XaiJu
Connecting The Dots
Connecting The Dots

patreon


The Future's So Bright (You Gotta Wear Shades)

Elon - sorry, Gorklon - recently posted that he's been thinking about the fastest way to bring a terawatt of compute online - roughly equiv to all the electrical power used in the US today. Following that, xAi employee Ross @rpoo invited people to join xAi


TBH, the following stocks should've SKYROCKETED following Gorklon's and Ross' posts: $NVDA, $TSLA & private XAi
Here's why... (OPTIMUS INSIDE!)


I'll assume Elon's considering this as a practical problem. If you think he's trolling or it's just theoretical, this post is not for you.

$NVDA should skyrocket


... because this signals a hell of a lot of demand for compute.
Whomever thought Deepseek signals a fall in demand for datacenters and GPUs should think again, and then raise by an order of magnitude or two. IMO Demand will only rise in the foreseeable future

I do expect significantly more energy efficient GPUs (possibly quantum) to arrive - possibly from other players besides Nvidia, and there will be cheaper, more efficient models and training, but for foreseeable future, demand will skyrocket, shooting up faster than supply


$TSLA should skyrocket


... because there will be extreme demand for megapacks and probably also Dojo and... OPTIMUS.
It's a double edged sword, bc if Tesla is the customer of these than it's expenses, not revenues, but that is a shortsighted way to view it...


As we've seen with industry's reaction to Cortex 1.0 and Collosus (example: pic), whatever Elon does others soon follow
Sure Tesla could use a lot of megapacks, but so will it's competitors. The energy business should be booming

and I said OPTIMUS...
If Tesla needs so much compute, it's NOT for FSD:
🔸As Chinese launch showed us, not a lot of learning required to enter new countries.
🔸Cars have their own inference, all that's needed for TRAINING are edge cases, and these should get FEWER

No... Growth is for Optimus, and if Elon's thinking a terawatt of compute, imagine how many Optomi he's gearing for... (yep, Billions)
Inference is done in the robots themselves and by "getting a terawatt of compute online" I guess Elon means centralized datacenters - so training. Tesla could sell millions of robots for factory, warehouse and specific business uses alone, with no need for massive training. If they're gearing for so much compute, this spells a huge number of GP robots for retail uses as well.


And xAi valuation?


I think that's obvious. Gorklon knows how to do things efficiently as well as monetize what others cannot. If xAi is building compute, they could use it themselves or sell it to others.






Going back to the start. Elon is gearing up for building a whole lot of compute. This post was written a few hours before the Saudi-US Investment Forum talks were held, in but Jensen's, Elon's, Davis Sacks' and Trump's speeches there further cement the idea.

So whether Elon finds a way to connect a full terawatt of compute online, or maybe just a portion of it - whichever way you look at it, following Elon's post and the Saudi-US investment forum talks, investors of all three companies should be...

Comments

True. Always a race between requirements and capabilities.

Connecting The Dots

No matter how big it is, soon it will not be enough. Such as the way it always goes with power and compute.

ericvfx

The reason I think he referred to a centralized solution - not necessarily a single giant data center, but def data centers, not edge computing on cars and bots - is that he measured the compute in Watts not flops etc. Raising inference compute is not a power problem. Just make millions of cars and bots, and the global grid will either absorb them or grow to accommodate them. It's mostly not a Tesla/X problem, although Tesla does consult utilities on how to scale up. Also, building cars or bots for each chip means it would take much more time to reach a terawatt - not something Elon would have to bother himself with right now. But building huge data centers definitely is a huge power problem, so I'm pretty sure that's what he meant. BTW, in the Saudi-US Investment Fund talks, they revealed that they're about to do a 500 MW deployment with NVidia. (they too measured in Watts, a new trend) 500 MW is huge in current standards (roughly 715k H100s) - yet only 0.05% of what Elon's wracking his mind on🤯

Connecting The Dots

Though I don’t disagree with your premise that Elon is thinking of centralized computing, he didn’t specify. He has also mentioned using Tesla cars’ idle computing in the past. I haven’t run the numbers so maybe this type of computing, regardless if it is in the millions of cars, might just be a rounding error though.

Dax Mickelson


More Creators