Noreva on PJM’s Price Collar: The Great Data Center Migration Begins

What the Bloomberg coverage examined

The article previewed a capacity auction carrying more weight than usual, and set out why (Bloomberg, July 14, 2026):

• The auction sought supply commitments from generators and other electricity suppliers for the June 2028 to May 2029 period.
• PJM serves 67 million customers across 13 states and had failed in the previous two auctions to attract enough commitments to cover its reliability requirements.
• The price collar was capped at $325 per megawatt-day, marginally lower than the previous sale in December, a mechanism criticized as a disincentive to power plant construction.
• The December auction left households and businesses on the hook for a record $16.4 billion to secure power for the year starting June 2027, and still fell short of reliability requirements.
• An emergency auction had already been scheduled for later in the year to cover any shortfalls.
• A Federal Energy Regulatory Commission technical conference is scheduled for July 23 to take up reforms to PJM’s governance.

Bloomberg quoted PJM Senior Vice President Adam Keech, who had written that recent auction results showed power demand growing faster than new generation can be built, and that the challenges facing the electric system “require longer-term solutions.” Evercore ISI analysts noted that a recent heat wave that triggered record electricity demand was a timely reminder of how burdened the system has become.

Gardett’s read went to the consequence rather than the clearing price. As quoted in Bloomberg: “Another print at the top end of the price collar will only increase the pressure to push data center demand off the grid, even at the cost of servicing the AI economy.”

And, in the same remarks: “We are witnessing the beginning of the great data center migration.”

Why a capped price becomes a migration signal

The logic is worth following carefully, because it inverts how capacity prices are usually discussed.

A price collar is designed to protect consumers by capping what capacity can cost. But a cap that binds is also a cap on the signal that would call new generation forward. When an auction clears at its ceiling and still falls short of the reliability requirement, the market is saying that the price required to attract capacity is higher than the price the mechanism permits. The shortfall does not resolve itself; it just stops being priced. Gardett’s reference to “another print at the top end of the price collar” points to auctions already running into that ceiling.

On his read, that leaves the adjustment to happen somewhere other than in supply. Political pressure builds on the demand side, because someone has to answer for a record $16.4 billion consumer bill, and the load that grew fastest is the obvious candidate. The pressure then moves data center demand off the grid: behind-the-meter generation, co-located supply, or interconnection queues in other regions. His phrase for that displacement is the great data center migration, and his sharpest point is that it would proceed “even at the cost of servicing the AI economy.” On this argument, the grid does not optimize for the AI buildout. It responds to the pressures it can act on.

For asset owners and investors, this reframes what a capacity print means. A cleared price at the collar is not a straightforward positive for generator valuations in the zone, because it also marks the point at which the largest source of load growth starts looking for an exit. Migration changes where load lands, which changes locational value, which changes what a given asset in a given zone is worth over the life of a contract. Behind-the-meter and co-located configurations remove that load from the capacity obligation entirely.

This is the question Noreva’s capacity work is built for: not what the auction cleared at, but what the clearing price does to the location of demand in subsequent years. Noreva models capacity price forecasts, resource adequacy, and ELCC accreditation alongside nodal power pricing, so that a migration scenario can be priced rather than described.

Update: what the auction did

PJM published its 2028/2029 Base Residual Auction report on July 14, 2026. It printed at the cap. All prices cleared at $325.00/MW-day, as the prior auction had cleared at its own cap of $333.44 (PJM, July 14, 2026).

The rest of the report bears directly on the argument above:

• Total cleared value was $16.4 billion, unchanged from the 2027/2028 auction.
• The auction procured 6,831.3 MW UCAP below the RTO Reliability Requirement, against a 6,516 MW shortfall the year before. The gap widened.
• RPM cleared at a 14.4% installed reserve margin, 5.6 points below the 20% associated with a one-day-in-10-year loss of load expectation, and the second consecutive auction more than a point below.
• Offered supply rose 3,447 MW, but new generation and uprates accounted for only 524.7 MW UCAP of that increase.

PJM also published a simulation of the same auction without the cap and floor. It clears at $554.72/MW-day, and at $776.69 in COMED, for a total of $29.7 billion, or $13.3 billion above the actual result (PJM, July 14, 2026). PJM notes that its simulation used the offers actually submitted, and that it cannot know how sellers would have bid under different rules.

That difference is the clearest available measure of what the collar suppressed. PJM states the mechanism in its own executive summary: a price cap set below the established VRR curve “can reduce the amount of investment, and therefore supply, in the PJM region.”

So the auction did not resolve scarcity at a lower price. It declined to price it. Two consecutive auctions at their respective caps have been accompanied by roughly half a gigawatt of genuinely new capacity, while the shortfall against the reliability requirement grew. Gardett’s pre-auction read, that the response to another print at the ceiling would appear on the demand side rather than the supply side, is consistent with what the report shows. Whether the displacement he described follows is the question the next several delivery years will answer.