How Magnetar Capital Engineered a $4.3 Billion CoreWeave Windfall — and Why Silicon Valley's Biggest VCs Stayed Out

When CoreWeave priced its initial public offering on the Nasdaq in March 2025, the prospectus mentioned a single outside name 157 times — more than twice the count for the company's own chief executive. That name was Magnetar Capital, an alternative-asset manager run by former Citadel traders and best known for its bets on subprime mortgage CDOs before the 2008 financial crisis. At the $40-per-share IPO price, Magnetar's $850 million equity position in CoreWeave had grown to roughly $4.3 billion. A separate $50 million loan the firm had made four years earlier had become a multi-billion-dollar stake. The structure of that financing — and the conventional-wisdom blind spot it exploited — is the story of how one of the most lucrative private bets of the AI infrastructure boom was assembled without a single cheque from Sequoia, Andreessen Horowitz, or any other marquee Silicon Valley venture firm.

The mechanism was a convertible-note architecture collateralized against compute contracts. As The Information reported, Magnetar lent an additional $730 million to CoreWeave against the company's agreements to sell high-performance computing capacity to Microsoft and NVIDIA. CoreWeave has paid roughly $66 million in interest on those loans so far, a sum that represents less than 10% of the principal. The loans mature by the end of the decade. Magnetar invested across every funding round from 2020 — when the company was issuing convertibles at $2 per share — through the Series C at $38.95 and a later tender offer at $47. Magnetar and Co2 also received put options allowing them to sell their stakes back to CoreWeave if the stock declines in the two years after the IPO, an unusual concession that effectively transfers the downside risk to the issuer.

The put-protection clause illuminates the deal's economics for the late-stage participants. Investors in the round that bought at $38.95 lose money if the stock falls below the $40 IPO price. Several funds — including Jane Street, Fidelity, Macquarie, BlackRock, and Newburger Berman — bought $650 million of CoreWeave stock at $47 in November 2024; those positions are now underwater by roughly 15%. Magnetar, by contrast, holds downside protection on a position that is already worth multiples of its cost basis. The asymmetry is not a market inefficiency; it is the negotiated price of capital that the issuer needed when no one else would provide it on reasonable terms.

The conventional-wisdom explanation for why the largest venture funds sat out the deal is straightforward and was articulated directly by Nick Carter, a venture capitalist and personal CoreWeave investor. "The conventional wisdom as a VC is you don't want to invest in capital intensive businesses," he said. CoreWeave's debt-to-equity ratio is approximately 800%, built by borrowing billions to construct data centers serving the demand from frontier AI labs. The leverage is collateralized against the same compute contracts that backstop Magnetar's loans, but the structure would have been anathema to funds whose limited-partner agreements prohibit the kind of balance-sheet exposure that comes with a hyperscale build-out. David Snyderman, Magnetar's managing partner, framed the opportunity in characteristically direct terms: "Sometimes the stars just align. I think we were the first firm to get comfortable lending against that asset called High Performance Compute."

A venture desk's blind spot

The pattern is not unique to CoreWeave. Crusoe, a competitor operating in the same GPU-cloud niche, raised $225 million of debt using existing NVIDIA hardware as collateral — another transaction Silicon Valley venture firms did not lead. The risk profile is closer to a project-finance deal than to a software investment, and the returns are calibrated to a different probability distribution. Magnetar's willingness to underwrite that risk — combined with its track record of trading structured credit through dislocations — made it a natural counterparty. The firm's subprime-CDO experience, widely cited as a cautionary tale, is in fact the relevant skill set: pricing collateral, structuring senior and mezzanine tranches, and surviving a cycle in which mark-to-market losses can erase equity entirely.

That same skill set now appears to be the gating factor in one of the largest private rounds of the cycle. Magnetar is reportedly positioned to write one of the largest cheques in OpenAI's $40 billion funding round, a transaction first reported by Bloomberg. The firm has also launched a venture arm with $50 million of capital from CoreWeave to invest in AI startups in exchange for access to NVIDIA chips — a vehicle that monetizes chip allocation rather than equity upside. The investment thesis has migrated from lending against compute contracts to capturing the scarcity rent that those contracts represent.

A contrarian return profile

The CoreWeave deal also produced an unusual retail-adjacent winner. Les Wexner, the controversial former Victoria's Secret owner linked to Jeffrey Epstein, invested $1 million in CoreWeave through a trust — a position that is now worth approximately $800 million, a roughly 700x return. Wexner holds 3% of the company. The presence of a single large insider with a low-cost basis creates its own overhang, but the mark illustrates the magnitude of the upside available to investors who entered the cap table early. Tane Ward, a venture commentator who has tracked the deal, captured the prevailing view in a single line: "CoreWeave is the web three to AI pivot done right."

The capital-intensity exception

The deal's structure also frames a broader question about whether the AI infrastructure boom can be financed through conventional venture capital at all. Dwarkesh Patel, the podcaster and author who has tracked the economics of AI scaling, put the constraint in numerical terms. "We can keep increasing computers that we're putting into these systems for maybe the next five, 10 years," he said. "But at some point, right now, we're spending 2% of GDP on compute and data centers. You can't just keep 10xing that forever." If compute spending plateaus — and Patel assigns a 10% to 20% probability to a stagnation scenario in which AI does not accelerate for fifty years — the debt that financed the current build-out becomes harder to service. Magnetar's loans are short-dated and collateralized; venture equity is perpetual and subordinated. The risk allocation in CoreWeave's capital stack reflects that asymmetry in stark form.

What the structure signals

The CoreWeave financing is best understood not as a venture investment that happened to use debt, but as a private-credit transaction in which the equity component was a kicker. Magnetar made the loan first, when CoreWeave was a three-year-old Ethereum-mining pivot with no operating history in AI. The convertibles that followed were priced for a company whose access to capital was limited. The IPO unlocked the equity, but the loan — with its $66 million in interest already paid and its multi-billion-dollar collateral position — was the actual trade. The fact that no major Silicon Valley venture firm touched the deal is not a sign that they missed CoreWeave. It is a sign that they recognized the deal was not, in any meaningful sense, venture.

The question for the next eighteen months is whether the structure is repeatable. If the AI infrastructure boom continues, expect more deals of the same shape: convertible debt against compute contracts, put-protected secondaries, and capital pools sourced from credit desks rather than venture funds. If the boom stalls, Magnetar's loans will be the first instruments tested. The firm's investors — and the limited partners who underwrote the OpenAI cheque — are betting that 2026 will look more like 2024 than like 2008. The structure of the CoreWeave deal suggests that Magnetar, at least, has already priced both outcomes into the position.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://www.youtube.com/watch?v=wFONmUOz5sk