News: This might be the most important structural shift in US power markets this year. PJM filed a proposal at FERC on Aug 13 to formally stop planning around a class of customers. The largest grid operator in the country, 67 million people across 13 states, told data center developers better bring their own power. Facilities built after mid-2027 that don’t secure their own generation go first in line for curtailment during shortages, and starting with the 2029-30 capacity auction, loads that show up empty-handed won’t even be counted when PJM calculates how much supply to buy for everyone else.
The grid has always been a really good deal: hook up, pay your tariff, and the system builds capacity to serve you. When the deal is too good, the line gets extremely long. ERCOT interconnection queue grows to 474GW, 90% data centers, 5X the grid’s all-time peak. PJM is staring at a 6.8 GW reliability shortfall and has run out of polite options. So it’s splitting the grid into two tracks: the socialized one you and I pay for, and a private one where hyperscalers self-provision or wait.
We’ve covered the causes and implications of BYOP a few times now. Old story: when a system runs out of headroom, somebody gets deprioritized. This week I went looking next door. Our food system runs on some of the same scarce inputs as AI infrastructure. How is it doing?
When Covid first started, my friends and I joked about buying farmland in Canada. The theory was that a warming climate would extend the growing season further north. Being a landlord also has its own appeal, collecting a small toll on something as old as human history.
We never did it, and it’s probably a good things as landlord economics have become less glamorous. In 2021, an Idaho farmer with a 150 horsepower irrigation pump paid about $700-800 per month. This year, the same pump would cost $1,800 to run. American agriculture depends on 3 physical inputs: diesel, water, and flat land sitting near power. For most of a century it held first claim on all three, because in rural America nothing else wanted them. That is no longer true.
Nitrogen fertilizer is more than 70% natural gas by variable cost, which makes every acre of corn a long position on the gas curve. When the Strait of Hormuz effectively closed this spring, urea went vertical and we’ve seen lots of fertilizer crisis stories. Outside of the war factor, the high gas price will continue to limit marginal nitrogen producers’ operating rates. Meanwhile RBC has US data center gas demand heading toward 6.1 Bcf/d by 2030, with something like 100+ GW of onsite generation announced purely to skip the interconnection queue. USDA reported farm electricity spending rising 48% between 2019 and 2026. Roughly 2/3 of planned DCs are slated for rural counties. Texas’s agriculture commissioner said developers target the most productive ground precisely because it’s flat and already has water, power and roads, the same features that made it good farmland in the first place. Ammonia and AI, corn and compute, now shop at the same counter.
One thing I don’t quite understand is while ag inputs are up this much, why the food inflation index is only up by 3%, so consumer is not feeling any of this? But if farm operators are eating up all margin compression, that’s probably not where we want to invest.
Two ideas:
Ag water. Roughly 9% of the Ogallala’s original volume is gone, and on current rates more than 1/3 the Southern High Plains can’t support irrigation within 30 years. Texas is deploying $450M into precision irrigation, rural piping and storage, with a state board monitoring usage in real time. Kansas has its LEMAs. Quota plus meters is exactly how a resource becomes an asset. Mobile drip retrofitted onto existing center pivots saves 30 to 60% of the water on installed base. Can we fund a fleet of mobile drip + irrigation automation retrofit and monetize saved water rights? I’d rather sell those water rights to whoever needs water and social license and has a balance sheet, than collect cash from a farmer.
Cold chain storage. The average US cold storage facility is 40+ years old, that can’t support modern throughput and lags in energy efficiency control. Demand is rising due to unstable trade policy: when export windows are political, somebody has to hold the crop, and cold storage can become both a transit function and a strategic buffer. Power accounts for 18% of facility operating cost. A refrigerated warehouse is a very large thermal battery that happens to contain food. Can we automate refrigeration control, pre-cool on cheap power, coast through the peak, get paid for the flexibility. Slate and Hamilton Lane just took majority control of Cold-Link Logistics; DHL is building 5 million sqft with RLCold. Can we fund the facility upgrade/retrofit, making them more energy efficient, and take capacity payment?
The core assumption behind both ideas is that water and power will stay expensive, which I believe it will remain true for a while.