$CRWV IPO: The AI Trade Just Got Cheaper

Flat launch. Full throttle thesis. I’m in—and staying in.

Ok. First thing’s first. I thought (futuristed?) that the CoreWeave (heretofore referred to as CRWV) IPO would be a grand slam; a raise the range, price above, first day pop kinda deal. (See: “Netscape moment.”) I was wrong. And when I’m wrong I will concede that I am wrong. But….I will also note three things (because there are always three) and a disclaimer that I am and remain at the time of this writing a CRWV investor via a pre IPO SPV.

1. I did refer to this as the “YEAR of the GPU Cloud IPO” not the “DAY.” So I can bullshit my way through that on a technicality (sorta).

2. As I’ve said before, fortune tellers like myself (and my close friend Nostradamus) are nearly always right because we say things cryptically and if we wait long enough we’ll be right at some point (or dead and it won’t matter).

3. Big picture: nothing has changed. My thesis remains intact and is happening. And, for reference, that thesis is:

We are amidst a fundamental shift in computing and software architecture from a traditional front end / logic / database stack to one that is based on AI- and LLM-like architecture that collapses all of those layers into a single logic layer.

That logic layer MUST sit on GPU and GPU-like silicon because legacy processors and data centers are not capable of managing these workloads.

This means a complete rebuild (not upgrade; a rebuild) of the current data center infrastructure that will require $1-2T (that’s trillion) or more investment over the next ten years.

So what happened? Three things (because there are always three):

1. The overall market has tanked (thanks, Donald) and specifically AI, i.e., NVDA is down ~30% from its recent (and all time) high (thanks DeepSeek / China / scaredy cat investors).

2. The press and analysts absolutely beat the living shit out of CRWV in advance of the IPO - a combination of clickbait and genuine / fair questions to ask about the company.

3. CRWV (or rather its bankers and investors) priced the IPO at $40/sh, below the bottom of the $47-55/sh range and they cut the deal size as well. It opened, broke, drifted just above the IPO price, and closed its first day of trading at $40.00 - exactly the IPO price, like magic.

I’m not going to begin to decipher or guess wtaf the “government” may do that will move markets in either direction. But I can (try to) comment on some of the critical reports on CRWV that I’ve seen or heard. Here goes….

1. “Waning AI Demand” - Those writing about this clearly are not anywhere close to reality or what’s really happening on the ground. I am (on the ground, that is; my attachment or otherwise to reality has always been dubious). And here is what I see and hear…. Demand is booming. AI is advancing. Models are getting bigger and better. More use cases are appearing. Enterprises are adopting AI faster than any other platform I’ve ever seen in my life. Every new phone, computing device, TV, appliance, toothbrush, etc is being rebuilt ground up with native AI. The world is spinning faster etc etc etc. (The last couple may be a stretch, but you get the point.)

2. “Canceled Contracts” - The FT reported in early March that Microsoft was canceling contracts. MSFT is indeed CRWV’s largest customer, representing a whopping 62% of last year’s revenue, which is, for sure, a gigantic risk. But CRWV went so far as to issue a press release the same day denying any contract cancellations from MSFT. The nuance may be reports that MSFT didn’t sign a $12B (additional) contract with CoreWeave, but OpenAI did sign a $12B contract and invested $350M into the company at the last round pre IPO valuation. And also recall that MSFT had been rumored to be using CRWV to satisfy its GPU cloud commitments to OpenAI. So it is possible that MSFT did walk away from the additional $12B contract and that OpenAI did “swoop in” to take that contract, but that could also mean that OpenAI went directly to CRWV for service rather than through a Microsoft Middle Man (well, person, but… alliteration).

3. “GPU Obsolescence” - From Seeking Alpha: “CoreWeave sits on 250,000 GPUs from last season's Hopper, which is already outdated with Blackwell and will become ancient next year…..” Yes, Jensen Huang calls himself NVDA’s Chief Revenue Destroyer because the company’s new generations are always far superior to existing generations. (He is quite the showman.) But saying CoreWeave’s (or anybody’s fleet) of 250K H100s are ancient or unusable or otherwise worthless is patently absurd. I’ll try and keep this short but…. The actual useful life of most of these chip generations is about 7 years (yes, 7 years, maybe more), not the 4-5 years over which they depreciate them, and certainly not the 1-2 years since introduction to new release. Aside from how long it takes to get new (Blackwell, for example) fleets up and running and how CoreWeave is among the most favored recipients for the chips given its close relationship with (i.e., significant ownership by) NVDA, raw performance is not the only measure buyers look at. More important is price versus performance. And older generation chips can outperform newer generation chips on price / performance basis often; see here for a (very) technical explanation.

4. “Collapsing H100 Pricing Dooms CRWV” - From what I understand and from what I’ve worked on in GPU acquisition financing exercises, the “typical” structure is borrowing with the chips as collateral but against contracts signed to rent those instances for a period (typically 1-3 years). And my math (from then) was that the contracts essentially covered the (entire) cost of the chips so when pricing comes down (or collapses) the GPU cloud provider is still making money (not just revenue) on those chip purchases (because the GPUs were already paid for by those shorter term multi year contracts). Think of it this way: borrow money to buy the chips, sign a contract to pay for the cost and financing of those chips (entirely), continue to extract value from those chips after they’re entirely paid for - like gravy. (And see above for the real useful life of these things.)

5. “Long term property leases and short term cloud contracts” - It is accurate (I think) that the company’s leases are (far) longer than their GPU cloud contracts. But remember that the most material cost is chip acquisition - NOT leasing a building. Is there risk? Yes. Is it significant? Maybe, but not nearly as much as it sounded. (TLDR: this isn’t WeWork.)

So what now? Three notes (because, well, you know….)

1. As per above and repeated ad nauseum in past posts and nearly any conversation I’ve had before or after the CRWV IPO: my GPU cloud thesis remains intact.

2. CRWV did almost $2B in revenue in 2024 and I’ve heard estimates ranging from $4-8B for 2025. Big range, but it’s 100%+ growth regardless on a business that can (easily) render cash flow, depending upon growth investment. (See what I did there? Hint: “render.”)

3. Remember another misunderstood IPO once called Facebook that broke (badly) because investors thought their entire user base was moving to mobile (it did) and that they had no revenue model for mobile (they did). The IPO valued FB at ~$100B; it halved shortly after because of the aforementioned misplaced concern; and now the valuation is up ~10X from the IPO price (and ~20x from the bottom). I wrote about it when it cratered here. (And I took some obnoxious victory laps here, here, and elsewhere.)

Will CRWV drift down or even collapse to half its current valuation in the short term? Maybe, but I hope not. (I was and remain an awful trader / short term market timer.) But I still submit the company is worth $50-100B (i.e., at least 2x the IPO price) and possibly more, especially if CRWV remains the top rated player in nextgen / neo / GPU cloud as SemiAnalysis said it was only last week. (If you don’t read SemiAnalysis, read SemiAnalysis - easily the best resource for any sort of information on this space.)

As for me and CRWV? I still own it (through a pre IPO SPV). I can’t sell it right now if I wanted to. But I don’t want to sell it and I wouldn’t sell it here now if I could. I don’t give out investment advice, but I can share my opinion (above) and my track record (here): I’m always right… in some way and at some time - and if not, I’ll be dead at some point and it won’t matter anyway.