Nuclear Gets Its First U.S. SMR Construction Permit
The NRC issued the first U.S. construction permit for a BWRX-300 small modular reactor, a 10th-generation boiling water design from GE Hitachi. The thread was notably skeptical. The best-case bet from one commenter: 15 years and $10 billion to first power. Someone else pointed out the BWRX-300 footprint is large enough that you might as well build a full-sized plant for 3-4x the output. The permit is real, but the gap between a permit and a working reactor in the U.S. is historically enormous.
This sits directly next to the data center secrecy story. Most data centers are refusing to disclose water and electricity consumption, and the EU is watching AI infrastructure demand grow while feeling paralyzed to respond. The energy demand from AI is real and accelerating. Nuclear SMRs are the only serious answer that doesn't require either massive grid upgrades or accepting intermittency.
The counterpoint in the thread: 15 years is too long to matter for current data center planning cycles. But the permit is a forcing function that didn't exist before.
So what?
If you're building infrastructure-heavy AI products, the energy constraint is not a background concern anymore. It is a first-order constraint that will shape where compute gets built, at what cost, and for whom. Founders building on top of cloud compute should track energy costs as a structural risk to their margin assumptions.