What about the compute deals? $3.1B a month for half their compute? Elon built $6B of rev in 122 days. What can he do now that he's got a $85B war chest of cash?
xAI was retardedly late to the party. When Elon Musk founded xAI in July 2023, competitors had already spent years buying up massive silicon footprints and locking down cloud partnerships.
OpenAI had a 3 year lead, anthropic 2.5 years, and google was building TPU's since 2015... what are you talking about?
First, many more people then just them buy these things and have ramped up their spending over the last 2 years.
And others if they were earlier, the explosion and demand of investment increased exponentially over the last 2 years, as we can see in the profit margins of the memory companies for example.
My claim was not that they were the first ever, but that Musk mobilized a huge amount of resources fairly quickly in a time where prices were exploding and before many others.
Early enough that compared to now he got a ok price and he can rent it more expensive then he bought it.
I'm not saying its not a reasonable business, but we are not asking 'can it be a business', we are asking can it be one of the most insane companies in world history. The suppliers have realized whats going on and they are racking in insane profit.
That's why all the Musk fans have realized that they need to scream 'Terafab'. But that is a fanciful concept and unprecedented in history of semiconductors.
>we are asking can it be one of the most insane companies in world history.
"So we went from, he just got lucky cause he got in early", to, "okay, he wasn't early he was just really fast", to, "the suppliers now see he's making a ton of money so they are going to raise prices" to, "terafab is impossible", QED?
You seem to misunderstand how tier 1 capital allocation actually works. You don't get fucked by your suppliers when you are adding $B's to their bottom line. XaI never "Stopped" purchasing GPU's in 2023. They signed multi-year long deals for as much hardware as nvidia, supermicro, dell and micron/skhynix can provide. They are the last to get squeezed. They never stopped adding on to Collosus 2, and they are actively building collosus 3.
Lastly, nothing is "Fanciful" about terafab. ASML makes the machines, Intel provides the cleanroom tech, and 14a pipeline... these are known variables that need to be tackled. Tesla and SpaceX are world renowned, high throughput manufacturing facilities with the ability to make lines more profitable than their competitors. .. no one is claiming he's building a TSMC-style silicon foundry. They are building a massive compute fabrication/clustering footprint. It is a formalized $55B industrial plan filed in Texas, backed by a hard partnerships.
They have quite short cancellation notices and there is a question why do you need to rent it if you are an ai company? How is it grok not running the world using that compute.. And possible cancellation will erase a large income stream, think when gpus get old or sooner when Google does not need it anymore.
an RTX 3090 or older Quadro/Tesla enterprise cards are still selling/ renting... 5 year depreciation is only for tax purposes.
Anthropic is not building their own datacenters... They are mainly giving equity for colocation with AWS. I cant find one example of them ground breaking on anything... Can you give me a source?
yeah, and power is not availabe at capacity until late 2029? They haven't even signed anything, the 'deal' is in principle and BP has been actively trying to challenge it because they want the same power for a hydrogen facility.
So, to circle back to my original point, Anthropic has not broken ground on actively building any datacenters.
The only way you can see this continue as a model if space based data-centers actually can beat everything land based. And that is very questionable.
> What can he do now that he's got a $85B war chest of cash?
Increase the profit of Micron and NVidia?