Company Deep Dive #1 - COREWEAVE
Dissecting the mechanics of selling AI cloud infrastructure
In today’s edition, we’ll dive deep into the business of CoreWeave and understand how its movements will impact infra startups.
1. CoreWeave at a Glance - Why People Are Talking About It
CoreWeave (ticker CRWV) rents out banks of high-end Nvidia GPUs that companies need to train and run modern AI models. In 2024, the company’s sales jumped from $229 million to $1.9 billion. This is a 730% surge that put it on every growth investor’s radar.
A big reason for that leap is a series of long-term “take-or-pay” contracts. Customers reserve capacity and pay even if they end up using less. Those signed orders (called backlog) stood at $25.9 billion on Mar 31 2025, more than 5x last year’s revenue as reported here.
The concentration of business is stark. Microsoft alone provided 62% of 2024 revenue. And the top two customers made up 77%. OpenAI has since committed another $11.9 billion contract and taken a $350 million equity stake, further tying CoreWeave’s fate to a handful of tech giants.
Growth is being financed with heavy borrowing. Industry press reports show CoreWeave will need to issue another $8–11 billion of debt in 2025 on top of the $2.6 billion that matures the same year. The first-quarter 2025 results underline the weight of interest costs. Revenue was $982 million, but net loss reached $315 million largely because interest expense rose 6x year on year.
They’re scaling up fast. CoreWeave now runs about 250,000 GPUs across 33 data-centers in the US and Europe. Roughly 4x the capacity of the largest pure-startup rival.
2. The AI-Computing Boom in Everyday Terms
Training an LLM is like running a blast furnace. The job demands enormous, continuous power for a short time. A cutting-edge chatbot may need thousands of GPUs working together for weeks. Public clouds such as AWS or Microsoft Azure offer GPUs. But customers complain about shortages, high prices, and slow setup.
That gap has created a specialized AI-infrastructure market that analysts at IDC expect to grow from roughly $79 billion today to more than $200 billion by 2028. Some forecasts stretch toward $400 billion by 2027. Even if only the conservative half of that range materializes, the pie is still several times larger than it is today.
Because demand is exploding faster than traditional clouds can add supply, second-tier providers like CoreWeave have room to carve out a niche by promising faster availability or lower cost. The trick is staying ahead while the giants race to catch up. Amazon, Google, and Microsoft are all building their own in-house AI chips and adding tens of thousands of Nvidia GPUs at the same time.
3. How CoreWeave Makes Money and Why It Grew So Fast
CoreWeave’s business model is simple to state. Buy the newest Nvidia chips, wire them into giant clusters, rent them out by the hour, and let software handle every step from scheduling jobs to billing. Two details explain the company’s rapid rise:
3.1 Priority Access to Hardware
Nvidia holds more than 90% share of the AI-GPU market. Because Nvidia also owns a 7% stake in CoreWeave and touts the firm as a reference customer, CoreWeave often receives chips earlier than smaller rivals. Early delivery means CoreWeave can promise capacity when others cannot, winning large contracts and collecting deposits months in advance.
3.2 Take-or-Pay Contracts
Roughly 96% of revenue comes from multi-year deals that bill customers whether they use the hours or not. That structure converts hardware scarcity into predictable cash flow, at least while demand exceeds supply.
With utilization high, gross margin (the revenue left after paying direct costs such as power and data-center rent) was about 49% in 2024 and edged above 50% in early 2025. Stripped of interest and depreciation, adjusted EBITDA margin hit 62% last quarter. Those figures rival the software industry and are striking for what is in essence a hardware-driven business.
4. The Hidden Snags: Debt, Dependence, and Dilution
4.1 A Mountain of Borrowing
CoreWeave’s expansion is fueled by debt. The company has already borrowed several billion dollars and will have to raise another $8–11 billion in 2025 to fund an ambitious $20 billion capital-spending plan. Interest expense alone was $264 million in the most recent quarter. If credit markets stay calm, rolling over loans may be routine. If not, cash could get tight fast.
4.2 One Whale Client
Microsoft’s sheer share of revenue is a double-edged sword. 62% in 2024 and 72% in the first quarter of 2025. The software giant is credit-worthy and keen on more AI capacity, but it also wields negotiating power. Even a modest 10% price cut on Microsoft’s contract would shave roughly $120 million off annual revenue.
4.3 Dilution and Lock-Up Expiry
CoreWeave raised $1.4 billion in its March 2025 IPO, issuing new shares to do so. A further share issue is part of its acquisition of data-center operator Core Scientific. Early investors are restricted from selling until mid-September 2025. Heavy sales then could weigh on the price.
5. A Look at the Competition - Startups and Giants Side by Side
Public-Cloud Titans: Amazon, Microsoft, and Google collectively operate millions of GPUs and spend tens of billions of dollars each year on data centers. Their deep pockets allow them to match or undercut CoreWeave’s pricing once they have stock.
Specialist Startups: A host of younger companies target similar customers:
CoreWeave has 250,000 GPUs.
Lambda Labs has 60,000 GPUs.
Nebius has 30,000 GPUs. Plans for 60,000 more.
Voltage Park has 24,000 GPUs.
These numbers illustrate just how dominant CoreWeave is within the specialist camp. It controls roughly 55% of the combined GPU stock held by the five best-known AI-cloud startups. (250k out of ~454k total.)
6. How CoreWeave’s Moves Ripple Through Infrastructure Startups
6.1 Chip Supply and Pricing
Nvidia’s manufacturing slots are limited. Because CoreWeave receives large early allocations thanks to its size and Nvidia’s equity stake, smaller startups often receive hardware weeks or months later and pay higher prices. That trickles down into the rates they can offer customers. When CoreWeave sells hours at aggressive discounts to fill new capacity, rivals must decide whether to cut prices (hurting margins) or wait for tighter capacity later.
6.2 Investor Sentiment and Valuations
Many venture investors use CoreWeave’s valuation as a benchmark. With CoreWeave trading around 11x expected 2025 revenue, startups pitching similar stories have raised rounds at price-to-sales ratios between 8 and 12. A sharp shift in CoreWeave’s multiple (e.g. after a debt hiccup or price war) would likely reset expectations for funding across the sector.
6.3 Partnership Opportunities
Some founders see CoreWeave less as a rival and more as a necessary supplier. Voltage Park, Nebius, and certain enterprise software vendors have signed wholesale deals to resell or burst into CoreWeave capacity when their own clusters fill up. If CoreWeave tightens those partner allotments to protect big anchor clients, downstream providers could face shortages. Conversely, adding more capacity quickly can relieve shortages for everyone.
6.4 Risk Concentration Across the Ecosystem
Roughly one-third of venture-backed infra startups tap into GPU leasing, hosting, or management. A prolonged dip in CoreWeave’s financial health would spook suppliers, banks, and customers. And this would raise borrowing costs or insurance premiums for lookalikes. The dependency is not just psychological. Power-grid upgrades, fibre routes, and chip-packaging plants financed on the assumption of CoreWeave-level demand could get delayed, squeezing the whole supply chain.
6.5 Correlation in Billing Models
Because CoreWeave popularized take-or-pay contracts, many startups copy the idea to secure financing. If large buyers push back (perhaps after noticing they are paying for unused hours), CoreWeave would be the first to renegotiate. But the precedent would carry over to smaller providers. Revenue recognition across the infra startup cohort would then sag in unison.
7. What to Watch Over the Next 12–24 Months
August 12 earnings call. Investors will scrutinize backlog growth, Microsoft’s share of sales, and whether gross margin remains north of 50%.
Debt refinancings. The terms on the upcoming $8–11 billion in new bonds will signal how comfortable lenders feel about the story.
Customer diversification. A major contract in banking, healthcare, or government would cut the Microsoft dependency and reassure the market.
Nvidia pipeline. Any hint that Nvidia favors a different partner or that hyperscalers no longer share GPU shortages could compress CoreWeave’s edge.
GPU prices in the spot market. When on-demand rates fall below CoreWeave’s contract levels, customers may delay signing new deals, slowing backlog growth.
Startups should track the same indicators. Lower utilization or price cuts at CoreWeave often lead the market, showing up in independent benchmark sites and trickling to competitors within weeks.
8. Takeaways for Investors
CoreWeave sits at the heart of a rapidly expanding AI-computing industry. Its early access to hardware and willingness to spend big have yielded eye-popping growth (along with amazing gross margins). At the same time, the company leans heavily on debt and a single giant customer, which leaves little margin for error.
Why it matters beyond one stock: CoreWeave’s decisions ripple through a whole class of infrastructure startups. Roughly one in three funding dollars in the sector aims at GPU hosting, AI-oriented data centers, or related tooling. Whether you own CoreWeave shares or venture stakes in Voltage Park, Lambda, or a smaller seed-stage platform, the same macro levers apply: chip supply, power costs, and buyer appetite for multi-year commitments.
Bottom line: CoreWeave provides a useful barometer for the health of the specialized AI infrastructure niche. Keep an eye on utilization, debt terms, and the Microsoft share of revenue. If those metrics hold steady while new industries sign on, CoreWeave and its startup cousins may keep compounding. If they wobble, expect a downturn to hit the wider infra startup space in short order.
If you are getting value from this newsletter, consider subscribing for free and sharing it with 1 infra-curious friend: