Newsletter · · Ashutosh Agarwal

Uranium Hits a Record as New Nuclear Hits a Financing Wall - Powering AI: Grid, Gas, Generation & Nuclear - Week of July 17, 2026

Powering AI infrastructure newsletter for the week of July 17, 2026. Uranium's term price set an all-time record even as new nuclear stalled at a financing wall, gas turbines stayed sold out for years, and the week's podcasts debated who actually eats the risk of building AI's power.

Powering AI: Grid, Gas, Generation & Nuclear

Week of July 17, 2026: Uranium Hits a Record as New Nuclear Hits a Financing Wall

Everyone agrees the demand is real and enormous. What the podcasts spent this week arguing about is quieter and more useful: who actually eats the risk of building the power. New nuclear keeps hitting a financing wall even with government money on the table, gas turbines are sold out for the better part of a decade, and the one number that moved without any drama was the uranium term price, which just printed an all-time high. If you own anything in this complex, this was a week about the plumbing, not the hype.

TL;DR

  • Uranium's term price hit a record ~$95/lb, the highest ever, while spot sits around $85. Multiple podcasts frame the miners as having badly underperformed the commodity.
  • New nuclear's problem isn't the reactor, it's the contract. Hyperscaler "deals" are mostly non-binding, dumping ~100% of construction risk on the developer, even as one startup (Aalo) says it reached criticality on July 4.
  • Gas turbines are effectively sold out (7-8 year waits), so data centers are grabbing anything that burns, from jet engines to diesel gensets, and that's pushing power rates up and pulling copper, gas and gensets along with it.

What's new

Uranium's term price quietly set an all-time record. On Sprott Radio, Per Jandier, director of nuclear fuel at WMC, said the monthly term price, the price utilities actually sign multi-year contracts at, hit $94 on UX and $95 on TradeTech, "the highest it's ever been," matching only the 2007 spike. Spot is stuck around $85. His key point for anyone trading the space: the term price sets a floor under spot, because when the gap gets past about $15 a pound "that carry trade becomes a no-brainer", a trader just buys spot, carries it, and sells it forward to a utility. He also flagged that utility contracting activity picked up "over the last two weeks," India locked two big contracts with Kazatomprom and Cameco, and supply keeps stumbling (Kazatomprom short of sulfuric acid and sticking to "value over volume," Cameco's spring flood washout in Saskatchewan now fixed). Why it matters: a record term price with spot lagging is exactly the setup contrarians look for, and it feeds straight into Cameco's realized pricing.

The financing wall for new nuclear got laid out in painful detail. The bear case of the week came from Decouple, where Michael Seely (of The Atomic Blender) explained why almost nothing is getting built despite incentives that cover up to half of project cost. Small modular reactor costs are drifting back toward big-reactor reality, Ontario's Darlington first unit is running about $4.5 billion in US dollars, nowhere near the "$1 billion a reactor" pitch, and NuScale's flagship Utah (UAMPS) deal already collapsed. The killer detail: the power-purchase agreements hyperscalers are signing are non-binding, "we'll buy the electricity if it exists, from this date, at this price", which "shifts 100% of the risk onto whoever's developing that project. The customer is taking on zero risk." That, Seely argues, is why suppliers can't place firm component orders and the whole chain stays stuck at one-off demonstration plants instead of an order book of ten. Why it matters: it's a direct caution on the SMR equities and a reminder that signed offtake, not enthusiasm, is the catalyst to watch.

Meanwhile, one reactor startup says it actually turned on. On Energy News Beat and CC Pod, Yasir Arafat, CTO of Aalo Atomics, said the company designed, built and reached "criticality" (the point a reactor sustains a nuclear reaction) on a brand-new reactor on July 4, 2026, in under twelve months, one of the winners of the Trump administration's DOE reactor pilot program, with Energy Secretary Chris Wright signing the go-ahead on-site in Idaho. Arafat's design pitch is the interesting part for investors: a 50-megawatt "pod" built from five small reactors and three turbines with built-in redundancy, sized to sit next to a data center and replace a gas turbine, with a long-run cost target of 3 cents per kilowatt-hour. Aalo is scaling a factory from 40,000 to 200,000 square feet this year, and, crucially, uses standard low-enriched uranium fuel rather than the still-scarce HALEU everyone else needs. Why it matters: it's the operator counterpoint to Decouple's skepticism, real hardware exists, but note the tension: Aalo is private, and Seely's point about who pays still applies once you leave the demo stage.

Gas turbines are sold out, so data centers will burn anything. On Renewable Rides, Robbie Orvis of Energy Innovation said if you order a new grid-scale gas turbine today, GE or a peer will tell you "see you in seven, eight years." The result is a scramble for "behind-the-meter" power, data centers "taking old jet engines, basically anything they can find," and in his own northern Virginia, diesel generator fumes. He said cost simply "is not an issue." His modeling shows wholesale power rates rising about 74% by 2035 as cheap clean supply rolls off and pricier units set the price, with distribution-system costs now one of the biggest drivers of household bills. Why it matters: multi-year turbine lead times are the single clearest bottleneck in the whole chain, great for the OEM order book, and the reason off-grid gas and gensets keep winning deals.

Texas is about to try to double its grid in five years. On the Energy Capital Podcast, Bryn Baker of the Texas Energy Buyers Alliance walked through ERCOT's new "batch" process for connecting giant loads: roughly 35 gigawatts of firm new load plus another ~65 gigawatts studied, for up to 110 gigawatts over five years, against a current system peak near 86 gigawatts. About $37 billion of transmission is already baked in, pushing costs up ~3.5% a year for everyone even before the new loads, and there's a live fight over how to split the bill (the "4CP vs 12CP" cost-allocation debate) plus a proposed minimum charge on big loads so they can't dodge their share. Her plain-English line on transmission: it's like converting "two-lane highways to eight-lane highways" with new 765 kV lines. Why it matters: this is the rate-base growth story utilities and equipment makers are counting on, and Texas is the leading edge of it.

The debate

Bull. The through-line across the week is a genuine, multi-year supercycle that lifts grid, generation, gas and uranium together. On Mission Matters, Jim Struger of Kineo Capital sized the US power shortfall at 60 to 90 gigawatts by 2030, "in nuclear reactor terms... between 60 and 90 nuclear reactors short", and pointed to government putting real weight behind it: the Department of Energy invoking the Defense Production Act to treat AI compute and the energy stack as national security, about $18 billion of DOE loans toward ten Westinghouse AP1000 reactors, and FERC telling hyperscalers to "bring your own power." He leans on Nvidia's Jensen Huang forecast of $3-4 trillion of annual AI-infrastructure spend by 2030 (a ~48% annual growth rate off a ~$500 billion base) and $8.5 trillion through 2040, and, tellingly, invests "all around the data center... out to the grid," never in the software on top. The optimists' framing was loudest on The Angry Clean Energy Guy, where Assaad Razzouk called data centers "the ultimate catalyst and financier" of grid modernization, citing $6.7 trillion of AI/data-center spend by 2030 with $1.3 trillion-plus of it flowing into generation and transmission, and Texas demand headed to 367 gigawatts by 2032.

Bear. The skeptics don't dispute the demand, they dispute the timeline and who pays. Decouple's financing wall is the sharpest version: without firm offtake and a construction-risk backstop, the projects stall no matter how good the incentives look. On Money of Mine, Matt Griffin, a uranium bull on the commodity, was blunt that SMRs won't move the needle "any time this decade" and that the working-capital cycle for new mines is "brutal" (Lotus won't see first cash until roughly a year after commissioning). And on Open Circuit, the panel showed the grid is already flying blind, PJM projected 156 gigawatts of summer peak and got about 166, while gas can't deliver into PJM until 2031-2032, leaving a messy, chaotic gap that distributed and behind-the-meter players are only partly set up to fill. The honest read: this week the bear case was about sequencing and financing, not about the demand being fake. Nobody credibly argued the load growth won't come.

The names in play

Cameco (CCJ) is the cleanest beneficiary of the record term price. Sprott's guest noted Cameco is largely meeting production targets after fixing its Saskatchewan flood disruption, and on Money of Mine, Griffin said Cameco is "locking in deals above $100 a pound on a long-term basis... ceilings up to $140, $150." The recurring caution across both uranium episodes and In it to Win it (Nomi Prins) is that the developers and juniors have been the disappointment, Prins still likes "domestic and Canadian" producers into the second half, but Griffin won't touch pre-production names and holds only Paladin. Next catalyst: the next monthly term-price print and whether utility contracting keeps accelerating.

GE Vernova (GEV) and Caterpillar (CAT) came up as the equipment winners of the gas-for-power scramble. On Squawk on the Street, Jim Cramer and David Faber flagged both as beneficiaries of the Microsoft-Chevron deal to pump gas straight to on-site turbines, with Cramer arguing Caterpillar's diesel-and-gas gensets clustered next to Marcellus and Utica gas are "the cheapest way to be able to have gigawatts" completely off the grid, and noting GE Vernova is now a ~$290 billion company that started life as the unloved piece of GE. This is pundit commentary, not company guidance, so weigh it accordingly, but the seven-to-eight-year turbine backlog Orvis described is the fundamental underneath it.

Constellation (CEG) and Vistra (VST) were named on Stock Club as the blue-chip way to own nuclear versus the "moonshot" pure-plays, with the Microsoft-backed Three Mile Island restart cited as a near-term Constellation catalyst. On the SMR side, the same hosts framed NuScale (SMR) (targeting a reactor by 2030), Oklo (OKLO) (first unit targeted 2027, selling directly to data centers), and the fuel "picks and shovels", Centrus as the only US-licensed HALEU supplier, while acknowledging only two SMRs run commercially today, both over budget and behind schedule. Read that alongside Decouple's warning before extrapolating the equities.

Read-throughs

  • Gensets (CMI, CAT): the turbine backlog is the gift that keeps giving to anything that makes on-site power. When Orvis says cost "is not an issue" and Cramer calls diesel-plus-gas the cheapest path to gigawatts, that's demand for engines and gensets that doesn't wait for the grid.
  • Copper and metals (FCX, and the cable/conductor complex): every one of these builds is copper-hungry. On Equity Mates, the hosts noted Microsoft's $500 million Chicago data center used 2,177 tons of copper and named BHP and Freeport-McMoRan as the biggest producers; data-center energy demand grew 17% in 2025 and is set to double by 2030. More concretely, Gunnison Copper says it's already selling cathode straight to Amazon Web Services and is chasing a project that could supply ~11% of US refined copper. Copper is the pick-and-shovel under the whole theme.
  • Gas-weighted midstream and LNG (WMB, KMI, LNG): the same Marcellus/Utica gas feeding off-grid turbines also feeds the export machine. RBN Energy detailed a revived Philadelphia-area LNG project (Eddystone, 950 MMcf/d) that would lean on Williams' Transco and Enbridge's TETCO, and counted nine US LNG projects that have reached final investment decision, about 11 Bcf/d of new export capacity by the early 2030s. Domestic data-center gas demand and LNG export demand are now competing for the same molecules.
  • Enrichers and fuel-cycle names: Aalo's decision to use standard low-enriched fuel is itself a read-through, it says the HALEU supply chain most advanced designs need is still "a couple of years out," which is bullish scarcity for the few licensed enrichers (Centrus was singled out) and a gating item for the rest of the SMR field.
  • Hyperscalers funding the deals (META, MSFT, AMZN): they're the balance sheet behind everything here, Meta's ~5 gigawatt project with Entergy, Microsoft-Chevron's off-grid gas, Amazon buying copper cathode direct. But per Decouple, their PPAs so far transfer very little risk, so watch for the first genuinely firm, signed hyperscaler commitment, that's the moment the developer economics change.

What changed

The new wrinkle this week is policy pushback on the demand side. On Closed! NYC's Real Estate Podcast, New York State Senator Kristen Gonzalez laid out a one-year data-center moratorium bill, noting the state has 28-plus hyperscaler proposals totaling roughly 9,000 megawatts, about a third of New York's current usage, and that US data-center power use, ~4% of the total today, is expected to triple within three years. Paired with the "bring your own power or pay a premium" stance echoed by Governor Hochul, it's a reminder that local siting and rate-shock politics are becoming a real gating factor, not just an afterthought, worth tracking as more states follow.