Well, I'm not sure about that, there is still VC money to throw into it. The mag7 is still not going into any serious debt. This is most likely not the top, but we're getting there.

---

Eh? You missed the main ones - if you’re an accredited investor, buy into a fund that bought interests in cap table VC funds. Or go to EquityZen/Forge and indicate your interest. They email you the next time they have some, and you buy it. The former is a bit Wild West, but the latter’s been good in my experience.

---

Yeah that assessment seems pretty accurate. The demand right now is huge, but it’s coming from unprofitable LLM labs. If they can figure out a way to dramatically reduce burn rate it might all work out, but so far they are dependent on VC capital and debt and seem like they’re in a precarious position financially. If one or more of the model companies goes under, a lot of the demand for compute vaporizes overnight

---

Agreed, OpenAI is clearly one of the weak points in a very interconnected ecosystem. A lot of the RPOs getting everyone jazzed at earnings are basically just IOUs from openAI, and bigger companies are making huge CapEx investments assuming those will be honored through the 2030s. VC capital is indeed drying up, and the small and medium sized labs and neoclouds are turning to private credit debt, while even the hyper scalers are issuing bonds. It’s a house of cards that requires flawless execution or it all collapses

---

The “compare launch MSRP vs launch rental rate” lens is useful, but it’s only part of the story. I'm looking at today’s economics. I.e. what people actually paid for H100/A100 hardware vs what they can rent it out for now . See my other comments where I did the math using today's rates. If older GPUs naturally drift cheaper as H200/Blackwell arrive that’s fine ONLY if the owner has largely paid down the original capex. The problem is a lot of this gear was bought in the last 1-3 years at peak pricing during the VC spending spree via these providers. When rental rates get pushed down toward or below full cycle breakeven while the loans are still outstanding, the age doesn’t save you. it just means you’re servicing yesterday’s gear / capex with today’s compressed cash. In that sense falling H100/A100 rental prices are a red flag. They show current supply/demand and ROI pressure on recent purchases, not just a normal generational price. And it’s not like most of this gear was bought for cash. Look at the public disclosures from the AI infra players we CAN see. CoreWeave has raised well over $10B in secured debt facilities for GPUs and data centers, and Oracle AI build out is explicitly tied to a debt load north of $100B. That’s before even getting into Nvidia‑linked financing and co‑signed deals. This isn’t a clean, pay as you go upgrade cycle. It's a lot of leverage that only works if rental economics hold up...which right now they don't seem to be at all. It looks more like scalp demand padding Nvidia’s books than durable, self funding infrastructure...and that can only run so long before the financing and ROI math collide and investors get wise to all of this.

---

I don't think a lot of people understand how similar the current Silicon Valley/VC sphere is to pre-collapse Rome/Greece It's a bunch of wealthborn yuppies taking drugs and regurgitating contemporary philosophy at each other at orgy parties and then barking orders down to their employee peons during the business week

---