It is thrilling to see *an entire field* emerge. When I was writing Tubes, the academic work on the geography or materiality of the Internet (as something distinct from “telecom”) was very close to zero.
(A sprinkle of optimism.)
Google doing PPA's like this were what drove early utility-scale solar investment. Amazing to see it happening again with fusion. blog.cfs.energy/google-deal-...
For MIT Technology Review's "Power" issue, I traveled to Nebraska to look at how one electric utility is confronting the most challenging and dynamic moment the grid has ever faced. ter.li/AndrewB
Sometimes we do new things!
I'm going to guess they're pretty interested.
"while CFS plans to sell the electricity ARC generates into the wholesale energy market, the company is also in discussions with large corporate buyers interested in procuring the environmental benefits of this clean energy"
🙋🏼♂️ — working in the background here, with my book out in 2026.
Love that, thank you. And I hope you’ll like the new one (coming in ‘26). It’s got some Sheldon Kimber in it!
Data centers going to power isn’t new. That’s why Google went to The Dalles in 2006. What’s astonishing here is building power and compute *together* — much less gigawatts of it, spread out over tens of thousands of acres.
“An AI campus that would use about 1 gigawatt of electricity would require a gigawatt each of wind, solar and battery storage with two to four hours of duration, sprawled over at least 10,000 acres.”
(I also wish the paper of record didn't refer to ECMWF's EPS as "a service that 35 nations rely on to produce their own weather forecasts.")
I should say the *supremacy* of physics-based models.
Yes. And I wish that much of the coverage (not yours) was clearer about how dependent the AI models are on the existing modeling infrastructure. And yet, for the bleeding edge to be AI-based, not physics-based, is really something.
The era of physics-based numerical weather models.
And credit where credit is due: "We’ll be releasing our model’s code, weights, and forecasts, to support the wider weather forecasting community."
I was skeptical for a long time, but Google's AI weather modeling is astonishing. When I wrote THE WEATHER MACHINE, I did not realize I was chronicling the end of an era in numerical weather prediction. deepmind.google/discover/blo...
Meta out with a nuclear RFP. Whoosh.
“Advancing the technologies that will build the future of human connection — including the next wave of AI innovation — requires electric grids to expand and embrace new sources of reliable, clean and renewable energy“ sustainability.atmeta.com/blog/2024/12...
Plus ça change.
"unlike previous infrastructure projects in which Meta has been involved, such as the 2Africa cable, the company will own the new cable outright...Speculation is understandable, therefore, that Meta might prioritise its own data on the cable, and that of any companies it favours."
This feels like a through-the-looking-glass moment, where Ryan Maue—well known as a climate change denier at the other place—convinces the NYT to print his NOAA administrator cover letter. www.nytimes.com/2024/12/01/o...
I’m not quite going to make 2025, but my energy book is coming: ELECTRIC EARTH www.linkedin.com/posts/ajblum...
In 2022, I wrote for TIME about Amazon's big push in purchasing renewables. (That's still happening -- but not necessarily fast enough to keep up with AI energy demands.)
time.com/6213666/amaz...
Big tech seems more worried about falling behind in AI, than falling behind in their climate goals.
Party tricks included.
I think $20/month/seat gets you unlimited, with AI. It’s not clear to me how it’s handling PDFs, so it might require a manual process to import them as text.
I’m also a bit haunted by the energy usage of this, compared to local search of notes, the old fashioned way.
for example! Impressive, but I’m debating how much more useful it is than a good search.
I’ve got a research database somewhere north of 500k words, in Notion: interview transcripts, notes on news articles (not complete text), some full text reports (if they’re foundational). Notion AI is chat based, and you can specify results from a certain document, or leave it open.
I’m doing this with @notion.com with decent results. It excels at pulling facts from my corpus (with great footnotes). It’s mediocre at any kind of synthesis.
Some reporting for the new book yesterday: an emergent moment of NYC’s biggest new piece of energy infrastructure.
Hey that’s my book
