The Fully Loaded Cost of AI Compute
AI infrastructure profitability depends on the complete cost of delivering computing services. This chapter examines facility expenditure, technology depreciation, operating margins and workload-level cost allocation.
← Flagship ContentsAI infrastructure profitability begins with the cost of delivering computational capacity.
The purchase of accelerators establishes an investment. Facility operations create recurring expenditure. Technology assets contribute depreciation throughout their accounting lives. Engineering and software sustain the computing environment.
Each cost enters the financial statements according to its economic character and accounting treatment.
For investors, the relevant question concerns the resources required to generate a dollar of compute revenue and the financial contribution remaining after those resources are accounted for.
The Economic Boundary of Compute
A computing service draws upon an integrated operating system.
Accelerators execute workloads. Networking equipment supports communication. Electrical infrastructure supplies power. Cooling systems maintain operating conditions.
Software and engineering sustain the platform through which customers consume these resources.
The financial boundary must reflect the service being evaluated.
At the facility level, analysis may focus on energy consumption and operating expenditure.
At the infrastructure-platform level, the cost boundary expands to include equipment depreciation and technical operations.
At the enterprise level, the assessment incorporates broader operating expenses and capital requirements.
The selected boundary determines the meaning of the resulting margin.
What CoreWeave's Financial Statements Reveal
CoreWeave's fiscal 2025 results provide a documented illustration of infrastructure cost classification.
The company reported $5.131 billion in revenue and $1.453 billion in cost of revenue.
Its technology and infrastructure expense reached $2.929 billion.
Operating expenses also included $144 million in sales and marketing and $651 million in general and administrative expenditure.
The resulting operating loss was $46 million.
These amounts establish the company's reported operating economics for the year ended December 31, 2025.
| FY2025 financial measure | USD millions |
|---|---|
| Revenue | 5,131 |
| Cost of revenue | 1,453 |
| Technology and infrastructure | 2,929 |
| Sales and marketing | 144 |
| General and administrative | 651 |
| Operating loss | (46) |
The figures demonstrate how the financial interpretation changes as the cost boundary expands.
Facility Costs and Infrastructure Operations
CoreWeave classifies data-center facility expenses within cost of revenue.
These include rent and utilities. Depreciation associated with power installation and distribution systems also enters this category.
Personnel supporting data-center operations and customer success contribute further expense.
Technology and infrastructure costs follow a separate classification.
This category includes depreciation associated with computing equipment and networking systems. Technical personnel and software expenditure also contribute.
The separation reflects the company's reported accounting policies.
During 2025, cost of revenue increased by $960 million compared with 2024.
Management attributed part of that increase to approximately $566 million in additional rent expense and $203 million in additional utilities expenditure.
Depreciation associated with power installation and distribution systems increased by approximately $84 million.
These are reported year-over-year expense increases. They illustrate the operating consequences of infrastructure expansion.
The Margin Boundary
CoreWeave's reported revenue and cost of revenue produce gross profit of $3.678 billion.
The corresponding gross margin is approximately 71.7%.
This measure reflects the company's reported cost-of-revenue classification.
A broader analytical calculation incorporates technology and infrastructure expense.
Revenue of $5.131 billion less cost of revenue of $1.453 billion and technology and infrastructure expense of $2.929 billion leaves $749 million.
That represents approximately 14.6% of revenue.
| Analytical measure | FY2025 |
|---|---|
| Gross profit | $3.678B |
| Gross margin | 71.7% |
| Revenue after cost of revenue and technology infrastructure expense | $749M |
| Corresponding analytical margin | 14.6% |
| Operating margin | −0.9% |
The 14.6% figure is an analytical subtotal. Its purpose is to illustrate the effect of expanding the expense boundary.
Sales and marketing expenditure and general administrative costs remain outside that subtotal.
The company's full operating statement produces the reported $46 million operating loss.
Margin interpretation depends on the expenses included in its calculation.
Depreciation and Capital Consumption
Computing equipment represents a significant capital asset.
CoreWeave reported approximately $2.3 billion in depreciation and amortization within technology and infrastructure expense during 2025, compared with $843 million in 2024.
The increase reflects the expansion of the company's computing infrastructure.
Depreciation allocates qualifying asset costs across their estimated useful lives.
The corresponding capital expenditure follows the timing of acquisition and payment.
These measures serve different financial purposes.
Depreciation contributes to accounting profitability. Capital expenditure affects investment cash flows. Asset replacement introduces future capital requirements.
A comprehensive infrastructure assessment must consider all three.
The resulting financial analysis should connect current operating performance with the investment required to sustain future computing capacity.
From Infrastructure Cost to Workload Cost
Company-level financial statements establish aggregate operating economics.
Workload economics require a more specific allocation.
A GPU-hour represents a defined unit of accelerator availability or consumption. Its cost depends on the resources assigned to that service.
For a consumption-based model, the calculation may allocate relevant annual costs across billable GPU-hours.
A reserved-capacity contract requires treatment consistent with the contracted service and associated revenue.
The denominator must reflect the commercial arrangement.
Consider a hypothetical computing service generating six million billable GPU-hours annually.
Assume the relevant annual cost pool totals $18 million.
The allocated cost is $3 per billable GPU-hour.
At a realized commercial rate of $4 per hour, the contribution before expenses outside the selected cost boundary is $1 per billable hour.
This calculation assumes a consistent service definition and complete allocation of the stated cost pool.
The example illustrates how cost boundaries influence unit economics.
The Financial Cost of Capacity
Infrastructure providers invest in computing resources ahead of commercial consumption.
Capacity availability carries costs through the operating life of the assets.
Electricity expenditure varies with operating conditions. Equipment depreciation follows accounting schedules. Facility obligations depend on contractual arrangements.
Commercial utilization influences how these costs are absorbed across revenue-producing services.
A higher volume of billable computing activity can improve cost absorption when additional revenue exceeds incremental service costs.
The resulting financial contribution depends on realized pricing and the resources required to fulfill demand.
The economic assessment must therefore connect cost structure with contract terms and workload characteristics.
Investment Judgment
AI infrastructure profitability is determined by the relationship between commercial revenue and the full economic resources required to deliver computing services.
Facility expenses establish one operating boundary. Technology and infrastructure costs establish another. Corporate operating expenses complete the reported operating picture.
Capital expenditure and financing introduce further investment considerations.
The relevant measure depends on the decision being made.
The economics of compute become meaningful when the cost boundary reflects the assets and operating resources required to deliver the service.
Financial productivity emerges through sustainable contribution and disciplined capital recovery.
- 2025 Form 10-K CoreWeave Inc.March 2, 2026Company Filing
Fiscal year ended December 31, 2025.
- Relevant disclosures:
- Consolidated Statements of Operations and Comprehensive Loss.
- Management's Discussion and Analysis: Components of Results of Operations.
- Cost of Revenue.
- Technology and Infrastructure.
- Property and Equipment.
- Primary source:
CoreWeave financial figures reflect the company's FY2025 reporting classifications.
The 14.6% analytical margin uses an expanded expense boundary and represents an independently calculated subtotal. The GPU-hour example uses hypothetical inputs.
Company-level financial statements provide evidence of aggregate operating performance. Workload-specific profitability requires measured resource consumption and attributable cost data.