Margin Traded for Capacity, and the Old GPUs Held Their Price Again: Maintaining CoreWeave at Outperform

Results vs. Consensus

Full Year 2025

The four reported quarters sum to $5,131.1M against the reported full-year figure of $5,131M, so the annual column foots to the quarterly disclosures. CoreWeave became, on management's claim, the fastest cloud in history to reach $5B of annual revenue, and it did so while swinging from $324M of operating income in 2024 to a $46M operating loss in 2025. Both facts are true and neither is more important than the other.

Quarter-Over-Quarter Comparison

Quality of the Miss

Revenue: Fine, and the least interesting line in the report. A 1.4% beat on a $1.55B consensus, up 15.2% sequentially and 110% year over year. The full-year outcome of $5,131M landed in the upper half of the $5.05–5.15B range management set in November after the powered-shell delay, which means the company delivered the revised plan despite a fourth quarter in which it was simultaneously running the largest deployment programme in its history.

Margins: This is the whole quarter. Adjusted operating income of $88M against $217.2M in Q3 is a 59.5% sequential decline on 15.2% sequential revenue growth. The mechanism is unambiguous and management stated it in a single sentence: infrastructure was deployed ahead of expectations, so the costs of energizing 260 MW landed in Q4 while the revenue from that capacity begins in 2026. The CFO framed it as an overshoot on execution rather than an undershoot on cost control, and the capex number corroborates that: $8.2B in one quarter against $1.9B in Q3, with full-year capex of $14.9B exceeding the $12–14B guide.

EPS: Adjusted loss per share of $(0.56) against $(0.49) expected, and a GAAP net loss of $(452)M against $(110)M in Q3. There is no tax benefit flattering the comparison this time, and the sequential deterioration is honest: pre-tax loss went from $(236.8)M to $(467)M as operating income swung negative and interest expense grew another 24.9%.

Business Drivers

Capacity: 850 MW Live, 3.1 GW Contracted, and a 5 GW Objective Beyond That

The fourth quarter was the largest capacity increase in the company's history by a wide margin, and the size relative to the base is the reason the income statement looks the way it does. Adding 260 MW to an installed base that ended the third quarter at roughly 590 MW means roughly 30% of the year-end fleet was energized in the final three months, incurring lease, power, staffing and depreciation costs against revenue that starts in 2026.

The deployment ratio finally improved: active power as a share of contracted power moved from roughly 20% to roughly 27%, the first meaningful improvement in our coverage, because active capacity grew 44% sequentially while contracted capacity grew only about 7%. Management expects virtually all of the 3.1 GW contracted position online by the end of 2027, which is a specific and gradeable claim.

The new item is the 2030 objective: more than 5 GW of additional data center capacity beyond the already-contracted footprint. That is a statement of ambition rather than a contracted position, and it should be read as such, but it does frame the scale management is planning toward.

"We ended the year with more than 850 megawatts of active power, adding approximately 260 megawatts in the fourth quarter alone across 43 active data centers, up from 32 at the start of the year. We contracted close to 2 gigawatts of additional power in 2025, ending the year with more than 3.1 gigawatts of contracted capacity, virtually all of which we expect to come online by the end of 2027." — Michael Intrator, Chairman and CEO

Assessment: Execution on capacity was the strongest part of the quarter and it was accomplished while clearing the third-quarter delays early. The deployment ratio improving is the structurally important marker, because it is the ratio that converts contracted power from a capital obligation into an earning asset. At roughly 27% and heading toward roughly 55% by year-end 2026, the conversion is finally visible.

Backlog: $66.8B, Longer Duration, and Broader

Duration is the most valuable of these. A weighted average contract length moving from roughly four years to roughly five extends the period over which each deployed megawatt is contractually spoken for, which does three things at once: it lengthens the revenue tail on assets already financed, it strengthens the collateral case for the asset-level debt that funds them, and it implicitly validates customers' own view of how long the hardware stays useful.

The customer-breadth metrics moved the same way. Roughly twice as many new reserved-instance customers were added in Q4 as in any prior quarter, and the count of customers committed to at least $1M of annual spend grew nearly 150% across 2025. Named additions in the quarter (Cognition, CrowdStrike, Cursor, Mercado Libre, Midjourney, Runway) sit well outside the frontier-lab core.

Assessment: Backlog growth of $11.2B is a deceleration from the third quarter's $25.5B, which is expected given that Q3 contained both the Meta and OpenAI expansions. The composition improved on every disclosed axis. The gap in the disclosure is that the concentration statistic management led with last quarter, no single customer above roughly 35% of backlog, was not repeated. We do not read a single omission as deterioration, but a metric introduced in one quarter and dropped in the next is worth tracking rather than assuming.

Capital Structure: $21.4B of Debt and a Falling Cost of It

Debt grew 52% in a single quarter and 170% across the year. Against that, three genuinely favourable financing facts: the weighted average interest rate fell 300 basis points during 2025 and nearly 600 basis points since 2023; there are no maturities until 2029 other than self-amortizing contract-backed debt and OEM vendor financing; and the company secured more than $18B of debt and equity in 2025 across more than 200 investment partners. The inaugural convertible offering raised approximately $2.6B after being upsized on demand, and the revolver expanded to $2.5B from $1.5B.

Management sized the rate improvement usefully: the 300 basis point reduction represents nearly $700M of annualized interest savings against the fourth-quarter debt balance. That is a large number, and it is also a reminder of how large the balance has become, because $700M is what 300 basis points is worth on $21.4B.

The working capital picture deserves separate attention. Receivables reached $3,169M against Q4 revenue of $1,572M, which is roughly 185 days of sales outstanding, up from 112 days in Q3 and 145 days in Q2. Set against that, deferred revenue reached $8,185M. The net customer position is therefore a positive float of roughly $5.0B: customers have prepaid CoreWeave far more than CoreWeave has extended to them. That is the correct way to read the balance sheet, and it materially softens the receivable concern. But a 185-day collection cycle in a business with $21.4B of debt is a metric that deserves an explanation, and none was offered.

Assessment: The financing execution remains the operational strength of this company, and the path toward investment grade that management named is credible given the duration and credit quality of the backlog. The concern is not the cost of the debt; it is the quantity, and the quantity is a direct function of a capital programme that management has just guided to more than double again.

Key Topics & Management Commentary

Overall Management Tone: Assertive and unusually forward-committing, with the assertiveness pointed squarely at the margin question that dominated the call. Management pre-empted the criticism rather than defending against it: the margin was spent deliberately, the mechanism was explained twice in prepared remarks and twice more in Q&A, and a long-term target and two forward run-rate figures were volunteered to frame it. The posture on demand was the most confident of the four quarters we have covered. Where it was least convincing was on the near-term revenue guide, where a first-quarter number 15% below the Street was presented without acknowledgment that it would be read as a shortfall.

1. What the Margin Bought

The fourth quarter's adjusted operating margin of 5.6% is the lowest of any quarter in our coverage, and the explanation is contained in a single capacity number. CoreWeave energized approximately 260 MW in three months, against roughly 120 MW in the third quarter and an installed base that entered the quarter at roughly 590 MW.

"We bought on 260 megawatts worth of power in Q4. It's fully 1/3 of our installed capacity. So the variance that you're going to see there is higher. As we continue to grow and scale our company, as the incremental data center capacity that we bring on becomes relatively smaller, you will see less variance from us." — Michael Intrator, Chairman and CEO

The variance argument is mathematically correct and it is also the argument that stops working if capacity keeps compounding at this rate. Management guides active power to double again in 2026 to more than 1.7 GW, which means the incremental deployment stays large relative to the base for at least another year. That is precisely why Q1 2026 adjusted operating income is guided to $0–40M.

Assessment: We accept the explanation because the capacity, capex and construction-in-progress numbers all corroborate it independently. What we do not accept is the implicit framing that this is transitional. On management's own guidance the margin does not return to low double digits until the fourth quarter of 2026, which is four more quarters of paying for capacity before it earns. The Q4 print is not an anomaly; it is the first quarter of a two-year pattern.

2. Residual Value: Observations Two and Three

Last quarter's disclosure of a 10,000-plus H100 contract recontracting at within 5% of its original price was the basis of our upgrade, and we noted that one observation is not a curve. This quarter management supplied two more, on a portfolio basis rather than a single contract.

"We are also seeing a significant increase in demand for prior generations of GPU architectures, where supply also remains constrained. Average H100 pricing in Q4 was within 10% of where it started the year, while average A100 pricing increased in 2025. From our customers, we understand the demand for this infrastructure is largely for inference use cases, which are proliferating rapidly. We are signing this infrastructure into new reserved instance contracts ahead of when it becomes available, firmly putting to bed concerns about demand for older generation SKUs." — Michael Intrator, Chairman and CEO

Three things are load-bearing here. First, these are average portfolio prices across a full year rather than a single renewal, which is a materially stronger form of evidence. Second, A100 pricing increased during a year in which two newer architectures shipped, which is the opposite of what a commodity-depreciation model predicts. Third, the capacity is being contracted before it comes off its existing contracts, which means the constraint binds even at the trailing edge of the fleet.

The stated reason is inference. Inference workloads are less sensitive to peak training performance and more sensitive to cost per token, which makes older, cheaper silicon economically rational rather than merely tolerable. If that demand is structural, prior-generation assets have a genuine second life at close to first-life pricing.

Assessment: This is the most important thing in the quarter and, as with the H100 renewal last quarter, it was not what the market traded on. Depreciation is 45 percentage points of margin in this business (see topic 4), so the useful life applied to accelerators is the single largest determinant of terminal value. Three consecutive favourable observations, now including a portfolio-level average, is enough for us to treat the melting-asset thesis as substantially weakened rather than merely unproven.

3. The 2026 Plan: $12–13B of Revenue for $30–35B of Capital

Guidance for 2026 is revenue of $12–13B (approximately 140% growth at the midpoint), adjusted operating income of $900M–1.1B, capex of $30–35B, and active power doubling to more than 1.7 GW. Management stated that substantially all of the capex is tied to already-signed customer contracts intended to come online during the year.

"We expect 2026 CapEx of $30 billion to $35 billion, which is more than double our 2025 investment. Substantially, all of it is tied to our already signed customer contracts that we intend to bring online this year as we expect to double our active power capacity to more than 1.7 gigawatts by year-end." — Nitin Agrawal, Chief Financial Officer

The efficiency of that capital is calculable. Fourth-quarter revenue annualizes to $6,288M. Management guides an exit-2026 annualized run rate of $17–19B. So roughly $10.7–12.7B of incremental annualized revenue is being purchased with $30–35B of capital, a ratio of approximately 2.8 times at the midpoints. Applied to the mature contribution margin management cites, that is roughly $2.9B of annual operating profit against $32.5B deployed, a 9% pre-tax return before financing costs. Applied to the mature EBITDA margin management also cites, it is roughly $8.2B of annual cash profit against the same capital, a 25% cash return before financing.

Assessment: The two return figures bracket the entire investment case, and the difference between them is again depreciation. At a 9% return, this capital programme barely clears its own cost of debt and creates little equity value. At a 25% cash return over an asset life longer than the depreciation schedule, it compounds substantially. The pricing evidence in topic 2 is what tips the estimate toward the second reading, which is why the two disclosures have to be read together rather than separately.

4. The Long-Term Margin Target, and Why Its Two Halves Are Consistent

Management put a long-term adjusted operating margin target of 25–30% on the record for the first time, supported by two supplementary figures: mature revenue contracts generate contribution margins in the mid-20s, and the EBITDA margins on those same contracts are in the 70s.

"As our business and growth normalize, we remain confident in our ability to achieve 25% to 30% margins over the long term. Our mature revenue contracts generate contribution margins in the mid-20s which combined with the ramp-up of margin-accretive products and services we continue to unlock gives us conviction in our ability to achieve this target." — Nitin Agrawal, Chief Financial Officer

Those two margin figures are worth checking against each other, because together they encode the depreciation assumption. A roughly 45-percentage-point gap between a 70% EBITDA margin and a mid-20s contribution margin is depreciation. Applied to the roughly $11.7B of incremental annualized revenue the 2026 programme is meant to produce, that is roughly $5.3B of annual depreciation against roughly $32.5B of capital, which implies an asset life of approximately six years. The two disclosures are therefore internally consistent with the company's stated depreciation schedule, which is a useful check that the target is not being constructed from incompatible assumptions.

Assessment: The target is arithmetically coherent, and the honest caveat is that the mid-20s contribution margin sits below the 25–30% target range. Getting from one to the other requires the margin-accretive products and services (storage, software, the cloud-stack licensing) to contribute, and those are today a rounding error on revenue. The infrastructure business alone gets you to the bottom of the range at best; the target's upper half is a software bet.

5. The Q1 2026 Guide, and the Number Nobody Wants to Look At

First-quarter revenue is guided to $1.9–2.0B against a Street looking for $2.29B, with adjusted operating income of $0–40M, capex of $6–7B and interest expense of $510–590M. Management called Q1 the trough of the annual margin trajectory.

The pairing to sit with is $510–590M of interest expense against $0–40M of adjusted operating income. Even at the most favourable combination, interest is roughly thirteen times operating profit; at the unfavourable end, operating profit is zero and the ratio is undefined. For the four quarters we have covered, that ratio has gone 1.34, 1.43, 4.41, and now guided to at least 12.8. It has never improved.

Assessment: This is our first bear pillar and it is now at its most extreme. The defensible reading is that a company whose active power doubles annually will always show interest ahead of operating profit, because the debt precedes the revenue by design, and that the ratio is a measure of growth rate rather than of solvency. The undefendable reading is that a business can plan its way to a quarter with zero operating profit and half a billion dollars of interest and describe it as a trough. Both readings are available from this guidance, and the first quarter of 2027 is when we find out which was right.

6. The Third-Quarter Delay, Cleared Early

In November management committed categorically that the overwhelming majority of the powered-shell delay would be resolved within the first quarter of 2026. It was resolved in the fourth quarter of 2025 instead.

"We delivered these results while quickly resolving the data center delays we discussed last quarter, delivering the impacted deployments ahead of our Q3 earnings call expectations. … And in total, we have now delivered more than 50,000 Grace Blackwells to the impacted customer, deploying servers on a rolling basis and delivering them within weeks of receiving access to the requisite data center infrastructure." — Michael Intrator, Chairman and CEO

This matters beyond the quarter's numbers because it was the single largest credibility risk we identified at the last report. Management made an unhedged commitment about a third party's schedule, and beat it. The operational detail supports the claim: more than 50,000 Grace Blackwell systems delivered to one customer, on a rolling basis, within weeks of each tranche of data center access.

Assessment: Credibility earned. Equally, it is the direct cause of the margin miss, which is a useful illustration of how this business trades one line for another. Beating a delivery schedule is unambiguously good and it lands in the P&L as a cost overrun. An investor who cannot hold both of those thoughts will misread every quarter of this company's reporting.

7. NVIDIA: Investor, Supplier, Distribution Partner

In January, NVIDIA expanded its commercial relationship with CoreWeave and invested $2B. Separately, NVIDIA intends to test and validate CoreWeave's platform, software and reference architectures toward inclusion in NVIDIA's own reference architecture for cloud, enterprise and sovereign customers. Management also flagged that select customers already license SUNK, CoreWeave's Slurm-on-Kubernetes scheduler, as their default cluster management platform across multi-cloud footprints.

The strategic prize is licensing the proprietary cloud stack outside CoreWeave's own data centers, which converts a capital-intensive infrastructure business into something with a software revenue line attached. Management explicitly excluded any of this from the 2026 guidance while saying it expects contribution to begin in 2026.

Assessment: Genuine optionality, excluded from guidance, and therefore free in the current price. The caution is the same one we raised at the NVIDIA backstop agreement last quarter, only larger: NVIDIA is now CoreWeave's principal supplier, a shareholder, a demand backstop, and a distribution channel. Each relationship is individually sensible and the aggregate is a concentration of dependency on one counterparty that no disclosure currently quantifies.

8. Where the Diversification Stands

The enterprise and AI-native cohort produced its best quarter of disclosure: approximately twice as many new reserved-instance customers as any prior quarter, a nearly 150% increase across 2025 in customers committed to at least $1M of annual spend, and named additions including Cognition, CrowdStrike, Cursor, Mercado Libre, Midjourney and Runway. Storage attach among customers spending at least $1M per year reached approximately 80%.

Management also said contracts with enterprise customers "largely look very similar to the hyperscale contracts in tenure," which if accurate means the diversification is not being bought with worse terms.

Assessment: The 80% storage attach rate is the most useful number in this cluster, because it is the first evidence that the platform strategy produces multi-product relationships rather than single-product ones. The gap remains that none of these metrics is denominated in revenue, so their materiality is unmeasurable from outside. A customer count growing 150% tells you nothing about the revenue share it represents.

9. Working Capital: 185 Days Out, $8.2B In

Receivables reached $3,169M, roughly 185 days of sales outstanding on the fourth quarter, against 112 days at the end of the third. The receivable build consumed $1,496M of cash in the quarter and $2,749M across the year. Deferred revenue meanwhile reached $8,185M, having grown $2,772M in the quarter and $4,174M across the year, which is why full-year operating cash flow was positive $3,058M despite a $1,167M net loss.

The netting matters. Customers have prepaid roughly $8.2B against roughly $3.2B of unpaid invoices, a net float of roughly $5.0B in CoreWeave's favour. In cash terms the customer base is funding the build, not draining it.

Assessment: Less alarming than the headline days-sales figure suggests, and still unexplained. A 73-day sequential increase in collection time is the kind of movement that normally draws a question, and it drew none. Extended terms to large creditworthy counterparties is the benign explanation and is probably the right one, but it is an inference rather than a disclosure.

10. Contract Duration Extends to Five Years

The weighted average contract length increased from roughly four years to roughly five during 2025, and management indicated the trend is continuing into Rubin-generation contracts.

"… one of the trends that we're seeing is the extension of the contracts from an average of four years up to an average of five years. And obviously, that is great for us. It's stabilizing for our business. It gives us a lot of confidence around building and scaling the infrastructure. And so that's sort of the first thing I'd highlight to you is that the clients that are coming in and using this infrastructure, they're gaining more confidence in the longevity of the usefulness of the infrastructure." — Michael Intrator, Chairman and CEO

The last clause is the analytically important one and it connects directly to topic 2. Customers extending contract duration is customers expressing a view about how long the hardware stays economically useful, backed by their own money. It is an independent corroboration of the residual-value evidence, arriving from the demand side rather than the pricing side.

Assessment: A five-year weighted average against a roughly six-year depreciable life means the large majority of each asset's book value is contractually spoken for from the day it is energized. That is a materially stronger position than a four-year average against the same schedule, and it is the quiet structural improvement of the year.

11. The 2027 Framework: More Than $30B of Run-Rate Revenue

Management guided an exit-2026 annualized run rate of $17–19B and an exit-2027 run rate of more than $30B, and was pressed on how much of that is booked.

"… what we are doing is we are taking the contracted power that we have, and we are projecting out when the existing contracts that have already been sold, and like I said, we are virtually sold out in 2026 of all of our capacity and then continuing to add contracts that will be allocated once they come online in 2027." — Michael Intrator, Chairman and CEO

The answer separates the two years honestly. 2026 is described as virtually sold out, which is consistent with the claim that substantially all 2026 capex serves signed contracts. 2027 is contracted power that is not yet allocated to customers, so the $30B figure requires selling capacity that exists on paper but has no counterparty attached to it yet.

Assessment: A two-year forward run-rate framework is unusual disclosure and mostly welcome, but the two years carry very different evidentiary weight and management did not draw the distinction as sharply as the numbers deserve. We underwrite the exit-2026 figure, which is capacity-constrained rather than demand-constrained. We treat exit-2027 as an ambition supported by contracted power rather than by contracted revenue, and we do not use it in our valuation.

Guidance & Outlook

The guidance is internally consistent, which is worth stating because it is not always. Full-year revenue of $12.5B at the midpoint against a Q1 of $1.95B requires the remaining three quarters to average roughly $3.5B, which implies substantial sequential acceleration through the year. That acceleration is exactly what doubling active power to more than 1.7 GW produces, and it reconciles to the exit-2026 run rate of $17–19B (a fourth quarter of roughly $4.25–4.75B). The margin shape follows the same curve for the same reason.

What is excluded. Management stated that the 2026 guidance contains no contribution from monetizing the proprietary cloud stack to other NVIDIA Cloud, enterprise or sovereign customers, while expecting that contribution to begin during the year. Guidance that explicitly excludes an expected revenue source is unusual and, in a quarter where the near-term guide disappointed, arguably self-defeating. It is also the honest treatment of revenue that has no contracted base.

Guidance style, four quarters in. The pattern is now legible. Quarterly revenue guidance has been met or beaten every time; the full-year revenue guide was cut once, in November. Profit guidance has been cut once and missed once. Capex guidance has been missed in both directions within six months, low in Q3 and high in Q4, for the same underlying reason: deployment timing, which management does not fully control and has stopped pretending it does. The useful conclusion for a modeller is that CoreWeave's revenue line is more predictable than its margin line, and its capex line is not predictable at all on a quarterly basis.

Analyst Q&A Highlights

Why Faster Deployment Did Not Produce a Better Revenue Guide

The opening question identified the central tension in the print directly: capex ran ahead, delays resolved early, yet active power was only in line and the near-term revenue guide came in well below the usual level of upside. The answer located the disconnect in timing within the quarter.

Q: "You came in nicely ahead on CapEx, and it's great to hear the delivery delays resolved quicker than expected. Trying to align that with seeing the active power, which is more in line and revenue guidance in the range, which is well below like the typical level of upside. So I was hoping you could unpack some of those dynamics. If you're moving faster, why didn't that show up in the active power and the revenue?"
— Josh Baer, Morgan Stanley

A: "As we deploy capacity, a lot of that capacity came online towards the end of the quarter, and you're going to start seeing the monetization of it in 2026. We continue to build our capacity at a rapid pace. … Relative to the Q1 number, like we are basically providing the guidance for the first time for 2026 and Q1 at this moment."
— Nitin Agrawal, Chief Financial Officer

Assessment: The first half of the answer is complete and correct: capacity energized in December carries a full quarter of cost and almost no revenue. The second half is a deflection. Noting that this is the first time Q1 has been guided does not address why the number is 15% below where the Street had modelled it, and the question was asked precisely because the deployment beat should have argued for the opposite.

How the Cost of Capital Is Falling and What NVIDIA's Involvement Does to It

A multi-part question on financing drew the most quantitatively substantive answer of the call, covering the trajectory of borrowing costs, how data center operators underwrite CoreWeave as a tenant, and whether an investment-grade counterparty in the structure changes the arithmetic.

Q: "… can you sort of quantify where you estimate your blended cost of capital is? How has that really evolved over the last 12 months? And when you negotiate with these data center operators, how do they assess your credit profile? … And then just on the financing side, does the NVIDIA credit support, the guarantor framework help translate into a measurable step down in your borrowing cost, you think, in '26?"
— Amit Daryanani, Evercore

A: "You've seen our cost of capital drop 300 basis points in the last 12 months. You've seen it drop 600 basis points over the last 2 years. We expect that, that will continue. … obviously, working with an investment-grade counterparty as the offtake will have an impact on the cost of capital or the cost that is associated with the data center."
— Michael Intrator, Chairman and CEO

Assessment: The rate trajectory is the single most reliable positive in this story and it is the one management can most directly influence. The NVIDIA answer was deliberately hedged (management works with NVIDIA "selectively, but certainly not exclusively"), which is the correct posture: an investment-grade offtake behind a data center lease genuinely lowers the cost of that capacity, and building a financing model dependent on one counterparty's balance sheet would trade one concentration for another.

Whether the Guidance Framework Itself Is Reliable

The bluntest exchange of the call questioned the guidance process rather than any single number, noting that revenue landed in line, operating income below, and capex far above, all within one quarter of the ranges being set.

Q: "I know when you look at the revenue guide, you were in line, op income a little lower and your CapEx was way higher. I guess it just — it kind of illustrates even the guide you gave us all that some of the metrics can really vary. I'm just curious just in terms of how you're thinking about the guide going forward. … some of the numbers were effectively kind of outside the range of what you initially gave us."
— Brent Thill, Jefferies

A: "Q4 alone, we brought 30% of our total active power base, which naturally creates some near-term margin compression as capacity costs ramp ahead of full revenue maturity and recognition. As I mentioned in my remarks, Q1 represents the trough of what we would see in margins. And then from there on, as we scale into the capacity deployed, we will expand margins quarterly from there, returning to low double digits by Q4."
— Nitin Agrawal, Chief Financial Officer

Assessment: The answer explains the variance without addressing the question about the process. That said, the honest position is that quarterly capex guidance for a company whose spending recognition is gated on third-party construction schedules may simply not be a forecastable quantity, and pretending otherwise would be worse. An investor should model the annual figures and treat the quarterly ones as indicative.

Whether Unit Economics Differ Across Customer Types

A question on customer-cohort economics tested whether the enterprise diversification is being achieved by accepting worse terms, and produced a useful restatement of how contracts are structured and priced.

Q: "I'm curious if you can share with us your observations on how the unit economics or how the attractiveness of how these customers are using the CoreWeave platform is different between types of customers. … I know that your pricing model is based on dollars per GPU and then the length of the committed contract. But curious if you could share your observations on customer behavior across the different cohorts."
— Gabriela Borges, Goldman Sachs

A: "Fundamentally, as Mike described, like our contracts look mostly similar across our customer base with the exception, of course, the volume element that we look at things when we are talking about larger customers versus smaller customers. But across the board in our customer profile, we look to generate similar economics for the infrastructure that we are generating as with the market dynamics go on."
— Nitin Agrawal, Chief Financial Officer

Assessment: Economics that are similar across cohorts, with volume as the only material differentiator, means diversification does not come at the price of margin. That is an important and under-discussed point: many businesses buy customer diversification by discounting to smaller buyers. If CoreWeave is not doing that, the concentration improvement is free rather than purchased.

Whether the Capital Spending Rate Ever Moderates

A question on the interaction between capex growth and the long-term margin target pressed on whether the spending must continue at this rate to deliver the promised economics, and drew the clearest articulation of the success-based model.

Q: "… there's all this CapEx and not enough margin or cash flow necessarily in the near term. … What is your confidence in the rate of the CapEx growth? Like we understand you're spending now to get the revenue but what is — does that growth rate moderate more than we're thinking as those margins go up?"
— Benjamin Reitzes, Melius Research

A: "… our business is built on a success-based model where clients come to us and buy long-term contracts to get access to the infrastructure to build their business. … And so when a hyperscaler comes to us and buys infrastructure for 5 years, they are going to be purchasing that infrastructure from us at a fixed price for 5 years. … because we know not only what we're going to make now, but we know what we're going to make as those contracts move through the 5-year cycle that they have been put under a take-or-pay contract."
— Michael Intrator, Chairman and CEO

Assessment: The fixed-price, five-year, take-or-pay structure is the reason management can put a long-term margin target on the record at all, and the answer is the right one. It also quietly concedes the question: nothing in it suggests the capital spending rate moderates, because the model spends whenever a contract is signed and management has just told the market it is virtually sold out for 2026 and selling into 2027.

What Rubin-Generation Contracts Look Like

Looking forward to the next architecture, a question on contract structure surfaced the duration extension and a useful comment on how prepayments function as a financing lever that is becoming less necessary.

Q: "… as we look to Rubin contracts ahead, are you seeing demand for similar 5-year duration? And is there anything different about how these deals are priced, the amount of prepayment that you'd expect or anything else? And how should we think about the ROIC on these deals versus prior generation contracts and the economics that you've outlined in your original S-1?"
— Brad Zelnick, Deutsche Bank

A: "The pricing on the infrastructure, we kind of take a — we think about that from a margin perspective. And as the infrastructure changes in price, we are altering our pricing of how we deliver it to clients in order to kind of target the type of margins that Nitin spoke to. … Now as far as prepay goes, prepayment is always a lever which we can use with clients to change the economics around a contract. One of the things that's very exciting for us is as we continue to drive down our cost of capital, our dependency on prepayments is reduced."
— Michael Intrator, Chairman and CEO

Assessment: Pricing to a target margin rather than to a market rate is the answer of a supplier with pricing power, and it explains why margins on mature contracts are describable as a stable band. The prepayment comment cuts the other way for anyone modelling cash flow: the $8.2B deferred revenue balance that has been funding the build is a lever management expects to use less as debt gets cheaper, so operating cash flow should be expected to converge toward economic reality over time rather than continuing to flatter it.

What Underpins the 2027 Run-Rate Number

The final question of the call asked directly how much of the more than $30B exit-2027 run rate is already contracted, and the answer separated 2026 from 2027 in a way the prepared remarks had not.

Q: "Can you just speak to what gives you confidence in the $30 billion run rate by 2027? And how much of that is already booked versus business your team needs to go get?"
— Michael Turrin, Wells Fargo

A: "… we are taking the contracted power that we have, and we are projecting out when the existing contracts that have already been sold, and like I said, we are virtually sold out in 2026 of all of our capacity and then continuing to add contracts that will be allocated once they come online in 2027. And we have vast and sustained interest from our clients to get more capacity to bring on more compute."
— Michael Intrator, Chairman and CEO

Assessment: The most useful answer for a modeller and the one that should govern how the two run-rate figures are used. Exit-2026 rests on signed contracts against capacity that is already contracted and largely under construction. Exit-2027 rests on contracted power with customers still to be attached. Those deserve very different confidence weights, and we assign them accordingly.

What They're NOT Saying

1. Customer concentration, on any basis. The backlog-concentration statistic management introduced and led with last quarter (no single customer above roughly 35%) was not repeated. Revenue-basis concentration has not been disclosed since the second quarter, when it was 71%.
2. Why days sales outstanding rose 73 days in one quarter. Receivables grew $1.5B sequentially to 185 days of sales. No commentary, no question.
3. Free cash flow, or any framework for when it turns. Operating cash flow of $3,058M against capex of $14.9B is the defining financial fact of 2025, and neither figure was placed next to the other by management or by any analyst.
4. Utilization. Four quarters as a public company, no utilization disclosure of any kind. For a business whose entire margin structure depends on how fully its assets are consumed, this remains the largest single gap.
5. The size of the NVIDIA relationship in aggregate. NVIDIA is supplier, shareholder ($2B invested), demand backstop ($6.3B agreement), and prospective distribution channel. No disclosure sizes the combined exposure.
6. Depreciation life. D&A reached $821M in the quarter, 52% of revenue. The useful-life assumption is the most consequential estimate in the financial statements and drew no commentary, on a call where management volunteered two datapoints that bear directly on it.
7. What "substantially all" of 2026 capex means precisely. The claim that substantially all of $30–35B is tied to signed contracts is the foundation of the entire investment case for the spending. It was stated twice and quantified never.
8. Any bridge from the mid-20s mature contribution margin to the 25–30% long-term target. The target's lower bound equals the current mature-contract margin, so the entire upper half depends on products and services that management simultaneously excluded from 2026 guidance.

Market Reaction

• Pre-print setup: close of $97.63 on February 26, up 36.3% year to date from $71.61 at the end of 2025 but down 10.3% over the trailing 30 days from $108.86 on January 27. The year-to-date strength was substantially a January event, following the announcement of NVIDIA's investment.
• After-hours move: shares fell approximately 8% in extended trading on February 26.
• Next-day session: opened at $84.22, a 13.7% gap down; traded a $75.66 to $84.38 range; closed at $79.56, down 18.5% and $18.07 on the day.
• Volume: 67.9M shares against a 30-day average of 26.9M, a 2.5 times multiple.
• Benchmark: the S&P 500 fell 0.4% on the session and was up 0.9% year to date entering the print. The wider software complex had been selling off through February.

For the second consecutive quarter the reaction more than doubled overnight, from roughly 8% after hours to 18.5% at the next close. The pattern is now consistent enough to be informative: this company's disclosures require arithmetic that the after-hours tape does not perform, and the derived figures are worse than the headlines.

The first-quarter guide was the proximate cause. Revenue of $1.9–2.0B against $2.29B expected is a 15% shortfall at the midpoint, and it arrived alongside an adjusted operating income guide of $0–40M. For an investor whose model assumed a linear ramp from the fourth quarter, both numbers require a rebuild.

The capex number landed into an unreceptive market. Capital spending of $30–35B against revenue of $12–13B is a ratio of roughly 2.6 times revenue, disclosed in a February in which AI capital intensity had become the dominant concern in technology and the broad software complex had de-rated sharply. The same guidance in a friendlier tape would have been read as confidence.

The setup had already given back its cushion. The stock entered the print up 36% year to date, almost all of it earned in January on the NVIDIA investment, and had already fallen 10.3% over the preceding thirty days. There was no accumulated goodwill to absorb a guide that needed explaining.

Our read is that the market priced the near-term guide accurately and the structural disclosures at zero for the second consecutive quarter. Portfolio-level evidence that H100 pricing held within 10% across a year and A100 pricing rose, contract duration extending to five years, the delay clearing a quarter early, and a first quantified long-term margin target are collectively more consequential to terminal value than one quarter of deferred revenue recognition. They are also harder to put in a headline.

Street Perspective

Debate: Is Spending 2.6 Times Revenue on Capex Conviction or Recklessness?

Bull view: The bull case on the Street holds that $30–35B against a $66.8B contracted backlog, with substantially all of it tied to signed contracts and a weighted average contract length of five years, is the least speculative large capital programme in AI infrastructure. CoreWeave does not build and hope; it builds against take-or-pay contracts at fixed prices with pre-arranged financing, and it has now demonstrated it can deploy 260 MW in a quarter.

Bear view: The bear camp contends that a business generating $3.1B of operating cash flow cannot responsibly commit $30–35B of capital in a single year, that the funding gap is necessarily debt on top of $21.4B already outstanding, and that "substantially all tied to signed contracts" is an unquantified assurance about the most important number in the story. In this reading the equity is a thin residual behind a rapidly compounding claim.

Our take: The bull framing is better supported on the evidence, and the bear framing describes the real risk accurately. Both are right because they are answering different questions: the capital is contractually committed (bull) and the equity is structurally subordinated (bear). The resolution is not a debate about the spending, it is the asset-life question, which determines whether the contracted revenue outlives the debt that funded the assets.

Debate: Does Prior-Generation GPU Pricing Hold, or Is This the Top of a Shortage?

Bull view: Three consecutive datapoints now point the same way: a 10,000-plus H100 contract recontracted early at within 5% of original price, average H100 pricing within 10% of where it started 2025, and average A100 pricing rising during a year when two newer architectures shipped. Inference demand is structurally less performance-sensitive than training, so older silicon has a genuine second market rather than a salvage market.

Bear view: The skeptical view is that every one of those observations was generated inside the tightest capacity conditions the industry has known, and that the same shortage which makes an A100 valuable in 2025 disappears when the industry's collective 2026 and 2027 builds energize. On that reading the pricing data measures scarcity, not durability, and mean reversion in prior-generation pricing arrives exactly when CoreWeave's own first cohort of assets reaches the end of its initial terms.

Our take: Bull, with the timing risk acknowledged. The bear argument is coherent and we cannot refute it from current data. But the evidence has now updated in the same direction three times, including once on a portfolio-average basis, and the demand-side corroboration (customers voluntarily extending contract duration to five years) comes from parties with their own money at risk. We would need a contrary observation to move back to neutral.

Debate: Can a Company Guiding to Zero Operating Profit and $550M of Interest Be an Outperform?

Bull view: A growing view holds that the first quarter is a mechanical artifact of doubling active power in a year, that management has been explicit that Q1 is the trough with margins expanding each quarter thereafter to low double digits by Q4, and that judging a build-ahead business on the quarter with the heaviest build is precisely the wrong frame. The exit-2026 run rate of $17–19B is the number that matters.

Bear view: The bear framing is that the interest-to-operating-income ratio has deteriorated in every quarter of this coverage, has never once improved since the second quarter of 2025, and is now guided to a level at which it is barely calculable. Every quarter has carried an explanation and every explanation has been superseded by a larger build. At some point a pattern stops being a series of exceptions.

Our take: The bear has the stronger empirical record and the bull has the better model. We resolve it by not resolving it: our base case assumes the ratio improves through 2026 as guided, and our bear case assumes it does not, and the two produce a $12 stock and a $117 stock respectively. That dispersion is the honest output, and it is why the position sizing matters more here than the rating letter.

Model Update & Valuation Framework

Method change, stated explicitly. We move the valuation base from forward-year revenue to management's newly disclosed exit-2026 annualized run rate. The reason is that CoreWeave's reported annual revenue systematically understates its earning power at any point in time, because capacity energizes continuously through the year and each megawatt earns for only part of it. A run-rate base removes that distortion. We use the exit-2026 figure, which management describes as resting on capacity that is virtually sold out. We do not use the exit-2027 figure of more than $30B, because that capacity is contracted at the power level but not yet allocated to customers.

At the February 27 close of $79.56 and 506M shares, equity value is approximately $40.3B. Adding total debt of $21.37B and subtracting cash, restricted cash and marketable securities of $4.16B gives an enterprise value of roughly $57.5B, which is 4.6 times the midpoint of the FY2026 revenue guide and 3.2 times the midpoint of the exit-2026 run rate.

Twelve-month price target: $99 (range $81–117). The target falls from $107 despite a higher revenue base because the net debt assumption widens by roughly $10B on the doubled capital programme. That is the correct mechanical response and it illustrates the central dynamic of this equity: incremental capacity raises enterprise value and raises the claim ahead of the equity by a similar amount, so the per-share outcome turns on the spread between them rather than on growth.

What differentiates the scenarios. The bear case is a 2026 in which deployment slips materially, the exit run rate lands nearer $14B, and the market applies an infrastructure multiple to an equity behind $36B of net debt. Note that the bear equity value is small in absolute terms, which is not a modelling artifact: at this leverage the equity genuinely behaves like an option. The bull case requires no acceleration beyond guidance, only that the run rate lands at the top of the guided range and that the market pays a growth multiple rather than an infrastructure one, which in turn requires the residual-value evidence to be believed. The dispersion is extreme, and we would rather state it than smooth it.

Thesis Scorecard Post-Earnings

Grading Last Quarter's Commitments

Overall: Eight of ten commitments met or beaten, including the two categorical ones on which management had staked its credibility. The two failures are the fourth-quarter margin and the concentration disclosure. The pattern across four quarters is now clear enough to name: CoreWeave delivers on capacity, backlog, financing cost and asset-value evidence, and misses on near-term profitability, every time, for the same reason. An investor who owns this stock is being paid to tolerate the second in exchange for the first.

Action: Maintain Outperform with a $99 twelve-month target. Conviction is unchanged rather than raised: the structural evidence improved again, and the leverage that would punish a mistake grew faster.

What would take us to Hold: the guided margin ramp failing to appear, specifically a second-quarter adjusted operating margin that does not expand sequentially from the first-quarter trough; any new residual-value datapoint that contradicts the H100 and A100 pricing evidence; a data center or deployment delay that pushes the exit-2026 run rate below the guided range; or capex guidance rising again without a corresponding lift to the run-rate framework.

What would take us to Underperform: a counterparty credit event among the largest customers; a debt raise at materially wider spreads, which would reverse the single most reliable positive in the story and signal the credit market has repriced the collateral; evidence that the 25–30% long-term margin target is being walked back; or disclosure that the receivable build reflects collection difficulty rather than negotiated terms.