Under the Hood of CoreWeave’s IPO

Chip arbitrage, is CoreWeave (NASDAQ: CRWV) worth the $40/sh price point at IPO?

CoreWeave went public at $40 per share with a promise: it would be the GPU infrastructure layer powering the AI revolution. The pitch is seductive. With demand for compute at all-time highs and Nvidia chips in short supply, CoreWeave offers an elegant solution — buy GPUs in bulk, deploy them into purpose-built data centers, and rent them out to hyperscalers, startups, and enterprise labs at a markup. The problem? That arbitrage window may already be closing.

CoreWeave doesn’t make chips. It doesn’t build software platforms. It isn’t a vertically integrated cloud like AWS or Azure. It’s a middleman, albeit one that currently benefits from early access to Nvidia’s most in-demand chips like the H100 and GH200. Its business model hinges on the assumption that demand will continue to outstrip supply and that competitors won’t bypass it. But OpenAI is reportedly developing its own chips, codenamed "Stargate," while Microsoft, Google, Amazon, and Meta have all built in-house accelerators. Even Nvidia could shift to direct sales. If that happens, CoreWeave loses its edge.

If AWS is a full-service dealership, offering a bit of everything for everyone, from economy cars to luxury SUVs, CoreWeave is a specialty race shop. It doesn’t cater to the average driver. It’s built for performance at the edge, crafting custom, high-octane machines designed purely for speed and precision. Coreweave only builds elite compute environments for bleeding-edge AI workloads, custom-tuned, high-speed, and expensive. That’s great when everyone is racing, but the second buyers start building their own engines, or the parts become commoditized, the shop risks becoming obsolete.

CoreWeave has raised a staggering $1.5 billion at IPO for a $27B valuation. Its biggest backer is Nvidia, which holds a roughly 6% equity stake and is deeply intertwined as both supplier and strategic partner. Others like Magnetar and Coatue are also in deep with $350M and $1.1B of private capital deployed respectively. But the capital has come at a cost. According to its S-1, CoreWeave raised $7.5 billion in 2023 alone, much of it tied to hardware-backed loans, and they are currently burning through $300 million per month, resulting in a burn multiple of 2.33. Here, burn multiple is calculated as total yearly cash burn divided by FY2024 revenue subtracting FY2023 revenue to get a proxy for net new ARR, a high-level indicator of capital efficiency. At 2.33, CoreWeave is spending over $2 for every net new $1 in revenue generated, which, while not catastrophic for an infrastructure-heavy hyperscaler in a land-grab phase, signals a far less efficient model than leaner cloud or AI infrastructure peers like Cloudflare or Snowflake, whose burn multiples are closer to 1.0 or below.

With the capital structure creaking and debt servicing costs soaring, going public was the only viable option. CoreWeave’s IPO wasn’t a celebration of market readiness, it was a pressure release valve. The company had to go public to avoid drowning in its own financing. With $332 million in interest expense last year, more than 18 times what it spent on sales and marketing, CoreWeave is a growth story financed with borrowed time.

The $27 billion IPO valuation makes ambitious assumptions. At $40/share, investors are effectively betting that CoreWeave can grow revenue by 30% per year and reach $15.6 billion in top-line sales by 2032, while sustaining net operating profit margins around 27% according to Forbes. That would give it profitability comparable to software giants, despite it operating massive data centers with high variable costs and limited automation. For reference, this projection exceeds the combined current revenue of Digital Realty and Equinix.

As of now, the stock has popped post-IPO and trades near $49–51 per share, buoyed by AI optimism. But beneath the surface, the fundamentals remain deeply strained. Microsoft alone accounted for 62% of CoreWeave’s 2024 revenue. The relationship is more of a hedge than a partnership. Microsoft needs GPU access now, while it finishes building its own capacity. Once Azure’s in-house silicon and AI-optimized infrastructure come online, that need and that revenue could vanish. Reports from the Financial Times and TD Cowen have already suggested that Microsoft has walked away from a gigawatt of planned CoreWeave capacity. CoreWeave denies it. The market should not.

There’s more. CoreWeave’s GAAP net income was negative $863 million last year. Its cash was worse, negative $1.4 billion. Adjusted EBITDA, a non-GAAP metric, was $1.2 billion, but that figure masks reality. Massive add-backs like stock compensation, depreciation, and interest distort true cash flow. Meanwhile, the company has $1.4 billion in cash and nearly $8 billion in traditional debt, not including a further $15 billion in long-term leases for facilities, hardware, and power, most of which are off-balance-sheet.

This is a financial structure reminiscent of WeWork: scale fast using long-term commitments, keep liabilities out of view, and hope that growth will outrun scrutiny. If demand slips or interest rates spike, CoreWeave could be left with excess capacity it can't monetize and fixed costs it can’t escape.

The company’s internal controls aren’t much better. It disclosed material weaknesses in IT systems, segregation of duties, and staffing for core accounting and finance roles. Management doesn’t expect these to be fully resolved until 2026. In other words, investors are flying blind for the next two years.

Meanwhile, the company’s founders have already taken $500 million off the table pre-IPO. They’ve retained only 30% of the equity but control 82% of the voting power through dual-class shares. They can’t be voted out. Investors have no real say. It’s a classic arrangement: privatize decision-making, socialize risk.

In theory, the upside is still massive. If CoreWeave can transition from arbitrageur to foundational infrastructure provider, it could become a critical part of the AI economy. If it locks in long-term contracts, builds diversified demand, and maintains early access to hardware, the model could stabilize. But all of those things would need to happen in short order while debt costs mount, customer concentration persists, and competition intensifies.

More realistically, CoreWeave’s path is to sell itself. An acquisition by Microsoft, Google, or Oracle would bring compute infrastructure in-house for those giants and give CoreWeave the scale and security it cannot build on its own. It would also give public investors a tidy exit. The question is whether that deal happens before the music stops and the race shop runs out of customers.