Were We Wrong About CoreWeave?

A Post-IPO Reassessment of the Most Polarizing AI Infrastructure Play

Three months ago, when CoreWeave went public at $40 per share, the response from financial analysts, institutional investors, and media outlets was largely cautious – if not outright critical. At the time, the company looked like a high-risk bet on the AI boom, closer to a highly leveraged GPU middleman than a lasting infrastructure player. Our article covering CoreWeave’s IPO in April echoed a lot of those same feelings.

Frankly, it was hard not to be bearish.

The numbers painted a picture of fragility. The business was hemorrhaging cash, heavily indebted, and reliant on one outsized customer, Microsoft, for a majority of its revenue. It had material weaknesses in internal controls, a governance structure that concentrated power in the hands of its founders, and what appeared to be a very narrow window to transition from opportunistic chip reseller to bona fide AI utility.

But markets have a way of surprising you. Since that IPO, CoreWeave's share price has soared past $140, rising nearly 250%. It has closed a $2 billion unsecured debt refinancing led by top-tier banks. And perhaps most intriguingly, they were able to realize synergies from their acquisition of Weights & Biases (W&B) faster than expected. This has allowed CoreWeave to develop a full AI suite, a move that is already paying dividends.

With so many moving parts, and so much investor enthusiasm, it’s time to ask the question seriously: Were we wrong about CoreWeave?

Revisiting the Original Thesis: A House of Cards?

Our skepticism wasn’t without reason. Let’s recap the core pillars of the original bear thesis:

1. Chip Arbitrage Doesn’t Scale: CoreWeave’s original model was to buy Nvidia’s cutting-edge GPUs, deploy them in custom-built data centers, and rent them out at a premium. But arbitrage doesn’t scale forever, especially when your suppliers (Nvidia) and customers (Microsoft, OpenAI) have the means and incentives to cut you out of the loop.

2. Customer Concentration Risk: Microsoft alone contributed 62% of CoreWeave’s revenue in 2024. That level of exposure to one client, especially one building its own infrastructure, was a glaring red flag.

3. Structural Fragility: With nearly $8 billion in debt and another $15 billion in long-term off-balance sheet lease obligations, the capital structure looked eerily reminiscent of WeWork. A burn rate of $300 million per month only added fuel to the fire.

4. Governance Concerns: The founders held 82% of the voting rights despite owning just 30% of the equity. Add in disclosed material weaknesses in finance and IT systems, and you had a governance setup that was hardly shareholder-friendly.

5. Lofty Valuation: At IPO, CoreWeave was valued at $27 billion, more than the combined revenues of Digital Realty and Equinix, with no profit, little control over supply, and limited pricing power.

It’s not that these points were wrong. They were, and in many ways, still are, valid. But the pace and nature of CoreWeave’s strategic pivots post-IPO are shifting the narrative in real time.

The $2 Billion Refinancing: A Turning Point for Market Perception

In late May, CoreWeave closed a $2 billion unsecured bond offering with a 9.25% coupon and a five-year term. On the surface, that’s expensive money. But dig deeper, and this deal is a telling milestone.

First, it was oversubscribed and syndicated across elite financial institutions – JP Morgan, Morgan Stanley, Deutsche Bank, and others. These aren’t speculative venture lenders; they’re the heartbeat of institutional capital markets. For them to underwrite CoreWeave’s unsecured debt signals growing confidence in the company’s cash flow durability and long-term relevance.

Second, the bonds were unsecured and covenant-light. That’s critical. Earlier tranches of CoreWeave debt were backed by GPU hardware, a clear sign of lender risk aversion. The new structure provides the company with far more operational flexibility, freeing it from asset-level constraints.

Yes, the interest rate is high. But when weighed against the alternatives, such as further equity dilution or forced asset sales, this is arguably the most favourable capital move CoreWeave could have made in its current position.

The signal to markets? CoreWeave is becoming bankable.

Weights & Biases: The Real Strategy Behind the Hype

While the acquisition of Weights & Biases occurred in June 2024, before CoreWeave’s IPO, its strategic implications are now coming to fruition, and they may prove more consequential than initially understood. 

In our previous coverage, we skipped out on the W&B deal, thinking it was merely an MLOps play. We discounted how quickly CoreWeave has been able to integrate and extract value from the acquisition. Here’s why:

1. Vertical Integration: By bringing W&B in-house, CoreWeave gains over 700,000 users and a deep bench of AI-native enterprise customers. This isn’t just tooling – but control. It now owns a critical piece of the machine learning lifecycle. That means more lock-in, greater usage data, and the ability to cross-sell compute and software.

2. Flywheel Creation: With CoreWeave GPUs and W&B tooling deeply integrated, the platform becomes a one-stop shop for training and deploying models. Startups and AI labs don’t have to stitch together their infra stack – they can just plug into CoreWeave.

3. Defensibility: W&B is sticky. ML teams don’t easily switch off the platform they’ve built their workflows on. That stickiness could protect CoreWeave’s margins as cloud compute pricing eventually normalizes.

This faster-than-expected realization of value reframes CoreWeave not as a commodity GPU lessor but as a full-stack AI infrastructure company. It’s no longer just selling the race car – it owns the track, the telemetry software, and increasingly, the team garages too.

Financials: Not Ideal, But Less Alarming

Let’s be clear: CoreWeave is still not profitable. It’s posted a trailing net income margin of -38.7% and EPS of -$5.19. Cash flow from operations has slowed since the IPO, and despite over $1.2 billion in cash, the company’s debt and lease obligations remain daunting.

That said, top-line performance has been phenomenal:

• TTM Revenue: $2.7 billion (up from $229M in 2023)

• Gross Margin: 74%+

• EBITDA Margin: 55.7%

• 2026E Revenue: $11.6 billion

These aren’t the numbers of a struggling business, they’re the numbers of a hyperscaler in the making. When paired with remaining performance obligations (RPOs) north of $15 billion, there’s a case to be made that CoreWeave has growing visibility into recurring revenue, even as it battles capital costs.

Valuation? Still hot. CoreWeave trades at 28x trailing revenue and over 20x forward EBITDA. But in a world where Nvidia is pushing a $3 trillion market cap and Databricks is raising at $43 billion pre-IPO, relative multiples aren’t outlandish.

Microsoft: Still a Risk, But Less Existential

The original risk was simple: if Microsoft builds its own infrastructure and exits the CoreWeave relationship, the company could take a significant hit.

What’s changed? Two things:

1. Diversification is happening. RPOs suggest new enterprise clients are coming online. OpenAI, Anthropic, and high-intensity labs are increasingly spreading compute loads across multiple providers.

2. Microsoft itself has changed its posture. While there were reports earlier this year that Microsoft had pulled back from a planned gigawatt of CoreWeave capacity, the narrative has since softened. CoreWeave’s integration into Azure tools, and the W&B acquisition, may strengthen rather than erode the relationship.

This doesn’t eliminate the concentration risk. But it repositions Microsoft from a single point of failure to a strategic partner with options.

Governance and Control: The Founder’s Dilemma

Founders still control 82% of voting rights. Material weaknesses in accounting remain unresolved. For institutional investors who value transparency, this is a persistent drag.

But one thing is increasingly clear: public markets are willing to tolerate governance risk when growth is exponential and positioning is unique. Think Tesla in 2019, or Meta in its early days. If CoreWeave continues to dominate the AI training landscape, investors may accept weak governance as the cost of admission.

Still, this remains one of the few bearish points that hasn’t materially improved. It’s a watch item, especially as the company scales.

Final Thoughts: We Underestimated the Pivot

Were we wrong about CoreWeave?

The cautious answer is we were early. Our initial critique focused on real risks, debt, concentration, governance. Those risks persist. But what we failed to fully appreciate was the company’s ability to evolve rapidly in response to market conditions.

• The debt overhang? Refi’d.

• The single-product limitation? Now acquiring across the AI stack.

• The narrative of fragility? Being replaced by one of scale and defensibility.

CoreWeave today is not the same company that IPO’d in March. It’s a company with strategic optionality, expanding moats, and growing investor confidence.

The lesson? In high-growth tech, especially AI, narratives can shift faster than traditional models can update.

Disclosure: This article does not constitute investment advice. Always conduct your own due diligence.