420.8 GW is alot of compute... like an ungodly amount. A vera rubin rack is like 220KW.. so 1.9m racks. a rack is 72 GPU's per rack so 420.8GW of power is 137.7M GPU's. So lets say about 60% of the power goes to infra, we get 55.08M GPU's. So, with each GPU costing around 55k, that gives us a final cost of compute for the 420GW is 3.0294 trillion.
Obviously that is not real... Eithier double booked, or fake corps making a bunch of requests from thousand different vectors to resell access.
Also, companies like Base are making a killing off taking all the excess energy generated by wind at Night and selling it back to the grid.
So in final, we can support around a $6T AI buildout in Texas just on our exported Natural Gas alone before we really need to start expanding LNG supply.
[–] vel0city 2 days ago
[–] burnerRhodov3 2 days ago
Ahhhhh, Makes sense now. Page 7 of your slideshow, Key Takeaway: Solar and Energy Storage account for approximately 72% of the amount of generation seeking interconnection in the ERCOT grid.
Remember i mentioned Base earlier? Those interconnects are for peps selling or arbing in energy. Not to pull energy out of it.
[–] vel0city 2 days ago
All I'm certain of is all this new demand willing to pay any price means things will be more expensive just to try and not melt or freeze in the Texas weather.
FWIW I'm far from an AI detractor. But we need to make sure we're putting these costs where they belong, and some of these plans do seem insane.
[–] burnerRhodov3 2 days ago
What happens when companies have more GDP than the united states? Because i though the two would be correlated since American companies, but i guess if you can sell to a global market and the government can only tax it's citizens, that's kinda the inevitable outcome.
[–] vel0city 2 days ago
These hyperscalers are issuing so much debt and investors are certain enough of returns they're upending global bond markets (among other factors, it's not just AI). Tons of large countries are having to mess with interest rates because the groups who used to buy their debt are already tossing their money at AI datacenters at a massive scale.
> JPMorgan Chase estimated in June that $4.1 trillion in AI-related debt will be issued through 2030, as data center companies and others tied to the AI boom race to build up capacity.
And once again that's just the debt. They're also all making many billions of dollars every quarter in revenue already, they're planning on that increasing rapidly, and dumping that cash into building as fast as they can.
https://www.cnbc.com/2026/09/27/debt-hungry-data-center-comp...
So yeah, your napkin math of them spending trillions on datacenter builds you used to try and illustrate how impossible that could be...yeah...that's been the plan for a while. It's an insane number, but that's the number.
https://www.ercot.com/files/docs/2026/07/29/ERCOT-Senate-Jul...
And yeah, supposedly it's over $1T in computer capacity allegedly planned to be built.
> just on our exported Natural Gas alone before we really need to start expanding LNG supply.
That's natural gas which already has a buyer though. We're now adding a buyer who isn't batting an eye at spending over a trillion dollars. What price of natural gas do they stop buying at? What price will the people of Texas, surrounding states, and our international customers pay for gas until we can't afford it and freeze?
So yeah, if we do build out anything close to this I hope we manage to increase energy capacity similarly. Because I doubt I can spend as much to heat my home as these hyperscalers can spend to power their chips.
That's kind of the whole point of my "sucks the natural gas pipes dry". I'm not saying the gas pressure of the line going to my house is literally going to zero, but if it's so expensive I can't afford for it to flow it might as well be.
Also, your number of 72 GPUs per rack is a dated figure. They're targeting >500 per rack these days. That's also assuming GPUs. What's the same math on TPUs?