Gurley says he wouldn't have believed Mag Seven would turn $50-100B annual free cash flow to zero via CapEx.
“If you told me five years ago that these, mag seven would become worth $3,000,000,000,000 and then turn around and take their free cash flow from 50 to a 100,000,000,000 a year down near zero because they were gonna spend it all on CapEx, I'd have been like, no way. Like, I wouldn't have believed it.”
Patel explains scaling laws require ten times more investment for each model iteration with specific dollar amounts
“A log I e, it takes 10 x more investment to get the next iteration. Well, 10 x more investment, you know, you know, going from 30,000,000 to 300,000,000, 300,000,000 to 3,000,000,000 is relevant.”
Gurley notes shifting narrative suggesting inference scaling is preferable to training CapEx.
“There was a podcast recently where they kind of flipped everything on their head and they said, well, if we're not doing that anymore, it's way better because we can just move on to inference, which is getting cheaper and you won't have to spend all this CapEx.”
Gurley notes hyperscalers now spending $200 billion in total CapEx with questions about return on investment.
“In fact, I think if people are gonna be critical of anything, a lot of people are looking at the total CapEx of the hyperscalers now at $200,000,000,000 and saying, when are you gonna get a return on the dollars that you're spending?”
post Gurley suggests the commenter would enjoy living in New York, responding to concerns about data center tax revenue concentration
https://x.com/bgurley/status/2087231641857282195