The Magnetar Trade: CoreWeave (AI) Edition

I've barely slept in the past 72 hours.

All in attempts to understand this invitation to a retail investor holocaust disguised as a promising investment opportunity. 

There may be typos, but they don't matter. 

And you'll understand why very quickly. 

This IPO must be stopped to preserve the integrity of capital markets. 

Capital markets that should provide young people with risk capital, not capital punishment. 

There’s still so much more to share on this that I will update on a dedicated website for this over coming weeks, accessible below. 

Here is evidence that we learned nothing from the GFC. 

Jamie Dimon, faith in American capital markets, and capitalism itself, are at stake here.

And if responsibility drifts towards those that can shoulder it, you’re our Michael Phelps.

This structured (to fail) product masquerading as a tech stock has no place in public markets.

Be the hero this moment needs.

Stop the CoreWeave IPO.

#UnweaveCoreWeave

Please read online here or at https://renaissancecapitalist.beehiiv.com/

_________

Enron with GPUs

The last time I was so disturbed by a tech company's prospectus that it kept me up all night was with Aspiration.

A "green fintech" startup that planned a $2.3 billion SPAC – endorsed by Leo himself – while essentially monetizing environmental guilt. 

After that sleepless night, I warned Scott Galloway this was "the WeWork of ESG," staking my professional reputation as head of research on that assessment. 

Days later, our research was published in a newsletter like this, the SPAC collapsed, and eventually one co-founder ended up behind bars for fraud.

After reading CoreWeave's S-1, I haven't slept for days.

And as I write this, I realize, as cliched as it is, it is the WeWork of AI.

CoreWeave specializes in renting out GPUs that are “powering the AI revolution”.

In practical terms, they offer GPU computing and managed data centers that let companies run and train their hallucination-prone models without having to build their own infrastructure.

The company not only describes itself as "The AI Hyperscaler"…

...they trademarked it.

While still pending and unapproved, it's a flimsy veneer of legitimacy for an even flimsier business.

CoreWeave intended to go public at a $35 billion valuation.

It's not a household name like OpenAI, Microsoft, or Nvidia.

And it never planned to be.

From Defunct Hedge Fund to "AI Hyperscaler"

To understand CoreWeave's true nature, we must examine its peculiar journey from failed hedge fund to crypto mining to "AI hyperscaler."

Founded in 2017 as Atlantic Crypto Corporation, CoreWeave was a cryptocurrency mining operation started by three men with hedge fund backgrounds—Michael Intrator, Brian Venturo, and Brannin McBee. 

None had significant technical expertise in data centers or AI; they had worked together at Hudson Ridge Asset Management, a fund focused on trading natural gas futures that underperformed and shut down in March 2018.

These traders fumbled from a failed gas futures fund to crypto mining, then stumbled into AI.

Despite no technical background, co-founder Brian Venturo was the CTO until July 2024, when he became Chief Strategy Officer, admitting in an interview: 

Yeh, this guy sounds like a real operator.

Hot tip: it’s one of many existential risks for this structured silicon house of cards trying to transition to a business. 

When crypto markets crashed in 2019, they pivoted, renaming the company and buying up tens of thousands of GPUs. 

CoreWeave offered these GPUs to the (at the time) much smaller group of companies using them for tasks like 3D rendering and data analytics. 

This was a modest business with CoreWeave making just $15.8 million in revenue in 2022 with losses of $31 million.

Then ChatGPT hit.

“The Architects”

Let’s meet the masterminds rigging this high-stakes heist, a cast straight out of a Wall Street crime drama where the loot’s measured in GPUs and the victims are innocent people that – like me – want to trust the branding and leadership of JP Morgan, Morgan Stanley, Blackstone, and Goldman Sachs (last one’s sarcasm, obviously):

• Jim Prusko, Magnetar Senior Portfolio Manager: A 2008 Magnetar Trade vet, Prusko built toxic CDOs, bet against them, and cashed billions while the housing market burned. He dodged a 2003 market-timing scandal at Putnam Investments that sparked fraud charges. He’s a system-gaming maestro, and CoreWeave’s his latest canvas.

• Ernie Rogers, CoreWeave’s "Chief Architect, Strategic Financing": Magnetar’s ex-COO and ex-board member, Rogers’ title is a neon sign screaming financial puppeteer, not tech guru. He’s snagged a $450k salary, a potential $450k bonus, and about $30 million in stock vesting over two years. Fired without “cause” or quitting for “good reason”? That $30 million’s still his, hush agreement pending. He’s in, if Magnetar lets him go. His gig? Juicing the books—those $755.9 million “fair value adjustments” boosting EBITDA reek of his handiwork.

• Jessica Damrat, CoreWeave’s "SVP, Strategic Financing": Magnetar’s former CFO of their management arm, Damrat’s role is pure financial alchemy. She’s lined up for $450k in salary, another $450k in bonus potential, and roughly $10 million in stock over two years. Same deal: sacked or walking with “good reason,” she grabs the $10 million with a gag order. Her job? Keeping the (AI revenue ouroboros) musical chairs spinning.

With Rogers and Damrat jumping from Magnetar to CoreWeave, it’s not just influence—it’s a takeover. 

“Non-core” financing, you say?

Seems like the beating heart of this silicon sham – and a very expensive one. 

If the name Magnetar sounds familiar, you were scarred from the financial crisis in which Magnetar played its classic combination of both arsonist and firefighter. 

You may also recognize the name from The Big Short.

Magnetar partnered with patsy Wing Chau (later SEC-busted for fraud) to pack CDOs with junk, betting against them as the world imploded. 

They hold 34.5% pre-IPO, $544 million in debt, and a bizarre $50 million “investment” from CoreWeave into their coincidentally newly launched AI venture fund (no joke).

These aren’t architects of AI. 

They’re architects of financial exploitation at industrial scale, preying on the vulnerable and those not financially illiterate…

….in convoluted capital & governance structures, warrants, puts, converts, covenants, market mechanics, market structures etc.

(No joke, I had to have an LP of mine that invented structured products explain some of this to me).

Luckily the esteemed JP Morgan would only feed this type of thing to sophisticated investors. 

Not a range of opportunistic funds and clients that will flip it to innocent people that bought into the institutional narrative that it's the next NVIDIA.

And why wouldn’t they?

If it weren’t, surely Goldman Sachs wouldn’t be endorsing unscrupulously selling it to anyone with a wallet (heartbeat optional) like it was 2007. 

Two people they’re not selling to?

Fidelity and Blackrock.

While both are significant institutional buyers of IPOs, they were actually already shareholders.

But not anymore.

They dumped $1.3 billion in secondaries before this initial final public offering and stain on [IM NOT GOING TO SAY IT]’s reputation.

Maybe two of the world’s biggest investors needed the cash.

Or maybe they realized they got to know the crew behind Magne(Death S)tar.

(Lack of) Fundamentals

Despite seeing every corner of the amusement parks that are global capital markets, nothing could prepare me for the house of horrors that is the CoreWeave S1. 

I’ve worked at the banks on this deal (Goldman Sachs and Deutsche Bank), private equity, advise hedge funds, and been across every funky capital structure and structured deal you could think of. 

I’ve even personally coinvested with Macquarie Capital, another investment bank on this deal. 

But even I struggled to get my head around this scandalous prospectus. 

We could talk about CoreWeave’s explosive revenue growth. 

We could talk about how CoreWeave’s sales & marketing expenses only rose 42% to $18 million in 2024 while revenues grew 737% to $1.9 billion. 

We could talk about the (AI) revenue ouroboros in the orgy of clients, clients that are suppliers, and clients that are competitors.

We could talk about how CoreWeave spent ~$3bn to earn ~$2bn revenue, part of which seems to be debt (more on that later).

We could talk about its valuation as a multiple of those “revenues”.

Multiples which have somehow been pegged alongside names like Palantir.

I’m sure the founders hate being mistaken for (actual) tech founders Alex Karp, Peter Thiel, and Joe Lonsdale.

We could talk about how depreciating GPUs in a straight line over 5 years is nonsense, and how they decided to extend it to 6 years.

We could talk about “adjusted EBITDA” that omits non-non core financing. 

But what’s more telling? 

How interest expenses rose 1,170% to $361 million.

Why? 

Because this isn’t as much a company as it is a ticking timebomb of debt rigged to detonate on uninformed investors by design. 

And in this devilish design, who better to share grand expansion implosion plans with than Core Scientific, a conception so immaculate you’d think Jeffrey Epstein was its sperm donor.

Core Scientific went public via a Chamath in 2022, the same year it filed for bankruptcy, and reemerged from Chapter 11 in January 2024.

Being the power hungry guys they are at CoreWeave, they decided to get on the Magnetar trade bro and signed agreements with Core Scientific for “more than 500 MW of capacity." 

The issue? 

Core Scientific's primary business is bitcoin mining, not AI computing.

As of February 2025, Core Scientific had approximately 166,000 bitcoin miners—application-specific chips that only mine bitcoin and have nothing to do with GPUs or AI compute. 

While this hardware can technically be repurposed for training foundational models, it is less performant and more expensive to do so.

Core Scientific scraped together a measly $24.3 million in high-performance computing revenue in 2024—pocket change next to its $408 million from crypto mining—yet CoreWeave’s somehow locked in over 500 MW of capacity from them.

Core Scientific’s “1,317 MW of contracted power capacity” also mirrors CoreWeave’s “approximately 1.3 GW” like two drunks propping each other up.

How a company that just emerged from bankruptcy, with no meaningful AI infrastructure, is supposed to deliver over a gigawatt of capacity to CoreWeave is, however, irrelevant.

This equity offering has a preordained fate, a byproduct of structures and structured products that are meant to suffocate and obfuscate, creating a structured product of its own.

Specifically, a structured-to fail-product.

Because in the neocloud, all numbers are in the air.

CoreWeave admits their internal controls are a “material weakness”.

Their poor controls, financial reporting, and IT systems botched their books for 2022, 2023, and 2024. 

Errors galore, but don’t worry, they’re “remediating”... by 2026, long after your money’s in. 

No effective control environment? 

In a company with $8 billion in debt (just guessing on the Magnetar abacus to oblivion at this stage) and a $35 billion valuation?

 And $1.9 billion in apparent revenue, some of which they have to pay back to a shareholder with guaranteed interest?

What could go wrong right?

Proprietary Tech Nvidia Leftovers

CoreWeave floats in a GPU distortion field, where Nvidia’s name shields it from reality. 

It’s value derives derived from its relationship with Nvidia, which holds a reported 1% stake (down from earlier reports of 5%), giving CoreWeave access to scarce GPUs.

So absent any "proprietary tech" they have only preferential GPU access. 

Lose that, and they're just another generic cloud provider with nothing but unserviceable debt.

It competes with hyperscalers like Amazon Web Services (AWS), Microsoft Azure, and Google Cloud, while paradoxically, Microsoft is its largest customer, accounting for 62% of 2024 revenue. 

Having such competitors, whose flywheels spit out nearly offensive levels of free cash flow, is a suicide pact. 

Flywheels wde not only subsidize the cost of AI investment, but drive the commoditization of foundational models.

Foundational models that the (actual) AI Hyperscalers will eventually acquire, like Amazon will Anthropic (the Google investment in Anthropic was very clever btw). 

That commoditization has meant that breakthroughs and new model releases barely get news flow, which impacts the forward outlook of a core (only?) market for CoreWeave. 

And unlike its Client-Suppliers^TM and Client-Competitors^TM in NVIDIA and Microsoft, CoreWeave isn’t in the business of breakthrough technology, or technology at all. 

(See above, non-non-core financing.)

Said differently, this house of silicon cards falls over if Satya Nadella sneezes or Jensen says it’s GPU, not me. 

And both seem likely. 

The Governance Show (All Strings Attached & Collateralized Separately)

CoreWeave’s governance is a puppet show with extra strings – held by Magnetar, but like absolutely everything else, it’s charged to CoreWeave’s refugees for future shareholders.

Thankfully they have revenues that are out of this world, which brings us to the first of many governance gut punches. 

Magnetar’s $230 million MagAI prepaid revenue is refundable at up to $398 million. 

If it could speak, it might say it identifies itself as a ridiculously expensive shareholder loan trying to be recognized as revenue, adding bloat to the Remaining Performance Obligations (RPOs), the performant cousin of the now deceased Collaterized Debt Obligations (CDOs).

(Magnetar is an inclusive adopter of all forms of obligations and doesn’t see color any concern for the harm they may cause innocent families, their communities, or the global economy.)

Voting rights in tech companies led by visionary founders tend to command more voting stock, and tend to outperform because of it.

But these guys aren’t Zuck and Elon. They’re reformed excel jockeys. 

Yet they wield 79% of voting power.

And that’s after having sold out ~$500 million pre-IPO. 

The next governance failure might be my favorite: CoreWeave, a money losing furnace levered up like Lehman Brothers “invested” $50 million in Magnetar’s venture fund. 

Because, obviously, a $15 billion hedge fund needs a PortfolioCo-LimitiedPartner^TM in there. 

Magnetar’s 34.5% stake, $544 million in debt, and boardroom muscle make them the puppet master. 

A 12-month lock-up with them (not underwriters) is twice the norm, giving them leverage to orchestrate exits while founders dance to their tune.

Then there’s the OpenAI Dowry. 

A March 2025 Master Services Agreement with OpenAI dangles $11.9 billion in compute promises—great, until you see the $350 million in Class A stock gifted at IPO price, diluting everyone else. 

Days before the IPO, OpenAI announced a Magnetar investment. 

Coincidence? 

Or a tangled web where Sam Altman’s crew gets a cut for legitimizing this financial Frankenstein across the IPO finish line?

This isn’t governance. 

It’s degeneracy.

A casino where complexity hides the house’s edge, and who even really owns the house.

Magnetar Trade 2.0

Back in 2008, Magnetar made a killing by pumping up shaky mortgage deals with equity buys, then betting against them with credit default swaps as the housing market cratered. 

Fast forward to today: CoreWeave’s IPO is their new sandbox, and they’ve rigged it with the same cunning. 

Whether CoreWeave’s stock soars or the company defaults, Magnetar has woven a web of financial instruments that ensures they win: debt, warrants, registration rights, and a prepaid revenue they can clawback. 

(I realize that makes no sense, that’s why I’m writing this).

Magnetar holds warrants—think of them as coupons—allowing them to buy 12.1 million CoreWeave shares at the IPO price of $50 each. 

And let’s assume CoreWeave’s IPO pops as it's financially engineered to. 

If the stock hits $100, they can exercise these warrants, buying shares at $50 and selling them immediately at $100. 

That’s a $50 profit per share, totaling $605 million across their holdings. 

And if it really spikes, say to $200, it’s a gain of $1.8 billion.

Magnetar “registration rights” also seem to enable them to sell their shares without waiting through the usual post-IPO lock-up period that binds other investors. 

If the stock surges and they sense a peak, they can unload millions of shares quickly. 

In this scenario, Magnetar’s warrants and fast-exit rights turn a rising stock into a windfall. 

They ride the wave up, cash out at the top, and leave others to navigate the aftermath. 

And if CoreWeave stock craters, Magnetar still wins. 

Magnetar's $544 million loan is secured by the company’s GPUs. 

These loans come with covenants, strict rules CoreWeave must follow, like hitting revenue targets. 

If CoreWeave slips, Magnetar can demand the full $544 million back on the spot. 

Can’t pay? 

Magnetar seizes the GPUs. 

In today’s GPU-hungry market, they could fetch $25 billion, covering their loan and a small nation state. 

Magnetar’s position higher up the capital caste brings it other benefits. 

They can trigger a cascading cross-default, meaning one missed covenant could unravel CoreWeave’s entire financial Frankenstein. 

Suddenly, all debts come due, and Magnetar’s debt starts calling, and owning, all the shots. 

This amplifies their power to profit from failure, turning a small misstep into a full-blown takeover of CoreWeave’s assets.

What else can they clawback anytime they like? 

CoreWeave’s “revenue”. 

Specifically, that (very) expensive shareholder loan masquerading as the innovative MagAI vulture venture fund. 

If CoreWeave falters, they don’t just get a refund—they could claim up to $398 million, a $168 million bonus baked into the deal. 

In a default scenario, Magnetar can seize assets, force a collapse, or cash out with a premium, while other shareholders bear the brunt. 

And guess what? 

CoreWeave already has. 

The (Practice) Pump and Dump

In 2021, CoreWeave held a stake in TTM Digital Assets, which merged with Sysorex via a reverse triangular merger. 

According to SEC filings, CoreWeave traded 3,130 GPUs for 28.65% of TTM, receiving 35.58 million SYSX shares (though a contradictory 13D filing mentioned only 14.2 million shares).

What happened next looks like a pump and dump. 

SYSX stock surged 1,000% on the merger announcement on April 14, 2021, only to crash immediately afterward as CoreWeave began selling shares. 

By May 2022, CoreWeave had reduced its ownership to below 5%, leaving retail traders with losses exceeding 99%.

Fast forward to the present, and we see a concerning pattern continuing. 

So if CoreWeave is the backbone of AI, Wall Street needs to find a spine.

Don’t Get WeWorked,

Daniel

If you enjoyed this post or know someone who may find it useful, please share it and encourage them to subscribe here or at https://renaissancecapitalist.beehiiv.com/