Louis Winthorpe III @cfa_life - Twitter Profile

Yesterday I asked @Midnight_Captl about the AI compute trade: - He expects $2-3 trillion per year in hyperscaler capex by the end of the decade - Views the second derivative of hyperscaler capex as likely slowing starting next year, and the absolute dollars keep growing through the end of the decade - Limiter won’t be debt markets (lots of demand to issue debt), it’ll be the equity holders - Corporate America right now is going through pandemonium, everyone is being forced to do more with less or more with the same (regardless of AI usage) - If Meta rents compute, he views it as a yellow flag, because Meta should be one of the best positioned cos in the world to get ROI from internal AI spend - The marginal value of intelligence and thus frontier models is huge, much more than people think - Increased cloud spend makes cloud lockin a more salient concern for enterprises, this is an advantage for GPUs over TPUs (can move between clouds) 00:00 Memory: stocks parabolic but earnings parabolic too (SanDisk) 01:10 What previously drove memory cycles (e.g. Apple iPhone surprise) 02:14 Why this time is different 03:05 Anthropic ARR growth 04:04 The buildout and natural rate of capex expansion 04:29 Why the second derivative of capex slows next year (debt + dilution as the limiter) 09:18 Will the debt markets be the limiting factor? No, equity will be (Meta) 12:19 Google's hockey-stick backlog 14:40 Meta as the economic hurdle rate test for AI spend 18:09 If Meta can't clear it, why would Boeing? 20:20 Are GCP backlog customers actually clearing their hurdle rates? 25:42 Token spend instead of backfilling roles 29:27 How the labs manufactured enough urgency to move corporate America 31:21 Hyperscaler moats and lock-in as cloud spend grows from 2-4% to 10-20% of OpEx 35:44 Why Nvidia/CUDA matters: customers can switch clouds more easily than TPUs 37:25 The marginal value of intelligence is bigger than people think 40:01 Frontier vs local AI 41:12 Cloud-migration risk = minimal while everyone's capacity-constrained 43:15 Semis went from 6% to 22% of the S&P 500 in four years; "Mythos is going to be dial-up" Extra Q&A: A well-known X account you think is solid? @buccocapital A less well-known X account you think is solid? @TMTLongShort is a really excellent thinker. I don't always agree with him, but I always enjoy learning about how he's thinking. A more well-known company you love? Nvidia has to be it for me. I'm a student of Jensen.

8

123

18

208

23K

0

1

0

1

489

Louis Winthorpe III

@CFA_Life

8 days ago

@jaykaeppel SER

0

33

Louis Winthorpe III

@CFA_Life

10 days ago

@ContrarianCurse Double bonus if there are no earnings at all yet as improvement is theoretically infinite.

0

1

0

368

Louis Winthorpe III

@CFA_Life

10 days ago

@Appyg99 Hi, these helped me a little. I wish the same for you. Pet Loss & Healing Support Group - Dakin Humane Society https://t.co/r32nLXXiGa https://t.co/Tzb3d0tmw7

0

1

0

2

953