Ward 250 microreactor.

🟥 Things I'm looking at today 🟥 Powered by @yeet Subscribe: Strategy buys BTC, Hyperliquid talks US entry, Robinhood teases cat Crypto * BTC: 78,004 (0%) | BTC.D: 59.7% (-0.1%) * ETH: 2,455 (0%) | BNB: 687 (0%) | SOL: 102 (-1%) * Fear & Greed: 70 | 24h Liq: $143m * BTC ETFs: +$217M | ETH ETFs: +$88M * Top Gainers: MARSCOIN, CAG, USELESS, ARB, PONS * Crypto steady, as Strategy buys $370m BTC * HYPE rallies on US entry talks * SOL prints first green month since Sep 2025 * Strive buys $143m BTC * Metaplanet moves $377m BTC to Coinbase * BitMine buys $131m ETH, holds 5.9m of ETH * Solana non-vote txns hit record 191m * PONS, AI continue pump * Robinhood teases 8-bit cat * CME launches altcoin indices * OpenSea now supports Solana NFTs * DeFi Development prices $19.8m CHAD deal * ICE taps tZERO for NYSE tokenised platform * Telegram begins Gram Wallet rollout * Pi Network ships a DEX as market looks away * 1789 Capital to invest $300m+ in Polymarket * Solana AMM Aquifer hit by $2.5m exploit * Crypto hacks rose 67% in Aug to $136m * South Korea BTC premium back after 15 weeks * Payward to tokenize LSE stocks * Lazarus wallets move $30m via Hyperliquid * Japan moves to cut JPY stablecoin tax rules * Russia opens regulated crypto trading * Singapore proposes 100% reserve stablecoins * Ireland excludes crypto from tax saver accounts * Thai SEC eyes rules for retail crypto derivatives Macro * NASDAQ: $26,370 (-0.1%) | Gold: $4,420 (-0.6%) * S&P 500 posts ATH monthly close * Trump threatens new Iran strikes * Trump rules out nuclear weapon on Iran * Japan 10yr yield hits 2.98%, highest since '96 * Bessent tells BOJ’s Ueda to raise rates * Only way out of debt is to grow: Bessent * Gold worth $5,025 on M2 basis: Fidelity * ExxonMobil going back into Venezuela: Trump * SC allows Trump to build WH ballroom * 51% odds Dems sweep midterms * DHS plans voter fraud crackdown in 9 states * US drops Obama-era truck fuel rules * Russia hammers Kyiv for sixth straight day * Russia seeks AI, digital tie-up with China * Army chief Driscoll quits amid Hegseth row * US Army to install 20 nuclear micro-reactors AI & Tech * Outlook and ChatGPT Work hit by outages * OpenAI is destroying evidence: Apple * Anthropic, Lambda agree $35b compute deal * Sony sues Anthropic over AI music training * Instagram limits undisclosed AI profiles * Groq 3 LPX now in full production: Nvidia * Chinese AI firm API sales surge * Saudi Humain partners Together AI on DCs * Hugging Face’s duck robot selling fast * Tesla adds Cybercabs to Texas robotaxi fleet * FTC sues Amazon over advertiser overcharges @YEET: Crypto's Casino Just passed $3.2B in lifetime volume! gm