The Texas grid which is separate from the rest of the US has a size issue with balancing and peaking that ha shown up from time to time. Their grid collapse a few winters ago would (probably) not have happened if they were connected to the big grid like everyone else is.

There are two other US grids, the Eastern and Western Interconnects. They are not frequency synchronized, and there are minimal AC-DC-AC interties connecting them.

The big collapse in Texas involved about 30GW of generation going off line. Even if Texas were connected to the Eastern grid, it is unlikely that there would have been 30GW of spare capacity and 30GW of available transmission to draw on.

Perhaps relevant, ERCOT and the Texas PUC have approved the Southern Spirit 3GW HVDC transmission line that would connect ERCOT and the western edge of the TVA's transmission network. The other states that would be affected have refused to approve the project because they believe Texas will screw them over and drive their electricity prices up.

Worth saying: In Texas, it's easy to find 3.75 square miles that's so desolate it's not useful for anything else. My local power authority here in Colorado has a power-purchase agreement with a solar farm about that size. The land it's on is so poor no one has ever trying either growing crops or running livestock on it. With the panels channeling rain water into narrow strips, it might now support enough grass for a small number of sheep, but probably not enough to justify the effort. Even more convenient, the land was adjacent to an existing transmission line, so the connection cost was a smallish substation. The authority's first battery farm is going in right next to the substation. There will probably be more.

My GP was ahead of her time. Many years ago she used to tell me, "Climb ten flights of stairs over the course of each day, don't worry about your heart."

I checked at the web site, and like all the other AI chip designers, actual fabrication is by TSMC at 5nm and 4nm. My understanding is that TSMC's fab complex in Arizona can do 5nm parts, and will be able to do 4nm relatively soon. Also that they are at Arizona's capacity for those nodes doing work for Apple and AMD. So FuriosaAI's parts will be made in Taiwan.

The US AI stock bubble is hideously dependent on a few fabs in Taiwan.

...I remember when companies wanted to accelerate depreciation, so they could take the write-off sooner.

I also remember the days when the tax laws made me a depreciable asset in my role doing tech research. I never quite worked up enough nerve to call the accountants and ask what happened after they had depreciated away my full value :^)

The Fort St. Vrain commercial reactor in Colorado was gas-cooled and ran on a Th232-to-U233 fuel cycle. Fissile U235 was used to get started, but almost all of the energy output was from fissioning U233 produced in the reactor.

Current iron-air batteries have efficiencies around 50%. That is, put 2 kWh in, get 1 kWh back out. That ruins the economics of a lot of potential applications.

Wyoming is an interesting case. Data centers that have big NG-fired backup generation can connect to the grid only if they turn dispatch of those backup generators over to the grid operator. In high-demand intervals, the operator runs the backup generators rather than bringing in high-cost power from distant generators. Or runs them for frequency control if that's necessary.

There's been at least one case of a data center that signed the contract to allow the operator to run their NG-fired generation. When it turned out they had never actually implemented the control interface, the local utility cut them off cold. Want to buy a data center building with its own substation, the cooling infrastructure, etc? It's sitting idle just outside Cheyenne, with the current owners banned from connecting to the grid.

Did they also ban the use of Latin, abbreviations, and Latin abbreviations?

Back in July the 8th Circuit Court overturned the FTC's "one-click" rule which required that it be as easy to unsubscribe as to subscribe. This case is based on different misbehavior, in particular collecting billing information before revealing terms of service.

Wyoming's Public Service Commission has approved some interesting tariffs for large data center service. The data centers are required to make their backup generation facilities dispatchable under control of the utility most of the time. During periods of high demand, if the backup generators are the cheapest source of additional power, the utility gets to make the decision about running those generators.

The same arrangement probably can't be used in places where an ISO operates the grid through a pure market system.

Most of the biggest battery companies also sell utility scale installations. Tesla has said in public filings recently that they are selling MegaPacks faster than they can build them. My small local power authority -- peak summer demand around 750 MW -- has ordered a 400 MWh battery system. The batteries are being built by a Korean company that is already constructing a big US factory. The power authority would probably accept some delivery delay if they didn't have to pay the tariffs.

I've kept a copy of vile on every Windows box I've been forced to use. Any installation hassle was worth the effort to keep my fingers from having to learn a new editor.

I've always believed the headline writers were thinking they couldn't present some outlandish claim as fact, but rephrasing it as a question let them off that hook. So not "Peking U makes fastest transistor", which as several commenters have pointed out would require some knowledge and an actual explanation, but rather, "Did Peking U make fastest transistor?" From there, Betteridge's Law is just a corollary.

Just from a technology perspective, are there any of the bleeding-edge chip processes that yield radiation-hardened parts, or parts that take the kind of vibration that goes with launch to orbit? TTBOMK, and I'd love to hear from experts, the military is still sticking with 28nm parts for weapon systems. Can anyone even build a competitive AI or cloud data center with 28nm parts these days?