Comments by "Tony Wilson" (@tonywilson4713) on "The Geopolitics of...Gaming || Peter Zeihan" video.

  1. AEROSPACE ENGINEER HERE I have been highly critical of Peters reporting on technology and urged him to get some decent background supplied by engineers. I don't know if he's noticed my comments or the comments of others, but this is certainly one of his better (if not the best) take on a technology he's done YET. The real problem that's going to hit the AI people is the energy. If you consider that some of the proposed server centres that are around the US$2 Billion mark they need the equivalent of a Westinghouse-Toshiba AP1000 Nuclear Reactor. That reactor gets mentioned (more than others) because its one that's already been built and has all of the approvals which are 2 things none of the SMRs or other options can say. The other 2 things we know are how long it takes to build an AP1000 and how much it costs thanks to the Vogtle Plant in Georgia and that answer is 9 years and US$18.4 Billion for each reactor. Now the advantages of an AP1000 are: 1) They are designed to last 60 years, which can probably be extended and that means that no matter what the technology becomes in future you can keep using the same site to power your server for a long time. Plus if the next generation of chips that Peter described use a lot less power then you can sell the excess power into the local grid which will help pay-off that US$18.4B. 2) The AP1000 was designed to be built in modules. This is different to the SMR concept where each module is the same and you simply keep adding modules until you have what you want. The AP1000 modules differ in that each module is a piece of an overall plant that can be built off-site and shipped to site on a barge. There's nothing new to that concept I have seen it done for mineral processing and seen modules the size of office blocks. There's a massive advantage to this in that you end up with a module construction site that keeps developing its skill and knowledge bases allowing for manufacturing improvements over time. Once you sort out the module construction and site construction issues this method allows for more efficient or cheaper or quicker construction or a combination of all 3. The disadvantages of the AP1000 are the same as every other nuclear design: 1) They still take almost a decade to build. Sorry but like every other large complex engineering machine they take the time that they take. There's just no way to get around the time required for certain things like constructing the foundations. No matter what the "nuclear island" the reactor sits on needs a foundation to sit on. that and all the other things TAKE TIME. 2) If your construction site is not close to where the barges can deliver the modules the building advantages can go away. This was an issue at Vogtle. 3) What do you do with the spent fuel. Engineers have put forward proposal after proposal on what to do with spent fuel and its like cutting off the head of a hydra. For every proposal put forward 2 new objections grow out of it. We are engineers NOT mythological Greek heroes who can kill a giant multi-headed snake. So the AI crowd are kind of trapped right now. They want to build these monstrous server farms to support the LLM Systems but they can't find spare capacity in the energy supply for it and they can't build anything fast enough and their new chips wont be ready for deployment for at least 5 years. AND YES if Peter wants to give me or another engineer a job he's free to ask.
    1