2025: The Year of the GPU Cloud IPO(s) - Les Prophéties de 2025 du Porch(dog)....

2025 will be the year of the GPU cloud IPO(s). Here is how it will go (up and) down (and up again) and why it matters and what to do:

• CoreWeave will go public in (early) 2025. (Disclosure: I am a CoreWeave investor via a private market SPV, but this information is from public sources.)
• If the $8B 2025 CoreWeave revenue projections are remotely close, the valuation will be (way) higher than the aforelinked? $35B IPO speculation. Think: $50-100B (and see above investment disclosure again so I don’t go to prison).
• CoreWeave will be the only pure play AI (infrastructure) stock other than NVDA, whose $3T+ valuation suggests investors noticed that one already.?
• CoreWeave’s soaring valuation will whet investors’ appetite (ignite their insatiable lust?) to pour more gasoline-soaked dollars onto the billowing AI investment inferno so private companies like Lambda and Crusoe will come to market in triumphant blazes of glory. (Disclosure: I am a Crusoe investor via a private market SPV.)
• Any public company (kindling) with even passing exposure to GPU cloud will catch fire and rally while bankers trip over their own drool to do private placements, secondaries, and other financings. Look at stocks like: DOCN, NET, AKAM, EQIX, RXT, FSLY, NBIS, APLD…. (Disclosure: I have a teeny position in NBIS for fun; also, I took EQIX public, but that was over two decades ago.)
• Proximate spaces like crypto / blockchain and any seemingly AI-related hardware will feel the life-giving heat so look at publicly traded stocks like RIOT, HUT, and BTBT in the crypto (to GPU cloud) thesis and SMCI (not new) and CAN for hardware (to AI hardware) ideas. Bankers will trip and drool and swoon over these types of companies as well.
• If history is any guide (and From the Porch says it always is), this will be followed by a(n overbuild-related?) bubble-bursting bankruptcy bonanza after which everything will grow again and my $1-2T digital infrastructure rebuild prediction will play out; i.e., I’m gonna be right (again). Trust me (again).

TLDR: I’m doubling (tripling? quadrupling…?) down on predictions I made here, here, here, and here.

Exactly 367 days ago today I penned (and posted) Les Prophéties du Porch(dog) so today was only fitting for a 2025 update. Here is some (other) background and then I’m diving (very) deep into the weeds - read further, if you dare, but consider yourself warned….

In 2011 I wrote that social / mobile / cloud (dare I say Web 2?) would change everything again in a 1999-esque bacchanalian hysteria of technology and stocks and jobs and everything that comes with that. This (retrospectively correct) prophecy was based on what I experienced from 1995-2000 (i.e., Internet bubble). I earned more in the last year of that period (2000) than I have during the rest of my life. Wanna hear how I did it? Of course you do. Because it’s happening again and you can play it the (sorta) same way (maybe). (And remember, none of this is official investment advice in any way, shape, or form; got that, SEC?) Here goes….

Our first history lesson of the year.

Netscape's August 1995 IPO marked the beginning of the dot-com boom with its $14/share offering price doubling to $28 just before the S-1 ink dried. The opening trade was $71 and it closed just above $58. A fitting start to the era with a near $3B NSCP valuation (and retrospectively anticlimactic acquisition by AOL for what would then be $10B in stock three or so years later; foreshadowing? Doesn’t matter right now.) The “Netscape Moment” subsequently inspired ~200 IPOs in 1995 and, during 1999 and 2000 (the tail end of the bubble; aka “the blow-off”), ~400 “Internet” companies came to market. For perspective, there were a little more than 200 US IPOs in total (of any sort) in 2024.

This is how I played it (i.e., tripped ass backwards into good fortune).

1. Dotcom bubble: I was a research analyst (which in the 90s meant I was an investment banker, whether I realized it or not). TLDR: Internet stocks went crazy.?
2. Yes, I was early on “the Internet call,” (the gap between the 100% of US homes with computers but only 20% of those connected to the Internet would close to zero), but I didn’t bank AOL or Netscape or Amazon or Yahoo or many of the premier Internet stocks of the time. (I did cover them with requisite Buy ratings, but still….)
3. The “differentiating” calls I made, instead, were on the other Internet stocks, which led to a lot of investment banking deals and a lot of investment banking fees and a few big investment banking contracts for me before being unceremoniously laid off (i.e., fired) near the peak, which triggered an acceleration of all my (bank) stock vests and contract payouts that were due to me over the next four years. Tada.

Ok so here are the categories and stocks to keep in mind for comparison….

Tier 1 (Internet) stocks - stocks from the second point above:

• AOL.
• NSCP.
• YHOO.
• AMZN.
• Etc….

Tier 2 (Internet) stocks - stocks from the third point above:

• MSPG.
• ELNK.
• PRGY.
• GNET.
• Etc….

I wasn’t the first (or last) to cover any of those and other names, but with the world already making money on AOL & Co., I combed over public company listings for comparables (and found them in stocks like MSPG, ELNK, and GNET) and through private company lists for operators like Prodigy and Equinix. The call on each was something like: “If [AOL/YHOO/AMZN…] is trading at X, then [ELNK/MSPG/OZEMY…] must be at least worth Y….” And for a time, I was right. And during that time, those T2 companies raised giant piles of cash and paid (part of those) piles of cash to the investment banks, including those where I worked (namely: Furman Selz and H&Q, which is now called JPMC). And then everything fell apart just after I was let go. Phew / tada.

So where are we now?

Tier 1 (AI) stock”s”:

• NVDA.

Private / Pre-IPO companies:

• CoreWeave.
• Lambda Labs.
• Crusoe Energy.
• Etc….

Tier 2 (AI) stocks:

• EQIX.
• SNOW.
• NET.
• AKAM.
• NBIS.
• DOCN.
• APLD.
• MARA.
• RIOT.
• CLSK.
• WULF.
• HUT.
• CIFR.
• BITF.
• BTBT.
• ARBK.
• EBON.
• SMCI.
• CAN.
• Etc…?

Today, the only publicly traded, pure play, large / mega cap, liquid way to invest in AI is Nvidia, which is up about 10x in the last four years. If you (ironically) Google or ask ChatGPT for (other) top AI stocks you’ll get lists that include MSFT, GOOG, META, AMZN, IBM, AAPL, TSLA, ADBE, etc. Sure, they are all heavily invested in and building wonderful products and services in AI, but that’s basically a list of FAANG / Mag7 and other top tech leaders; all of which: (1) aren’t native AI companies and (2) have had pretty nice runs over the years as well.

So what are the “other” stocks with (potential) AI exposure? Three themes to start with:

(1) cloud -> AI infrastructure.

(2) crypto -> AI infrastructure.

(3) hardware -> AI infrastructure.

See a trend? <- AI infrastructure. Representative playbooks for each are DigitalOcean CoreWeave, and Lambda.

CoreWeave will be this cycle’s “Netscape Moment.”

Assuming CoreWeave’s IPO goes as planned (prophesied?) in 1H25, we will have our 2025 “Netscape CoreWeave Moment.” And to quote Ed Hajim, my first hedge fund portfolio manager and partner: “Katy, bar the door.” (Fwiw, I always took that to mean that big things were about to explode, but the Internet just told me that it means: “Take precautions; there’s trouble ahead.” Foreshadowing? Doesn’t matter right now.) The market will crank out all the IPOs it can, but if you can’t get your hands on private company shares, you may be able to play it via the cloud -> AI infrastructure theme.

The table above lists a handful of cloud infrastructure stocks (that I know next to nothing about). EQIX, a data center company, is already nearing $100B in valuation and SNOW, which calls itself the AI data cloud, is nearly half that already. NET, a CDN and security company, fancies itself the “AI inference at the edge” leader (though I’m not certain any of that really exists yet). Also, a few months back NET was rumored to be eyeing DOCN, a pure play cloud computing company that in 2023 acquired Paperspace, a pure play GPU cloud company. (Ask me offline about why I bought and eventually sold DOCN stock over the last 2 years; spoiler: it involves this theme.) AKAM is the granddaddy of all CDNs and in 2022 acquired Linode, another pure play cloud computing company, for ~$900M. NBIS is the “not Russian anymore” pure play GPU cloud spinoff from Yandex (NB: I own a few NBIS shares). And APLD is suddenly an AI cloud company that did a $160M private placement last year that included NVDA (which is also a gigantic CoreWeave backer).

Don’t like the cloud angle? Here’s another….

Crypto -> AI Infrastructure.

CoreWeave started in 2017 as a cryptominer. Its pivot to AI infrastructure sparked its GPU cloud revenue to go from $0 in 2022 to $500M in 2023 to a forecast of $2B in 2024 and media reports have speculated 2025 revenue could be as high as $8B. JFC. A year ago I failed at an attempt to buy private stock at a $9B valuation, but did manage to grab some (via an SPV) at $23B more recently. IPO speculation this year is a ~$35B valuation, but if that $8B is close to real then I don’t see how this company doesn’t IPO at $50-100B.

Taking note of CoreWeave’s move, nearly every crypto miner in creation decided a pivot to AI infrastructure wouldn’t be a bad idea. Investors (and bankers) took note. To wit, Starboard Value took a stake in crypto miner RIOT to compel a pivot to AI infrastructure, HUT popped on a $12B data center announcement and rumored META deal, and BTBT hired GPU cloud (and Paperspace / DOCN) alumni Ben Lamson and Tom Sanfilippo to run their fledgling GPU cloud infrastructure. The table above is a collection of publicly traded crypto miner stocks (that I know next to nothing about) that could be swept up in this trend. (NB: switching from crypto mining to GPU cloud is, to put it mildly, nontrivial; BTC miners use ASICs not GPUs so there likely isn’t a chip supply advantage if they’re mining BTC; and designing and operating data centers with the uptime requirements to which customers have grown accustomed is… hard; that said, these companies usually have access to some combination or hardware, space, and, most important, multi megawatt energy sources.)

Crypto / blockchain PTSD? One more idea….

Hardware -> AI Infrastructure.

I’ll confess upfront that hardware isn’t my strong suit, but let’s look at Lambda Lab’s transformation from hardware supplier to GPU cloud provider. The company started a dozen years ago selling GPU-accelerated hardware for AI / ML (i.e., they sold boxes). Hardware was essentially all of its revenue for most of that time until the company noticed that cloud services might be a more lucrative direction. Similar to CoreWeave (though not quite as large or well financed), over the last few years the company went from close to zero dollars in cloud revenue to several hundred million dollars and I’m (literally) guessing that number will near or eclipse $1B ARR any day now.

I don’t have a great handle on how to play this angle; SMCI makes AI server boxes and has a $20B valuation, though the company’s valuation was once 3x that and has since been plagued by auditor and China-related issues. CAN is a Chinese AI server box and ASIC (for crypto) maker, but, well, China (again). Is it time to revisit DELL, HPE, IBM, et al? Other chip manufacturers like AMD (not a secret), INTC (no), AVGO? MRVL? Despite once trading semiconductor stocks (professionally) for a year or more, I have no idea (which is perhaps why I didn’t do it that long). But maybe worth a deeper dive.

I'm a cowboy; on a steel horse I ride….

So as 2025 unfolds and CoreWeave’s forthcoming “Netscape Moment” ignites the blaze of IPOs, exuberant valuations, and feverish investor enthusiasm, the parallels to the dot-com era will be impossible to ignore - for better (bull market bubble!) or worse (infrastructure overbuild crash). But history reminds us that the infrastructure left behind after every bubble fuels the next wave of innovation. So, whether you’re betting on GPU clouds, pivoting crypto miners, next gen AI hardware, or the upstarts, remember that it’s all the same and only the names will change. The seeds of the next trillion-dollar rebuild are being sown in the (GPU) clouds of today.