CoreWeave, Inc. -- How the Business Works
CoreWeave is a GPU cloud infrastructure company that borrows capital to buy NVIDIA GPUs
and leases compute capacity to AI companies under multi-year take-or-pay contracts. Founded
as a crypto mining operation, the company pivoted to AI cloud in 2022 and IPO-ed in March
2025. FY2025 revenue of $5.1B (+168% YoY) with a $66.8B contracted backlog (13x trailing
revenue). Adjusted EBITDA margin reached 57% in Q4, but GAAP net loss was -$1.2B due to
$1.22B in annual interest expense and massive depreciation on GPU infrastructure. CapEx of
$14.9B in FY2025, guided to $30-35B in FY2026. Debt/equity of 8.94x with total debt
exceeding $14.6B. No free cash flow, no competitive moat, less than 13 months as a public
company. All three quality gates FAIL. Composite score 4.0/10 (NOT ACTIONABLE / Speculative).
Price / Composite Score
$82.24 / 4.0
NOT ACTIONABLE -- all quality gates fail
FY2025 Revenue
$5.1B
+168% YoY | $66.8B contracted backlog
Debt / Equity
8.94x
Total debt >$14.6B | Interest $1.22B/yr
FY2026 CapEx Guide
$30-35B
More than 2x guided revenue of $12-13B
Business model -- leveraged GPU infrastructure intermediary
Value Chain -- How CoreWeave Makes Money
Borrow Capital
Debt/equity raises, SPV financing
D/E 8.94x | >$14.6B total debt
→
Buy NVIDIA GPUs
Blackwell, GB200, Rubin (2026)
Single-supplier dependency
→
Build GPU Cloud
Data centers, networking, software
NVIDIA Exemplar Cloud status
→
Lease to AI Companies
Multi-year take-or-pay contracts
$66.8B backlog | ~70% target EBITDA margin
Revenue Trajectory -- Quarterly Progression
Q1 2025
$982M
+420% YoY
Q2 2025
$1.2B
+207% YoY
Q3 2025
$1.4B
+134% YoY
Q4 2025
$1.6B
+110% YoY | 57% Adj. EBITDA margin
FY2026 Guide
$12-13B
~140% YoY growth guided
Key Customers and Contracted Demand
$66.8B
Contracted Backlog
13x trailing revenue | 5-year take-or-pay
$11.9B
OpenAI Contract
Largest single contract disclosed
MSFT / META
Major Hyperscaler Customers
Plus CrowdStrike, Rakuten, MercadoLibre
~70%
Target EBITDA Margin
On seasoned contracts at scale
CoreWeave is a leveraged infrastructure intermediary, not a technology company.
The business model is structurally simple: borrow money, buy NVIDIA GPUs, lease them to
AI companies under multi-year contracts, and earn the spread between the cost of capital
and the lease rate. Revenue is growing at extraordinary rates ($5.1B in FY2025, guided
$12-13B in FY2026), and the $66.8B backlog provides visibility. But the model requires
continuous access to capital markets -- FY2026 capex of $30-35B dwarfs revenue. Interest
expense alone consumed 24% of FY2025 revenue. GAAP net loss of -$1.2B. The entire value
proposition depends on two things: (1) NVIDIA hardware remaining the industry standard,
and (2) sustained demand from a concentrated customer base dominated by hyperscalers.
If either breaks down, the leveraged model amplifies the downside.
Revenue and contract data from CRWV S-1, 10-K, 10-Q filings, and quarterly earnings releases. Backlog and capex guidance from management commentary.
NVIDIA dependency -- single-supplier risk at the core of the model
NVIDIA Relationship -- Supplier, Investor, and Kingmaker
Sole GPU Supplier
Entire infrastructure built on NVIDIA
Blackwell, GB200, Rubin (2026), Vera
$2B Investor
NVIDIA equity investment (Jan 2026)
Strategic alignment but also dependency
Exemplar Cloud
First cloud with GB200 Exemplar status
SemiAnalysis platinum ranking
Existential Risk
If NVIDIA shifts priorities or goes direct
No alternative supplier at scale
NVIDIA is supplier, investor, and kingmaker -- all in one.
CoreWeave does not own the core technology. NVIDIA does. The entire business depends on
preferred allocation of NVIDIA hardware, which NVIDIA can redirect to other partners or
sell directly. Custom silicon from Google (TPUs), Amazon (Trainium), and Microsoft (Maia)
represents a structural threat -- if hyperscalers increasingly use their own chips, demand
for third-party GPU cloud diminishes. Hardware transitions every 12-18 months (Blackwell
to Rubin) create continuous execution risk. The $2B NVIDIA investment aligns interests
today, but concentration on a single supplier with this level of leverage is a structural
vulnerability that cannot be diversified away.
NVIDIA relationship details from CRWV S-1, 10-K filings, and investor presentations. Custom silicon data from Google, Amazon, and Microsoft public disclosures.
Competitive landscape -- no oligopoly, no durable moat
Competitive Position Assessment (Oligopoly Gate: FAIL)
| Competitor | Type | GPU Cloud Exposure | Scale / Moat |
|---|---|---|---|
| Hyperscalers (Building Own Capacity) | |||
| AWS / Azure / GCP | Hyperscale cloud | $100B+ combined annual capex | Unlimited capital, custom silicon, vertical integration |
| Independent GPU Cloud Providers (Direct Peers) | |||
| Lambda Labs | GPU cloud | Pure-play GPU cloud for AI | Smaller scale, developer-focused, growing fast |
| Nebius (NBIS) | GPU cloud | Full-stack AI cloud | Yandex spinoff, European footprint, well-funded |
| Custom Silicon Threat | |||
| Google TPUs / Amazon Trainium / MSFT Maia | Custom ASICs | Displacing NVIDIA GPUs for internal workloads | Structural threat to the GPU cloud model itself |
No Oligopoly
GPU Cloud Market
AWS, Azure, GCP, Lambda, Nebius all compete
Speed
Primary Differentiator
Faster deployment than hyperscalers -- replicable advantage
Intermediary
Structural Position
Sits between NVIDIA and end customers -- can be disintermediated
FAIL
Oligopoly Gate
No durable competitive moat identified
CoreWeave is an intermediary in a market where both the supplier and the customers
are more powerful. NVIDIA controls the hardware supply. Hyperscalers (Microsoft,
Meta, OpenAI) are the primary customers but are simultaneously building their own GPU
capacity and developing custom silicon. CoreWeave differentiates on deployment speed and
NVIDIA relationship quality, but neither constitutes a structural moat. The company
essentially provides a financing and operations wrapper around NVIDIA hardware -- a
valuable service in a supply-constrained market, but one that becomes less valuable as
supply normalizes. The $66.8B backlog provides near-term protection, but long-term
competitive positioning remains uncertain.
Competitive data from company filings, industry research, and public disclosures. Custom silicon programs from Google, Amazon, and Microsoft announcements.
Key risks to the business model
| Risk | Timeframe | Severity | Detail |
|---|---|---|---|
| Extreme Leverage | Structural | Critical | D/E 8.94x, >$14.6B debt, $1.22B annual interest -- any disruption to capital markets access could be existential |
| NVIDIA Dependency | Structural | Critical | Sole GPU supplier -- if NVIDIA shifts priorities, goes direct, or custom silicon gains share, the model breaks |
| Customer Concentration | Structural | Critical | OpenAI ($11.9B), Microsoft, Meta dominate revenue -- hyperscalers could insource as their own capacity expands |
| No Free Cash Flow | Near-term | High | FY2025 capex $14.9B vs. $5.1B revenue -- FY2026 capex $30-35B vs. $12-13B revenue -- FCF negative for years |
| AI Capex Cycle Risk | Medium-term | Moderate | If AI infrastructure spending decelerates, CoreWeave is left with depreciating GPU assets and fixed debt obligations |
| Hardware Obsolescence | Ongoing | Moderate | GPU generations turn over every 12-18 months -- depreciating assets must be replaced continuously at massive cost |
| Unproven Through Downturn | Structural | Moderate | IPO March 2025 -- less than 13 months public -- no track record through reduced demand or capital market stress |
Risk assessment from CRWV 10-K, S-1, earnings calls, and analyst commentary. Beta and short interest data from market data providers.