PJM Price Cap Hit Again! CHR Explains Why

We Know It's Worth More Than $325

PJM’s 2028/29 capacity auction cleared this week at the FERC-approved cap of $325/MW-day across the entire footprint. Third consecutive auction at the ceiling. The auction procured 138 GW and still came up roughly 6.8 GW short of the reserve margin target, a bigger shortfall than last time.

Shocked but not shocked.

This on top of Q1 wholesale prices running 76% above last year, again attributed largely to data center load. Consumer advocates are already calling for large-load cost allocation reform.

Here’s the read I keep coming back to: when the price hits its ceiling three auctions in a row, the cap is actually censoring the signal. We don’t know what capacity is actually worth in PJM right now; we only know it’s worth more than $325/MW-day. Meanwhile the load class driving the shortfall is the least price-sensitive demand ever connected to this grid.

The Q2 CHR Index puts the blended Compute Heat Rate around $8,000/MWh, roughly 160x the gas heat rate benchmark at around $50/MWh. A demand class with that tolerance does not blink at a capped capacity price. The gap between what the auction is allowed to say and what the demand can pay is substantial, and possibly making the problem worse.

Prices at the cap, reserves shrinking, and the marginal buyer can’t be priced out. Something in this market design gives before the demand does. And there’s not enough generation being built to keep up, full stop.

“But AI is Unprofitable”

I love this one. One of the common pushbacks on CHR is that AI is actually a bubble and these data centers are unprofitable. While the bubble argument may be true and I have no crystal ball (but you can read more about my perspective on that in my piece “The Bubble Bet”), most people making that argument are conflating several things, falsely.

1. AI labs operating at net operating losses from a corporate P&L perspective is NOT THE SAME THING as individual data centers being built via non-recourse project finance and contracted cash flows.

2. Data centers with long-term take-or-pay contracts from creditworthy buyers are some of the most profitable and attractive infrastructure investments in history. And the value creation by converting electrons into tokens is immense, hence all of the CHR values.

If your argument is that the offtakers themselves will go bankrupt, those cash flows will stop, the GPUs will no longer convert electrons into intelligence and the value creation there goes away entirely, then fine. But seeing a headline where OpenAI is losing money (and to be fair, the numbers are eye-watering), is still not the same thing as data centers losing money. The latter are NOT.

For more on how lucrative project finance is for DCs, this recent podcast by Norton Rose sums it up nicely.

I also had a chance to attend the Infocast PowerUp Data Center conference this week, and took away a few notable observations:

• Speed to power is obviously the hot topic and anyone with guaranteed interconnection is the bell of the ball.

• RTOs are scrambling to compare notes on how they’re handling large load interconnects, and there were more questions than answers. RTOs and utilities move slow compared to tech, and the collision of those worlds will be interesting to keep following closely.

• All eyes on ERCOT, SB6 and Batch Zero and how that all plays out.

• Everyone is doing behind the meter, but only because they absolutely have to, and can’t wait for the grid to capture the juicy CHR spread. But everyone WANTS to be grid connected and will do so as soon as they’re physically able.

• Lots of the downstream long-lead items that otherwise would be provisioned for the grid are being bought for BTM applications, making the issue worse, not better. Case in point, I heard someone say, “Well if load goes behind the grid, that actually protects rate payers.” Which I totally disagree with. The grid needs all of it to upgrade.

• Rest assured any remaining slack on the line will be taken by edge compute or other loads seeking electrons - low hanging fruit of capacity on the grid is largely gone.

More soon!

Hans Royal is the originator of the Compute Heat Rate (CHR) framework. All views are his own and do not represent those of any employer or affiliated organization.

Royal, Hans, The Compute Heat Rate: Quantifying AI-Driven Electricity Price Tolerance and Its Implications for Wholesale Market Repricing.
Available at SSRN: http://dx.doi.org/10.2139/ssrn.6322318