Today on SHIFT KEY: @JodyFreemanHLS joins us to to unpack Trump's moves to kill regulation of greenhouse gases from vehicles, power plants, and everything else. We unpack how the EPA is attacking the cornerstone finding that underpins all federal GHG regs under the Clean Air Act.

I expected this was coming: OpenAI announced they have developed a high performing mini versions of GPT-o4 that will run locally on a laptop or a phone. This will shift a lot of inference off of data centers & onto our local machines. When the model can’t answer a question locally, it can say “wait just one moment while I confer with the hive mind…” or whatever & go do inference in the cloud. Big question for those of us in electricity & data center world: how will this impact data center build-out trends, their demand for electricity, and locations they are able/willing to build? My hypothesis: it puts a dent in demand for inference at centralized data centers, while also increasing the degree of latency that inference data centers can deal with (because low latency frequently needed tasks will be done locally).

RT @robinsonmeyer: Really good profile — captures @TheStalwart and @tracyalloway’s genuine curiosity, the breadth of their listeners, their…

What makes @Google's data center flex announcement today so significant, in my view: (1) These are the first known contracts between a hyperscaler and US utilities around AI data center flexibility; (2) This is the first known case of US data center flexibility being incorporated into the utility planning realm, instead of limited to operations; (3) This is the first time Google is providing data center flexibility via machine learning workloads; (4) The contracts are definitive and long-term in nature; and (5) The announcement notes that flexibility applies to “certain hours or times of the year,” indicating a time-limited arrangement.

You can’t power a state on fantasy. Nuclear: 10–15 years out Coal: No new builds planned, anywhere Gas: 5–7 year delays States banning solar & wind are picking losers—and guaranteeing failure. When gas finds out they're the *only* resource a state will allow, prices skyrocket.

In fact, all but 5 OECD countries have seen rapid declines in coal power... most driven by a rise in wind and solar generation...

What have wind and solar done for us? They've halved coal power in the developed world🤷‍♂️

Very interesting comments here by developers on gas turbine, transformer lead times + other constraints on new capacity Consistent theme that gas turbines are almost over discussed and labor + fuel can be huge constraints. - GE F class gas turbines (~450 MW each) available for 2028/29 - CCGT in 1x1 configuration $2200-2500/kW today. 2x1 configuration more economical but vendors not interested in spending time on that - Cost by the time you build it 4-5 years from now will be higher by 5% per year (!!) - For solar & storage projects, transformer is the longest lead-time (345kV 48 months, lower voltage ~30 months) - Fuel: Lot of attention on turbines but pipelines and storage capacity to handle natural gas scarce as well. For e.g. lot of demand in Georgia but only intermittent gas, pipelines cannot commit to firm deliveries - Labor shortage: finding an EPC to build a plant is serious constraint on new gas power plant capacity in next decade, heavy reliance on trade skills that are in short supply

The labor market is stalling. Manufacturing is getting hammered month after month. And for energy firms, AI remains the only game in town. I wrote about the ominous new jobs report for @heatmap_news:

Facts don’t care about your feelings, lib. But ah, mmm, they do need to care about the president’s feelings. Actually the more they can improve the president’s feelings the better.