Welcome back to CMJ,

20-second highlights:

Fifteen months ago, CMJ named the two mineral systems carrying the energy transition. The mechanism held. What changed is the order: compute pulled gas turbines forward, while nuclear stayed the long-duration answer.
We rebuilt the gas-turbine yttrium math from first principles: ~28 kg of yttrium oxide per gigawatt of new capacity. The number is small. Two North American coating suppliers paused production anyway.
USGS has no clean number to give here. No US trade code captures yttrium alone, and its own import estimate shifted between the 2025 and 2026 editions.
On the customs line that comes closest to yttrium, China-origin unit values went from $19/kg in 2024 to $172/kg by mid-2026, on falling volume. We tried twice to prove that's a licensing premium. Twice, the data said something more careful.
One June customs entry from Peru accounts for more than two-thirds of the apparent recovery in that trade this year. We don't yet know what it was.
U.S. import value for sintered magnets fell 27% since 2023. Tonnage barely moved. What moved is the origin: China's share dropped more than 11 points, to 82%, by mid-2026.
The Department of War put $25 million behind ReElement in July, for yttrium capacity. There's still no price floor, no offtake, no stockpile line, for the one rare earth whose market may be too small to finance on its own.

Illustration of GE Vernova’s head

What CMJ described in CW21 (2025), and where the sequence broke

In CW21 (2025 edition), CMJ described a mechanism: firm generation and torque density are separate industrial problems with separate supply chains, and any credible electrification path has to solve both. Uranium fuels the reactors. Magnetic rare earths spin the rotors. We called them the Power Couple.

The mechanism still holds. What changed in 2026 was the sequence.

CW21 assumed the energy transition would set the pace of demand for firm power. Fifteen months later, compute is one of the accelerants. The IEA's Electricity 2026 forecast puts global electricity-demand growth at an average 3.6% a year from 2026 through 2030, with its AI work pointing to global data-centre electricity use roughly doubling by 2030 from the mid-2020s base.

The volume of power required was expected. What's harder to predict is who wants it, how quickly, and which industrial constraints clear first as a result.

Hyperscaler nuclear commitments are real, but most are long dated. The nearest large restart is Constellation's Crane Clean Energy Center (the former Three Mile Island Unit 1), roughly 835 MW under a 20-year Microsoft power-purchase agreement. In November 2025 the U.S. government backed the restart with a loan guarantee of up to US$1 billion, and Constellation still targets a 2027 return to service.

That gap, between near-term load growth and long-dated nuclear delivery, has made gas more important, not less.

GE Vernova ended Q2 2026 with 116 GW of combined gas-turbine equipment backlog and slot reservations, up from 100 GW at the end of Q1, and expects at least 125 GW by year-end. It's on track for 20 GW of annual gas-turbine output in Q3 2026, rising to 24 GW in 2028 and targeting 30 GW in 2030.

Siemens Energy's August 2026 presentation shows 69 GW of gas-turbine backlog plus 26 GW of slot reservations, with 15 GW booked and 6 GW shipped in the quarter.

❝

A hyperscaler can sign nuclear for the next decade and still need gas before it arrives.

Both legs run at once, and they consume different bills of materials.

Heavy-duty gas turbines protect hot-section components with ceramic thermal-barrier systems that commonly use yttria-stabilised zirconia (YSZ), and yttrium sits on China's April 2025 export-control list for selected medium and heavy rare-earth items. Those April controls remain the relevant baseline. The later October 2025 measures were separately suspended for one year under the U.S.-China arrangement.

That makes an intuitive thesis tempting: the AI power build must be a large yttrium-volume event.

We tested it. Twice. And the arithmetic says something more useful.