CRWV 10-K Analysis: 168% Revenue Growth, 0.70x Debt Service Coverage

CoreWeave grew revenue 168% to $5.1 billion in fiscal 2025 and sits on $60.7 billion in committed backlog — more than 11 years of revenue at the current run rate. The GPU-specialized cloud infrastructure company, purpose-built for AI workloads, now operates 43 data centers across six countries and counts Microsoft as its dominant customer at 67% of revenue.

The headline numbers look like a generational growth story. Revenue nearly tripled. Remaining performance obligations grew 302% to $60.7 billion. EBITDA reached $2.4 billion at a 47% margin. Committed contracts — multi-year, take-or-pay agreements with large AI labs — accounted for 98% of revenue, up from 88% two years ago. On-demand revenue is negligible at just 2%.

The 10-K reveals something the earnings release did not. Debt service of $4.4 billion already exceeds the $3.1 billion CoreWeave generated from operations — a 0.70x debt service coverage ratio that means the company consumed $1.3 billion more in debt payments than it earned. Interest coverage has fallen below 2.0x. And the company must raise approximately $18 to $20 billion in 2026 just to refinance maturing debt and fund the infrastructure its backlog demands. The growth is real. The question is whether CoreWeave can afford it.

The 48-Point Margin Gap — Why EBITDA Misleads Too

CoreWeave's EBITDA margin of 46.9% looks like a healthy infrastructure business. Its GAAP operating margin of -0.9% looks like a money-losing startup. Both are the same company in the same year — separated by a 48-point gap that is the single most important number for understanding CoreWeave's economics.

The gap exists because depreciation and amortization ($2,454 million) is larger than EBITDA ($2,408 million). CoreWeave's GPU fleet — $30.6 billion in net property and equipment, depreciated over a 6-year useful life — creates a D&A charge so large that it alone pushes operating income into the red. Technology and infrastructure D&A nearly tripled from $843 million to $2.3 billion as the fleet expanded. Cost of revenue also absorbed $84 million in power infrastructure depreciation, $566 million in additional rent, and $203 million in higher utilities.

"An increase in depreciation and amortization of approximately $1.5 billion, from $843 million for the year ended December 31, 2024, to $2.3 billion for the year ended December 31, 2025, resulting from investments in our platform and servers, switches, and other networking equipment."

But EBITDA is not the right alternative either. Investors who dismiss GAAP operating income and substitute EBITDA are making a different mistake: EBITDA ignores the $1,229 million in interest expense that finances the very GPU fleet creating the D&A. CoreWeave didn't choose to be capital-light — it borrowed $21.4 billion to build its infrastructure. The interest on that debt is as fundamental to the business model as rent or utilities.

The more useful economic measure sits between these extremes. EBITDA ($2,408 million) minus interest expense ($1,229 million) equals $1,179 million — what we call "true economic earnings." That is a 23.0% margin, not the 47% that EBITDA suggests or the -0.9% that GAAP shows. And it is deteriorating: in FY 2024, the same calculation yielded a 43.8% margin.

The picture worsens further when you account for hidden interest. The filing discloses $159 million in capitalized interest — added to the cost of GPU assets rather than expensed on the income statement. Including this, all-in interest coverage drops from 1.96x to 1.73x. Investment-grade companies typically maintain 4x or higher. Stock-based compensation surged from $31 million to $630 million — a 1,932% increase driven by IPO-triggered RSU vesting — adding another layer of non-cash cost that will persist at approximately $157 million per quarter.

Every major expense line outgrew revenue in FY 2025: cost of revenue increased 195%, technology and infrastructure spending rose 205%, sales and marketing surged 700%, and G&A jumped 447% — all outpacing the 168% revenue gain. CoreWeave's $2.5 billion depreciation charge — larger than its $2.4 billion EBITDA — means the company's GPU fleet costs more to write down annually than the operating earnings it generates, leaving a "true economic margin" of just 23% after interest expense.

The Depreciating Collateral Loop — CoreWeave's Novel Financial Trap

CoreWeave has built an elegant but fragile financing machine. It purchases GPU servers using delayed-draw term loans (DDTLs) that are collateralized by the very hardware they finance. New customer contracts provide the revenue visibility that lenders need to extend credit. The GPUs fulfill the contracts. Revenue pays down the debt. On paper, it is a self-reinforcing cycle.

The 10-K reveals the trap embedded in this model. The filing states that DDTL borrowing capacity is "limited to a percentage of the depreciated purchase price of GPU servers and related infrastructure." GPU servers are depreciated over a 6-year useful life — approximately 17% per year on a straight-line basis. But the weighted average maturity of CoreWeave's $21.4 billion in debt is approximately 3.3 years.

"The total loans available are limited to a percentage of the depreciated purchase price of GPU servers and related infrastructure for the contract that the loans are being used to finance."

At the average debt maturity of 3.3 years, a GPU server has depreciated to roughly 45% of its original value. The collateral backing the loan has lost more than half its book value. To refinance the maturing debt, CoreWeave faces three options: accept worse terms against diminished collateral, inject equity to de-lever, or — and this is the reflexive loop — purchase new GPU servers to pledge as fresh collateral. Option three is what the company has been doing, and it requires new debt to fund the new GPU purchases, which will themselves depreciate and need refinancing in another 3.3 years.

The cost dispersion across instruments tells its own story. The earliest DDTLs carry effective rates of 15% — private credit territory, not investment-grade borrowing. Management describes having "pioneered and scaled innovative financing structures," but the cost of that innovation is embedded in rates that far exceed what hyperscalers pay. The most recent instruments show improvement (the 2031 Convertibles at 1.75% stated rate), but the legacy debt remains expensive.

Total capital obligations — debt plus $8.2 billion in operating lease liabilities — reach $29.8 billion, or 5.8 times annual revenue. Debt service of $4.4 billion already exceeds operating cash flow of $3.1 billion, producing a debt service coverage ratio of just 0.70x. CoreWeave consumed $1.3 billion more in debt payments than it generated from operations, even before a single dollar of growth capex. CoreWeave's delayed-draw loans are collateralized by GPU servers that lose approximately 17% of their value each year, creating a refinancing trap where the average debt matures in 3.3 years but the underlying hardware has only depreciated to 45% of its original collateral value.

The $154 Million Buffer — Can CoreWeave Fund 2026?

CoreWeave's total liquidity as of year-end was $6.9 billion. Its debt maturing in 2026 totals $6.7 billion. The buffer between having enough cash to refinance and not having enough is $154 million — before any growth spending.

Management has guided capital expenditures "increasing relative to 2025," when the company spent $10.3 billion. At a modest 10% increase, 2026 capex would be $11.3 billion; at 30%, it reaches $13.4 billion. Add the $6.7 billion in maturing debt, and CoreWeave's total 2026 funding need ranges from $18.0 billion to $20.1 billion — approximately 3.5 times its annual revenue.

"For the year ended December 31, 2025, our cash flows dedicated for debt service requirements totaled $4.4 billion, which includes principal payments of $3.4 billion and interest payments of $1.0 billion, inclusive of $159 million related to capitalized interest."

Where does $18-20 billion come from? Operating cash flow might contribute $3-4 billion, assuming continued contract revenue and working capital inflows from upfront customer payments. That leaves a funding gap of roughly $8-10 billion or more that must come from new debt issuance, equity offerings, or some combination. For context, CoreWeave added $7.4 billion in new debt in Q4 2025 alone — an astonishing pace of issuance that coincided with the stock's decline from approximately $90 to $64.

There is a mitigating factor. In November 2025, CoreWeave extended its revolving credit facility to November 2029 and tripled its capacity to $2.5 billion. The fact that banks tripled the facility signals that institutional lenders remain confident in CoreWeave's collateral and contract base. But a revolving credit facility is a bridge, not a destination — it provides liquidity flexibility without solving the structural funding gap.

The scale of capital required puts CoreWeave in singular territory among technology companies. Dell Technologies carries similar absolute debt ($23.8 billion) but on $104 billion in revenue — 20 times CoreWeave's top line — with a manageable 2.0x Net Debt/EBITDA. Snowflake, DoorDash, and MercadoLibre all carry net cash positions. CoreWeave's $6.9 billion in total liquidity would cover less than one year of debt maturities — leaving a $154 million buffer and zero dollars for the $11 billion-plus in capital spending that its $60.7 billion backlog requires.

Two Customers, $72 Billion in Commitments — The Concentration Paradox

Microsoft accounted for 67% of CoreWeave's revenue in FY 2025. That number alone would concern most investors. The trend is worse.

In FY 2023, Microsoft was 35% of revenue, Customer B was 15%, and Customer C was 21%. By FY 2024, Microsoft had risen to 62% while both Customer B and Customer C fell below the 10% disclosure threshold. In FY 2025, Microsoft reached 67%. The "top two customers" concentration that was once 77% across two distinct entities has collapsed into essentially one customer at 67%. CoreWeave's customer diversification is going backward.

"We recognized an aggregate of approximately 67% of our revenue from our top customer, Microsoft, for the year ended December 31, 2025. We recognized an aggregate of approximately 77% of our revenue from our top two customers for the year ended December 31, 2024."

There is a paradox here. Microsoft outsourcing GPU capacity to CoreWeave is the strongest possible validation of the business model — even the world's second-largest cloud provider, which builds its own data centers at massive scale, needs external GPU infrastructure. OpenAI's $11.9 billion commitment through October 2030 adds a second anchor, representing approximately 20% of total RPO. Together, these two relationships likely account for the majority of the $60.7 billion backlog.

"In March 2025, we entered into a master services agreement with OpenAI, a private company, pursuant to which OpenAI has committed to pay us up to approximately $11.9 billion through October 2030. Other significant customers include Microsoft and Meta."

But validation and fragility are two sides of the same coin. If Microsoft were to slow its AI capex — due to budget reallocation, in-house capacity coming online, or a strategic pivot — there is no second customer large enough to absorb the impact. CoreWeave's revenue, its ability to service $21.4 billion in debt, and its collateral base for refinancing all depend on decisions made in Redmond. And OpenAI, the second-largest disclosed commitment, is a private company that management explicitly flags for credit risk.

The commitment data reinforces the lock-in. Committed contracts rose from 88% of revenue in FY 2023 to 98% in FY 2025 — contractual protection against customer churn, but also confirmation that there is virtually no spot or on-demand revenue to attract new customers. Revenue growth from existing customers was 85% in FY 2025, with just 15% from new customers. Sequential quarterly revenue growth has decelerated from 31% to 15%, suggesting the existing contract base is being consumed while new customer acquisition has not yet reached meaningful scale. Microsoft's share of CoreWeave revenue has risen from 35% to 67% in two years — not because Microsoft is growing faster, but because every other large customer has fallen below the 10% disclosure threshold.

Among this peer set, CoreWeave is the only company with negative free cash flow, the only one with capex exceeding its revenue, and one of only two with any net debt at all. MercadoLibre grew revenue 39% at $29 billion scale while generating 37% FCF margins and 25% ROIC — what growth-with-operating-leverage looks like. CoreWeave's 168% growth at -141% FCF margins is the opposite: growth funded entirely by external capital, not internal earnings.

What to Watch

Five metrics will determine whether CoreWeave's growth story survives its capital structure — or whether the financing trap tightens:

At $71.61 per share, CoreWeave trades at 9.4x trailing EV/Revenue, implying the company reaches $10-12 billion in revenue within three years at infrastructure-company multiples of 4-5x. The $60.7 billion RPO backlog says the revenue path exists — 11.8 years of committed contracts at the current run rate. The Q4 annualized run rate of $6.3 billion already implies 23% revenue growth locked in for FY 2026. Contracted power of 3.1 GW is 3.6x active capacity of 850 MW, providing the physical footprint for years of expansion.

But the 10-K complicates the path from here to there. Debt service already exceeds operating cash flow. Interest coverage is below 2.0x. The $18-20 billion 2026 funding need requires perpetual capital market access. Each $1 of RPO conversion demands approximately $1.50-2.00 in capex — meaning the $60.7 billion backlog requires an estimated $91-121 billion in cumulative capital spending over its life, funded almost entirely by debt rather than retained earnings. The filing supports the growth trajectory. It reveals the cost of funding it.

The thesis breaks if CoreWeave demonstrates debt service coverage above 1.0x for two consecutive quarters while maintaining capex guidance — because that would prove the business model is self-funding and the reflexive debt loop is unwinding rather than tightening.

Frequently Asked Questions

Why did CoreWeave's operating margin swing from +16.9% to -0.9% despite 168% revenue growth?

Every major expense line grew faster than revenue in FY 2025. Cost of revenue increased 195%, technology and infrastructure spending rose 205%, sales and marketing surged 700%, and G&A jumped 447% — all outpacing the 168% revenue gain. The single largest factor was depreciation and amortization, which grew from $843 million to $2.3 billion in technology costs alone as CoreWeave's GPU fleet expanded to $30.6 billion in net property and equipment. Stock-based compensation also surged from $31 million to $630 million, primarily from IPO-triggered RSU vesting.

Is CoreWeave's $60.7 billion remaining performance obligation (RPO) backlog reliable?

The RPO represents contractually committed revenue recognized ratably over 1-6 year contract periods. Committed contracts accounted for 98% of FY 2025 revenue, up from 88% in 2023. However, two concentration risks exist: Microsoft accounts for 67% of current revenue, and the largest disclosed contract — OpenAI's $11.9 billion through October 2030 — represents approximately 20% of total RPO. Management explicitly identifies increased credit risk from customers in their early stages or private companies. The backlog is contractually binding, but conversion to revenue depends on CoreWeave having the infrastructure capacity, which requires $10B+ annual capex.

How does CoreWeave's debt compare to other technology companies?

CoreWeave's total debt of $21.4 billion on $5.1 billion revenue produces a Debt/Revenue ratio of 4.2x and Net Debt/EBITDA of 7.2x. Dell Technologies carries similar absolute debt ($23.8B) but on $104 billion revenue (2.0x Net Debt/EBITDA). Snowflake and DoorDash carry zero debt with net cash positions. CoreWeave's leverage is closer to a leveraged buyout than a technology company, driven by its project-finance-style GPU acquisition model where delayed-draw term loans are collateralized by the hardware they finance.

What is CoreWeave's interest coverage ratio and why does it matter?

CoreWeave's EBITDA-based interest coverage is 1.96x using the filing's $1,229 million interest expense, net — below the 2.0x threshold that typically signals credit stress. Including $159 million in capitalized interest (added to asset cost rather than expensed), all-in coverage drops to 1.73x. Investment-grade companies typically maintain 4x+ coverage. CoreWeave's level means a single weak quarter could push coverage below covenant thresholds.

Can CoreWeave become free cash flow positive?

In FY 2025, CoreWeave generated $3.1 billion in operating cash flow but spent $10.3 billion in capital expenditures, producing negative free cash flow of -$7.3 billion (-141% FCF margin). Management has guided 2026 capex higher than 2025. For FCF to reach breakeven, revenue must grow enough that OCF exceeds capex, requiring approximately $22 billion in revenue at current margins and $10B+ capex. Given the RPO backlog requires years of infrastructure buildout, FCF breakeven is unlikely before 2028-2029 at the earliest.

Why is Microsoft's revenue concentration increasing rather than diversifying?

Microsoft's share rose from 35% in 2023 to 62% in 2024 to 67% in 2025. Simultaneously, Customer B dropped from 15% to below the 10% reporting threshold, and Customer C dropped from 21% to below 10%. The acceleration suggests Microsoft is expanding GPU usage faster than other customers, while new customer acquisition at meaningful scale has not materialized. CoreWeave's 85% revenue growth from existing customers (15% from new) supports this interpretation.

What are the key risks in CoreWeave's debt structure?

Three structural risks emerge from the filing: (1) Maturity concentration — $6.7 billion matures in 2026 with total liquidity of $6.9 billion, a $154 million buffer before growth spending. (2) Collateral depreciation — DDTL loans are collateralized by GPU servers with a 6-year useful life, meaning collateral declines ~17% annually; at the 3.3-year average debt maturity, only ~45% of collateral value remains. (3) Cost dispersion — effective interest rates range from 1.75% to 15%, with legacy debt at private-credit rates. Debt service of $4.4 billion already exceeds operating cash flow of $3.1 billion.

How does CoreWeave's capital intensity compare to hyperscalers?

CoreWeave spent $10.3 billion in capex on $5.1 billion revenue — a 201% capex/revenue ratio. Peers average 2.8%: MELI at 4.6%, DELL at 2.5%, SNOW at 2.2%, DASH at 1.9%. Even hyperscalers with massive AI programs don't approach this ratio: Meta's capex/revenue is approximately 25-30%, Amazon's approximately 30-35%. CoreWeave's ratio is 6-8x higher because it builds dedicated GPU infrastructure from scratch rather than adding AI capacity to an existing cloud platform.

What is the reflexive debt-collateral loop and why is it unique to CoreWeave?

CoreWeave finances GPU purchases through delayed-draw term loans collateralized by the hardware itself. This creates a feedback loop: new contracts provide collateral for new loans, loans fund GPU purchases, GPUs fulfill contracts but depreciate over 6 years, at debt maturity (~3.3 years average) only ~45% of collateral value remains, and refinancing requires either new GPU purchases (more debt) or equity injection. Hyperscalers fund GPU purchases from operating cash flow and don't pledge hardware as collateral. Management itself describes having "pioneered and scaled innovative financing structures."

What would cause CoreWeave's stock to re-rate higher or lower?

Higher: (1) Debt service coverage exceeding 1.0x for two consecutive quarters, proving self-funding capability. (2) Microsoft concentration dropping below 60%. (3) Investment-grade credit rating, lowering refinancing costs. (4) Q1 2026 revenue above $1.9B, indicating faster capacity ramp. Lower: (1) Failure to refinance $6.7B 2026 maturities at reasonable rates. (2) Microsoft reducing GPU commitments. (3) Interest expense exceeding $1.9B (coverage below 1.5x). (4) OpenAI restructuring its $11.9B commitment.

What is CoreWeave's "true economic margin" and why does it differ from EBITDA margin?

CoreWeave's EBITDA margin of 46.9% excludes both depreciation ($2.5 billion) and interest expense ($1.2 billion). But the interest expense directly funds the GPU assets that generate the depreciation — they are economically linked. "True economic earnings" (EBITDA minus interest) equals $1,179 million, or a 23.0% margin. This is the cash earnings layer between EBITDA (which overstates by ignoring financing costs) and operating income (which understates by charging full depreciation on assets with remaining economic life). The gap between EBITDA margin (46.9%) and true economic margin (23.0%) represents the cost of funding CoreWeave's asset-heavy model.

Methodology

Data Sources

This analysis is based on CoreWeave's FY2025 Annual Report (10-K) filed with the SEC on March 2, 2026, accessed via the MetricDuck filing text extraction pipeline. Financial metrics were extracted using the MetricDuck XBRL metrics processor and cross-verified against filing text in the following sections: MD&A — Results of Operations, MD&A — Liquidity, footnotes on debt and commitments, risk factors, and accounting policies.

Peer data for MELI, SNOW, DELL, and DASH is sourced from MetricDuck pipeline data. All peer metrics are from the latest available fiscal year.

Key Derived Calculations

• DSCR: OCF ($3,058M) / Total Debt Service ($4,400M) = 0.70x
• Interest Coverage (filing): EBITDA ($2,408M) / Interest Expense, Net ($1,229M) = 1.96x
• All-in Interest Coverage: EBITDA ($2,408M) / (P&L Interest $1,229M + Capitalized Interest $159M) = 1.73x
• "True Economic Earnings": EBITDA ($2,408M) - Interest ($1,229M) = $1,179M (23.0% margin)
• 2026 Funding Need: $6.7B maturities + $11.3-13.4B capex (+10% to +30% of FY 2025) = $18.0-20.1B
• Weighted Average Debt Maturity: ~3.3 years (from filing maturity schedule, weighted by principal)
• GPU Collateral at Avg Maturity: (1 - 3.3/6.0) × 100% ≈ 45% remaining

Limitations

1. RPO duration not granularly disclosed. The filing does not provide an aging schedule for the $60.7B RPO. OpenAI's $11.9B through October 2030 is the only contract with disclosed duration.
2. No maintenance vs. growth capex separation. CoreWeave does not distinguish replacement capex from expansion capex. The "collateral maintenance" component is estimated, not disclosed.
3. Microsoft contract terms undisclosed. Termination penalties, renegotiation provisions, and exact duration are not in the 10-K. The take-or-pay characterization is inferred from the 98% committed contract disclosure.
4. Capex growth assumption is a range, not guidance. The filing says "increase relative to 2025" without quantification. The $11.3-13.4B range (+10% to +30%) is the author's estimate.
5. Peer comparison limitations. CoreWeave's GPU infrastructure-as-a-service model has no direct public comparable. MELI, SNOW, DELL, and DASH were selected for specific comparison angles rather than business model similarity.
6. Pre-IPO financials limited. FY 2024 balance sheet data relies on pipeline estimates because CoreWeave was private until March 2025.

Disclaimer: This analysis is for informational purposes only and does not constitute investment advice. The author does not hold positions in CRWV, MELI, SNOW, DELL, or DASH. Past performance and current metrics do not guarantee future results. All data is derived from public SEC filings and may contain errors or omissions from the automated extraction process.