Elon won after all
TL;DR
Compute is the real choke point: Theo points to Microsoft, Google, and Anthropic all saying some version of the same thing, demand is there but capacity is not, which means they literally cannot serve or monetize all the AI usage they could have.
Google and Anthropic are renting from a company they compete with: Anthropic is reportedly paying about $1 billion a month and Google about $920 million a month to SpaceX for spare compute, which Theo treats as the clearest sign the shortage is severe.
The bottleneck is an entire supply chain, not just Nvidia GPUs: TSMC fab capacity, HBM from SK Hynix, Samsung, and Micron, hard drives from companies like Western Digital, and even grid power all have to scale together or the whole system stalls.
Consumer hardware is getting squeezed out by AI demand: Theo highlights Micron winding down Crucial's consumer business and his own hard drive prices jumping from roughly $170 for refurbished 16 TB drives in 2024 to $360 plus, with 28 TB drives costing over $600 each.
Elon 'won' by overbuying compute early: Theo argues xAI and SpaceX made an aggressive bet on compute and power infrastructure, including Tesla batteries and gas generators, and now can rent Colossus capacity at rates that could repay a reported $3 to $4 billion buildout in about four months.
OpenAI also benefited from betting early on scaling laws: Theo says OpenAI's more generous usage limits versus Anthropic are partly explained by having secured far more compute ahead of the crunch, while Nvidia remains the biggest winner because demand still exceeds anything it can ship.
The Breakdown
Google is reportedly paying SpaceX about $920 million a month for compute, even while building its own TPUs, which Theo uses to argue the AI compute shortage is far worse than most people realize. His bigger point is that the bottleneck is not just GPUs but the whole stack, from TSMC and HBM to hard drives and power, and Elon happened to overbuy at exactly the right time.
Was This Useful?
Share
Keep Reading
Make Alcreon Yours
Tune your feedFive quick questions, and the feed ranks what matters to you first.Or just get notified
The weekly Echo. Signal worth keeping in your inbox.
Every new piece, announced on X.
Read Next
See all
Playbook
Cheap Models, Hard Tasks
Most agent workflows route every step to the frontier model by default. The bill scales with how chatty the agent gets, even when most steps don't need that brain.

Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.

Playbook
The Art of Tasteful Prompting
Learn how tasteful prompting helps you move beyond generic AI output by shaping context, style, and judgment from the start.