The Thesis Worked and the Price Caught It: Downgrading CoreWeave to Hold After a 43% Run

Results vs. Consensus

Year-Over-Year Comparison

General and administrative expense falling 6.3% year over year on 112% revenue growth is the one line that has not appeared before in this coverage, and it is worth marking. The prior-year quarter carried IPO-related costs, so the comparison flatters, but the direction confirms the CFO's claim that G&A is moderating relative to revenue. Every other expense line grew faster than revenue.

Quarter-Over-Quarter Comparison

Quality of the Quarter

Revenue: The best line in four quarters. Sequential growth of 32.2% is the fastest since we initiated, and it came in above a guide range management had set only ten weeks earlier. The composition is also better than the growth rate implies: pricing rose across every prior-generation GPU line, so this is not volume growth purchased with price.

Margins: Adjusted operating margin of 1.0% is the lowest in the company's public history, and it is exactly where management said it would be. That distinction matters. In the fourth quarter the margin missed the implied guide; this quarter it landed above the midpoint of an explicit range. The deterioration is now planned rather than surprising, which converts it from an execution question into a forecasting question.

EPS: Adjusted loss per share of $(1.12) against $(0.90) expected. The miss sits entirely below the operating line and is arithmetic rather than operational: interest expense doubled year over year to $536M, and D&A grew 159% to $1,147M as the fourth-quarter and first-quarter deployments began depreciating. There is no adjustment or one-off flattering or distorting the figure.

Business Drivers

Backlog: $99.4B, and the First Real Duration Disclosure

More than $40B of commitments signed in a single quarter, against a company that will book $12–13B of revenue for the whole of 2026. The named drivers are a $21B Meta agreement announced in early April, the addition of Anthropic, and a $6B capacity expansion from Jane Street. The counterparty roster now includes, on management's description, the world's four leading model developers and nine of the ten leading AI companies outside China.

The duration disclosure is new and useful: 36% of the backlog recognizes within 24 months and 75% within four years. On $99.4B that is approximately $35.8B by March 2028 and $74.6B by March 2030. The first of those figures is roughly three times the entire 2026 revenue guide, and it is contracted.

The credit-mix disclosure went the right way again. Commitments from non-investment-grade AI natives and foundation labs are now below 30% of the total, from an implied 40% two quarters ago. Two entirely new verticals crossed materiality thresholds: financial services is approaching $10B of backlog, and physical AI and spatial computing passed $1B.

"We delivered our strongest quarter for customer bookings, signing more than $40 billion of new commitments and growing contracted revenue backlog to nearly $100 billion. We generated approximately $2.1 billion of revenue, up 32% quarter-over-quarter and 112% year-over-year, and surpassed 1 gigawatt of active power, a milestone only a handful of cloud companies have ever achieved." — Michael Intrator, Co-Founder, Chairman and CEO

Assessment: The strongest bookings quarter of the coverage and, more importantly, the best-documented. Duration, credit mix, vertical composition and customer-count metrics were all disclosed together for the first time, which materially improves the ability to underwrite the number rather than take it on faith. The one caution: management noted these new commitments "mostly contribute towards our 2027 targets," so the $99.4B is increasingly a 2027-and-beyond asset rather than a 2026 one.

Capacity: Past One Gigawatt, With 3.5 Contracted

Crossing one gigawatt of active power is the milestone management chose to lead with, and the deployment ratio improved for a second consecutive quarter to roughly 29%. The quarterly cadence of contracted-power additions has been lumpy by design, running 700 MW, 200 MW and 400 MW over the last three quarters, or roughly 2 GW over twelve months against a portfolio that stood at 1.3 GW at the end of 2024.

The provider-concentration disclosure is the direct answer to the failure that cost the company its guidance two quarters ago: no single data center provider now delivers more than 17% of active infrastructure, across approximately 50 sites. That is a genuine structural response, not a reassurance.

All of the Q1 contracted-power additions came through long-term leases. The first self-build site is expected online later this year, which will begin converting lease obligations into owned assets with different capital and margin characteristics.

Assessment: Execution on capacity has now been strong for three consecutive quarters, including one in which the company cleared a third-party delay early. The deployment ratio improving from 20% to 29% over two quarters is the metric that converts contracted power from a liability into an earning asset, and it is moving in the right direction at an accelerating rate.

The Fit-Out Mechanics, Finally Quantified

For three quarters management has explained margin compression as "costs ahead of revenue" without saying how far ahead. This quarter the CFO put numbers on it, and they are the most useful disclosure of the call for anyone building a model.

"Upon receipt of a powered shell, we incur lease and power costs while depreciating server and other data center equipment during the fit-out process, which, on average, takes us about 1 to 2 months. During that period, we recognize costs but no revenue, causing these new deployments to run at negative contribution margins. By month 3, however, we are typically generating revenue with contribution margins normalizing in the mid-20s." — Nitin Agrawal, Chief Financial Officer

Three consequences follow. First, the drag is short: one to two months per deployment, not a structural feature of the asset. Second, it is proportional to how much capacity is energized relative to the installed base, which is why the margin trough coincided with the fastest deployment quarters. Third, and most useful, it means margin recovery is mechanically predictable from the deployment schedule rather than dependent on pricing or mix.

Revenue recognition was also clarified: revenue begins when GPUs have been deployed, tested and handed to the customer, then recognizes straight-line over the contract term. That explains why a quarter with enormous capacity additions can show modest revenue growth, and why the following quarter shows the opposite.

Assessment: This is the disclosure we have wanted since initiation, and it makes the margin story falsifiable. If second-quarter adjusted operating margin does not expand from the first-quarter trough, the explanation fails on its own terms. Management has handed the market a testable claim, which is to its credit and is also the thing we will grade next quarter.

Capital Structure: The First Investment-Grade HPC Loan

DDTL 4.0 is the structural event of the quarter and arguably of the coverage. An $8.5B delayed draw term loan, non-recourse to the parent, rated A- equivalent by Moody's, Fitch and DBRS, priced at an implied cost of less than 6% with a floating tranche at SOFR plus 225 and a fixed tranche around 5.9%, and carrying an asset-backed-style draw feature that unlocks a further $1B on stabilization of the underlying contract.

"This is the first ever investment-grade Delayed Draw Term Loan backed by HPC infrastructure, achieving an A- equivalent rating from Moody's, Fitch, and DBRS. The facility was nonrecourse to the parent. The transaction, which was well oversubscribed, was priced at a level implying a cost of less than 6%. … This is not an incremental improvement. This is a structural shift in how we expect to finance investment-grade customer contracts going forward in what is among the deepest parts of the capital markets." — Michael Intrator, Co-Founder, Chairman and CEO

The activity since quarter end reinforces it: more than $10B of additional debt and equity secured, with convertible and high-yield offerings both upsized on demand, a $1B strategic equity investment from Jane Street (a customer, which is its own signal), a fifth DDTL priced 50 basis points inside the initial marketing range and syndicated in the public loan market for the first time, and S&P moving the corporate outlook from Stable to Positive. Weighted average cost of debt is down approximately 80 basis points year to date on top of roughly 600 basis points from 2023 to 2025.

The working capital reversal deserves note as well. Receivables fell $1,049M to $2,120M, taking days sales outstanding from 185 to 92, and the net customer float (deferred revenue less receivables) rose to $5.4B. The 185-day figure that we flagged as unexplained last quarter reversed without ever being explained, which supports the benign interpretation.

Assessment: The financing story has moved from "good execution in high-yield" to "created a new asset class." An A- rated, non-recourse, HPC-backed facility at under 6% changes the arithmetic of every future investment-grade customer contract, because it lowers the discount rate applied to contracted cash flows. It does not change the fact that net debt grew 25% in a quarter to $21.5B, or that the equity sits behind all of it.

Key Topics & Management Commentary

Overall Management Tone: The most assured of the four calls in our coverage, and the most specific. Management arrived with numbers for the questions that had been left open (fit-out duration, provider concentration, backlog duration, inference share of compute, credit mix) rather than framing. The posture on the second-half ramp was flat conviction rather than argument: the guide was reaffirmed, the exit run-rate floor was raised, and the supporting evidence offered was supply-chain resilience rather than demand. Where the call was least satisfying was on why first-quarter revenue upside was not flowed through to the full year, an answer that came down to being sold out.

1. Bookings: More Than $40B in One Quarter

Backlog rose $32.6B sequentially to $99.4B on more than $40B of gross commitments, the largest quarter in company history by a wide margin. The Meta relationship expanded through multiple orders including a $21B agreement announced in early April; Anthropic joined as a customer for Claude development and deployment; Jane Street added $6B of capacity.

The composition matters as much as the size. Two verticals that did not exist as disclosed categories a year ago now carry material backlog: financial services approaching $10B, and physical AI and spatial computing above $1B. Ten customers are committed to at least $1B of spend.

Assessment: The strongest single datapoint of the quarter and the one that most supports the multi-year framework. The qualifier is that these commitments largely serve 2027, so they extend the visibility horizon rather than the 2026 outcome. That is a good problem, and it means the backlog number and the current-year guide have decoupled.

2. Pricing Rose on Every Prior Generation

The residual-value question has driven our rating since initiation. At Q3 2025 we got one contract recontracted at within 5% of its original price. At Q4 we got portfolio averages holding within 10% for H100 and rising for A100. This quarter the disclosure is unambiguous.

"Average pricing for the A100s, H100s and H200s and L40s all increased quarter-over-quarter, and we remain largely sold out for near-term capacity across our fleet. This validates what we have long believed. Demand for inference-ready compute across generations of GPUs is compounding. We expect this will be durable and accretive to our long-term margins and earnings power." — Michael Intrator, Co-Founder, Chairman and CEO

Every named generation rose sequentially, including the A100, an architecture first shipped in 2020. Management reinforced the point in Q&A: sold out in H100s, sold out in A100s, with price appreciation attributed to inference demand.

Assessment: The melting-asset bear case is now, on four consecutive observations, contradicted by the evidence. That does not make it wrong forever, because every observation has been generated inside a capacity shortage. But an investor who still holds that GPU assets decay to salvage value inside their initial contract term is arguing against the only data available. This is the pillar on which our upgrade rested and it has strengthened every quarter since.

3. Inference Crosses Half the Fleet

Management quantified something it has previously described only directionally: from power-draw profiles, materially more than 50% of compute is now serving inference rather than training.

"… we can look at the power draw and extrapolate what we believe to be, how much of the compute is being used for inference versus how much is being used for training. And when we do that, we think we are now — materially in excess of 50% of our compute is being used for inference. And that is a wonderful thing for a company like CoreWeave because it gives us tremendous confidence that the consumers of our compute are making a — are driving revenue at their entities because they are able to monetize their investment." — Michael Intrator, Co-Founder, Chairman and CEO

The analytical link to topic 2 is direct. Inference workloads are less sensitive to peak training performance and more sensitive to cost per token, so older silicon is economically rational rather than merely acceptable. A fleet majority-loaded with inference is a fleet whose prior generations have a genuine market, which is exactly what the pricing data shows.

The second-order point management drew out is about counterparty durability: customers running inference are customers earning revenue from the compute, rather than customers spending venture capital on training runs. That distinction, at scale, is what separates a durable contracted revenue base from a fragile one.

Assessment: The two most important disclosures of the quarter are this one and the pricing data, and they are the same disclosure viewed from two angles. Together they make the case that CoreWeave's asset base has a longer economic life and a more creditworthy demand profile than the market assumed a year ago.

4. The Second-Half Requirement

The full-year adjusted operating income guide of $900M–1.1B was reaffirmed. Q1 delivered $21M and Q2 is guided to $30–90M, so the first half produces roughly $81M at the midpoints. The implied second half is therefore approximately $919M, on implied revenue of roughly $7.9B against a first half of $4.6B.

Revenue must grow 71.6% from the first half to the second, and adjusted operating margin must go from 1.8% to 11.6%. Management's supporting argument was neither demand nor pricing but supply-chain readiness.

"… we have gone ahead and work through our supply chains to ensure that we are able to hit those numbers. We are in approximately 50 data centers. We have no single data center provider that delivers more than 17% of our active infrastructure. We have multiple OEMs, ODMs we are really built now to lean into a resilient supply chain that crosses all of the components that we are required to be able to deliver in order to drive that revenue." — Michael Intrator, Co-Founder, Chairman and CEO

Assessment: This is the single largest risk in the story and it is a forecasting risk rather than a demand risk. The fit-out disclosure makes the ramp mechanically plausible: capacity energized in Q1 and Q2 reaches full contribution by month three, so second-half margins inherit the benefit of first-half deployments while incurring proportionally smaller new drags as the base grows. Plausible is not the same as delivered. An investor buying this stock today is underwriting two quarters that carry 92% of the year's operating profit.

5. Component Inflation and the Pass-Through Defence

The 2026 capex guide moved to $31–35B from $30–35B, with management attributing the increase in the low end to component pricing. Given that memory and other inputs have inflated across the industry, the question is whether CoreWeave absorbs that or passes it on.

"… we are a success-based company, which means that we build our contracts to incorporate the cost of all of the components that are necessary to deliver infrastructure. And so by and large, we are insulated from the price inflation on some of the components because we include that in our pricing that we ultimately bring to clients in order to target the margins that Nitin spoke to, right, is up in the mid-20s is how we think about it on a unit basis." — Michael Intrator, Co-Founder, Chairman and CEO

On previously-signed contracts, management said pricing is set with purchase orders in hand and power costs contracted through the term, so both major inputs are effectively locked before a contract is signed. If accurate, the exposure to component inflation is confined to the gap between signing and procurement, which is short.

Assessment: Credible and consistent with the pricing power evident elsewhere in the quarter. A company selling out its capacity at rising prices across four GPU generations is not a company absorbing input inflation. The residual concern is that a higher capex number for the same revenue is, by definition, a lower return on capital, and the pass-through protects the margin percentage rather than the capital efficiency.

6. Customer Diversification Reaches a New Level

The concentration story that began this coverage at 71% of revenue from one customer has moved further. Ten customers are now committed to at least $1B of spend. Non-investment-grade AI natives and foundation labs are below 30% of backlog. Financial services approaches $10B of backlog on Jane Street's $6B expansion and the addition of Hudson River Trading. Physical AI and spatial computing crossed $1B with World Labs, Physics X and Sunday Robotics.

The one metric still absent is revenue-basis concentration, which has not been disclosed since the second quarter of 2025. Contemporaneous reporting notes that Microsoft was 62% of 2024 revenue, which gives a sense of where the series started, but there is no current figure.

Assessment: Genuine and now multi-dimensional: by customer count, by credit quality, by vertical, and by contract size. The credit-mix figure is the most valuable because a take-or-pay contract is worth exactly what its counterparty can pay. That said, three consecutive quarters without a revenue-basis concentration figure, while backlog-basis figures are volunteered, is a disclosure pattern rather than an oversight.

7. Platform Attach and the Four Product Lines

More than 90% of reserved-instance customers use at least two CoreWeave products and more than 75% use three or more. Storage, software, CPU and networking are each expected to exceed $100M of ARR by year end, which would put the non-GPU portfolio above $400M annualized.

New products shipped in the quarter span the consumption spectrum: Flex Reservations and Spot (both immediately oversubscribed), Dedicated Inference, ARENA for evaluating workloads in production-ready environments, Trust Center for enterprise compliance, and a set of cross-cloud tools (CoreWeave Interconnect with Google Cloud, SUNK Anywhere, LOTA Cross-Cloud). CoreWeave Omni goes furthest: deploying and operating the full CoreWeave stack inside customers' own data centers on customers' GPUs.

Assessment: Omni is the most strategically interesting item and the least quantified. It converts CoreWeave from a capacity provider into a software and operations provider for infrastructure it does not own, which is a structurally higher-margin, lower-capital business. Against $12–13B of 2026 revenue the entire non-GPU portfolio is still around 3%, so this is optionality rather than a driver, but it is the optionality that would change the terminal margin profile.

8. Spot and Flex Arrive, Two Years Late and Immediately Oversubscribed

At our initiation, management explained that a spot product could not be launched because every megawatt built was immediately absorbed by committed customers. Flex Reservations and Spot both launched this quarter and both were immediately oversubscribed.

The mechanism that made it possible is the NVIDIA agreement struck two quarters ago, under which reserved capacity can be interrupted and resold. That structure created the pool of interruptible capacity that a spot product requires without exposing CoreWeave to utilization risk on it.

Assessment: A two-year-old constraint resolved through contract design rather than through capacity abundance, which is the more durable solution. Immediate oversubscription tells you the demand was real and confirms that the diversification ceiling was a supply constraint, not a demand one. Watch whether spot pricing is ever disclosed; it would be the cleanest read on marginal compute pricing available anywhere in the market.

9. Self-Build Begins

All contracted power added in the quarter came through long-term leases, but the first self-build site is expected online later this year. Management framed self-build as a complement to leasing that provides operational control and long-term financial upside, alongside the NVIDIA relationship as a third avenue for accelerating the footprint.

The economics differ materially. A leased shell converts to an operating lease liability and a recurring rent expense; a self-built shell converts to owned capex, depreciation and interest. Self-build is more capital-intensive per megawatt and slower to first revenue, and it captures the developer's margin.

Assessment: Correct direction and small enough not to move the model yet. Its significance is informational as much as economic: a company that builds shells itself knows what a shell costs and how long it takes, which is precisely the knowledge gap that caused the guidance cut two quarters ago.

10. Capital Markets: A New Asset Class

Beyond DDTL 4.0 itself, the sequence since quarter end is the story: more than $10B raised across several transactions, all significantly oversubscribed, both the convertible and high-yield offerings upsized, a $1B strategic equity investment from a customer, a fifth DDTL syndicated in the public loan market at 50 basis points inside its marketing range and completing the financing of all existing OpenAI commitments, and an S&P outlook move to Positive.

Total debt and equity secured year to date exceeds $20B, against a full-year capital programme of $31–35B. Management's framing is that the capital raised is tied to customer demand rather than raised in anticipation of it.

Assessment: Financing capability has been our fourth bull pillar since initiation and it is the one that has compounded most reliably. Reaching investment grade on the asset-level facilities before reaching it at the corporate level is the right sequencing, because it prices the collateral rather than the parent. The remaining question is unchanged: cheaper debt on a larger balance is still a larger absolute interest bill, and $536M in a quarter is what that looks like.

11. The 2027 Framework Gets Contracted

At the fourth-quarter call, the exit-2027 target of more than $30B of annualized run-rate revenue rested on contracted power without customers attached. This quarter management stated that more than 75% of it is already contracted, excluding any not-yet-exercised customer renewals, and that the power capacity required is already secured.

That is a substantial de-risking of the single largest forward number the company publishes. It also explains where the $40B of Q1 bookings went: management noted the new commitments "mostly contribute towards our 2027 targets."

Assessment: The most important change to the multi-year framework since it was introduced. A run-rate target that is three-quarters contracted two years ahead is a materially different object from an ambition supported by megawatts. We now use the exit-2027 figure as a sanity check on terminal value, though we still value on exit-2026 because the intervening net debt build is the variable that determines what reaches the equity.

Guidance & Outlook

Why the Q1 beat did not lift the full year. Management's answer was that the company is essentially sold out of 2026 capacity, so upside in one quarter is a timing shift rather than incremental volume. The place it does show up is the exit-2026 run-rate floor, which was raised, and the 2027 framework, where more than 75% is now contracted. That is internally consistent, and it also means that for the balance of 2026 revenue is a function of deployment schedules rather than of demand.

The Q2 interest line. Guided at $650–730M against adjusted operating income of $30–90M. Even at the most favourable pairing, interest is 7.2 times operating profit; at the least favourable, 24.3 times. Interest expense will be roughly 26–30% of second-quarter revenue.

Guidance style, one year in. Four quarters of public guidance now show a consistent pattern that a modeller can use. Quarterly revenue guidance has been met or beaten every time, though the full-year revenue guide was cut once, in November 2025. Profit guidance was cut once (Q3 2025), missed once (Q4 2025), and met once (Q1 2026), with the miss and the cut both traceable to deployment timing rather than to pricing or demand. Capex guidance has been wrong in both directions. Multi-year run-rate guidance, introduced last quarter, has already been raised once and de-risked once. The company is more reliable on what it will sell than on when it will book the cost of selling it.

Analyst Q&A Highlights

How Component Inflation Passes Through, and What Changed With NVIDIA

The opening question tied the capex increase directly to the after-hours reaction and asked for the mechanism by which higher input costs reach the P&L, alongside a request to characterize what specifically changed in the NVIDIA relationship.

Q: "… maybe starting on the CapEx side of the equation and the higher component pricing. I think that's something that's weighing on the stock a little bit after hours. Can you explain to us how the higher component pricing works through the contract and how that ultimately affects your profitability? And then — and maybe if we could talk about the NVIDIA relationship and the expansion of that relationship, what fundamentally changed in that relationship this quarter?"
— Keith Weiss, Morgan Stanley

A: "… by and large, we are insulated from the price inflation on some of the components because we include that in our pricing that we ultimately bring to clients in order to target the margins that Nitin spoke to, right, is up in the mid-20s is how we think about it on a unit basis. … the two most important were the qualification of our software solution as a reference architecture for them. And that was an incredible validation of the quality of the software solution that we deliver to market."
— Michael Intrator, Co-Founder, Chairman and CEO

Assessment: The pass-through answer is the right one and consistent with pricing rising across four GPU generations. On NVIDIA, note what was not emphasized: management downplayed the 5 GW element of the partnership, pointing out that CoreWeave secured 2 GW on its own over twelve months and framing the NVIDIA capacity as opportunistic acceleration rather than a dependency. That is the right posture for a company already accused of circularity with its principal supplier.

What Underpins $919M of Second-Half Operating Profit

The most consequential question of the call put the second-half arithmetic to management directly, and drew both the reaffirmation and the specific evidence management is relying on.

Q: "Nitin, $81 million in EBITDA in the front half, but you're guiding to $919 million in the second half on the bottom line. What's giving you conviction in the bottom line build in the back half of the year?"
— Brent Thill, Jefferies

A: "… everything in our business is defined by active power and capacity ramp schedules, which are not necessarily linear. … We are reaffirming our full year guidance, including full year revenue, full year adjusted operating income and that we will exit the year with low double-digit adjusted operating income margin. We are reaffirming our active power, and we are raising the floor of our 2026 exit ARR guidance. All of these are indicators in the conviction that we have on our execution plan for the remainder of the year. What you will see us is executing through this plan and expect to see our adjusted operating income accelerate faster than revenue growth in the second half of the year."
— Nitin Agrawal, Chief Financial Officer

Assessment: The answer stacks reaffirmations rather than explaining the bridge, which is less than the question deserved. What makes the ramp credible is the fit-out disclosure elsewhere on the call, not this answer: a one-to-two-month cost lead followed by mid-20s contribution margins by month three means first-half deployments carry second-half margins. The CEO's follow-on about supply-chain resilience across roughly 50 data centers is the operational substance. This exchange defines the single largest gradeable claim in the story.

Whether Inflation Damages the Margins on Contracts Already Signed

A precise question separated new contracts from booked ones: for business signed before component costs rose, is the margin profile now worse than underwritten, and is there recourse?

Q: "To the extent that you have a business that you booked previously, and it's not fully up and running and now the cost of components has risen. I would think the cost of energy has risen, although maybe you have some of that locked in. Is the margin profile of those previously signed contracts slightly different than originally contemplated? Or do you have some recourse somehow to flow through increased pricing or maybe restructure some of those contracts?"
— Mark Murphy, JPMorgan

A: "… when we are pricing our deals, we are pricing them with purchase orders in hand for the infrastructure that is required to be able to deliver on it. We understand what the power cost is going to be today, next year and out through the term of the contract because that is contractually delivered to us. And so we've done a really good job of understanding what the components and electricity costs are going to be, we have it structured so that it is effectively passed through when we enter into the contract once again, to ensure that we're able to hit our targeted margins on a unit basis."
— Michael Intrator, Co-Founder, Chairman and CEO

Assessment: The strongest answer of the call and an important one for the backlog's value. Purchase orders placed at signing and power contracted for the term means the $99.4B backlog is largely insulated from input inflation, which is not true of every operator in this market. If accurate, it converts component inflation from a margin risk into a capital-efficiency drag, which is a materially smaller problem.

Why the First-Quarter Beat Did Not Reach the Full-Year Guide

A follow-up pressed on the mechanical consequence of reaffirming an annual number after beating a quarter: the implied rest-of-year forecast falls.

Q: "What holds you back from just passing through that revenue upside into the full year revenue guidance because, I mean, in some sense, technically will reduce our rest of year revenue forecast a little. Is there some other effect?"
— Mark Murphy, JPMorgan

A: "We pretty much remain sold out for our 2026 capacity. So that is continuing to be true for us at this point of time. Where you would see this kind of inflect for us is around two vectors. We've raised the guidance for the floor of exit ARR. … At the same point of time, where it's also inflecting is showing up in 75% of our 2027 ARR guidance that we provided, which is $30-plus billion already booked at this point of time, excluding any potential renewals, greater than 75% of that."
— Nitin Agrawal, Chief Financial Officer

Assessment: Coherent and, for a modeller, the key to reading this company correctly. When capacity is fully allocated, in-year revenue is a deployment-schedule variable and upside surfaces in exit run rate and in the following year rather than in the current-year guide. Investors modelling CoreWeave on quarterly revenue beats are watching the wrong series; the exit run rate and the contracted share of the following year are the ones that carry information.

What Fixes the Declining Gross Margin

A structural question about margin trajectory over five quarters produced the fit-out explanation in its most complete form, plus the clearest articulation of the scaling-variance argument.

Q: "… gross margin has been going down throughout. I look at the last 5 quarters. We started at 78%, now 68%. And I'm trying to understand the operating structure, how it changes over the next 2 years, let's say, meaning what are the drivers for gross margin improvement."
— Tal Liani, Bank of America

A: "The margin dynamic in our business is predominantly timing based, not economic. … That process takes us about 1 to 2 months. During that period, we are recognizing cost but no revenue. That is what costs us new deployments to be running contribution margin during the — negative during the deployment phase. … By month 3 in these deployments, we are typically generating revenue and the contribution margins stabilize normalizing to a ramped contract in about mid-20s."
— Nitin Agrawal, Chief Financial Officer

Assessment: The most useful answer of the coverage to date. It replaces a qualitative claim with a testable mechanism and a duration. The CEO's supporting illustration (adding 300 MW onto a 50 MW base versus 50 MW onto a 2,000 MW base) makes the scaling argument concrete. The implication is that margin recovery is a function of the ratio of new to installed capacity, which is knowable in advance from the deployment schedule.

How Much of the Fleet Serves Inference, and What That Does to Old Silicon

A question linking the inference transition to the economics of prior-generation GPUs drew both the quantification of workload mix and the clearest statement yet on the useful life of the asset base.

Q: "I'd love to understand how fast do you think inferencing as a share of your consumed power is growing right now in your installed base. And as that keeps going, what does that really imply for the utilization and contracting economics for the H100s, A100s over the next 12 to 18 months to you."
— Amit Daryanani, Evercore

A: "… we think we are now — materially in excess of 50% of our compute is being used for inference. … We are sold out in our H100s. We are sold out in our A100s. We are seeing price appreciation as more inference is coming in and making demands upon that compute work loads in order to be able to deliver to their clients. And we think that is an incredibly bullish signal to the space at large."
— Michael Intrator, Co-Founder, Chairman and CEO

Assessment: The two facts that most support the terminal value of this business, delivered in one answer. Sold out on two prior generations with rising prices, in a fleet where inference is now the majority workload, is direct evidence that economic life exceeds the depreciation schedule. The described customer behaviour (buy leading edge for training, cascade it down to inference, buy the next generation for training) is a coherent mechanism rather than an assertion, and it explains why nine of the ten largest customers hold multiple contracts.

Allocating, Not Selling, the Remaining Capacity

The final exchange tested the language management uses about capacity and, in doing so, surfaced what the real constraint now is.

Q: "… is the word allocate rather than sell accurate; and b, is the amount to allocate or sell, if you prefer that word, increasing or decreasing relative to a quarter ago."
— Nehal Chokshi, Northland Capital Markets

A: "I think the word allocate is probably accurate. It's unusual to be in a business where the demand for your product is so high that you get to really be thoughtful about which clients you want to bring on to your infrastructure … the truth of the matter is the limiting factor isn't just power, it's labor, it's memory, it's storage, it's our ability to bring up infrastructure."
— Michael Intrator, Co-Founder, Chairman and CEO

Assessment: The constraint has migrated again. It was GPUs, then powered shells, then powered shells plus construction throughput, and is now a multi-input problem that includes labour, memory and storage. Each migration has been toward inputs that are more elastic over time, which supports the view that CoreWeave's advantage is a multi-year head start rather than a permanent scarcity rent. The ability to allocate on counterparty quality rather than price is worth real money in the interim, and it is visible in the credit-mix improvement.

What They're NOT Saying

1. Revenue-basis customer concentration. Three consecutive quarters without it. Backlog-basis credit mix is volunteered in detail; the revenue figure was last disclosed at 71% for Q2 2025.
2. Utilization. Five quarters as a public company, still no disclosure. Management can infer training-versus-inference mix from power draw, which means it can measure utilization precisely.
3. Spot and Flex pricing. Both products were "immediately oversubscribed" and no pricing was given. Spot clearing prices would be the single best read on marginal compute pricing in the market.
4. Free cash flow. Operating cash flow of $2,984M against $7,695M of property and equipment purchases in one quarter. Neither figure was set against the other by management or by any analyst.
5. The size of CoreWeave Omni's opportunity. A product that deploys the CoreWeave stack in customers' own data centers on customers' GPUs would change the margin structure of the company, and was described in two sentences with no sizing, no pricing model and no pipeline.
6. Depreciation life. D&A of $1,147M is now 55% of revenue and grew 159% year over year. On a call that produced the strongest evidence yet that assets outlive their schedules, the schedule itself went unmentioned.
7. What the 25% of the 2027 run rate that is not contracted depends on. More than 75% contracted is a strong disclosure; the residual quarter of a $30B target is roughly $7.5B of annualized revenue with no counterparty named.
8. Any framing of 2028 capital intensity. The 8 GW-by-2030 ambition implies capital deployment beyond 2027 at a scale that has never been discussed, in a business where capex has run at roughly 2.5 times revenue.

Market Reaction

• Pre-print setup: close of $128.84 on May 7, up 79.9% year to date from $71.61 at the end of 2025, up 51.1% over the trailing 30 days from $85.24 on April 7, and up 140.4% over the trailing twelve months from $53.60. The stock entered the print 30% below its 52-week closing high of $183.58 but after one of the sharpest one-month rallies in its short public history.
• After-hours move: shares fell as much as 10% in extended trading on May 7.
• Next-day session: opened at $118.46, an 8.1% gap down; traded a $110.55 to $123.75 range; closed at $114.15, down 11.4% and $14.69 on the day.
• Volume: 48.5M shares against a 30-day average of 30.6M, a 1.6 times multiple, the lowest reaction-day volume multiple of the four prints we have covered.
• Benchmark: the S&P 500 rose 0.8% on the session and was up 7.2% year to date entering the print, so the move was entirely company-specific.

For the first time in this coverage the after-hours move and the next-session close were consistent, at roughly 10% and 11.4%. The previous two quarters saw the reaction roughly double overnight as investors worked through arithmetic the headlines did not contain. That the two matched this time suggests the market has learned how to read this company's disclosures in real time, which is itself a maturation signal.

The proximate cause was the second-quarter guide. A $2.53B midpoint against a $2.69B consensus is a 6.1% shortfall. For a stock that had risen 51% in thirty days, a guide below the Street was always going to be sold regardless of what accompanied it.

The capex revision compounded it. Raising the low end of the 2026 capex range to $31B from $30B is a $1B change to a $33B midpoint, which is immaterial in isolation. In a market already anxious about AI capital intensity, a capex number moving up in the same breath as a revenue guide moving down is read as a margin story regardless of the stated cause.

The setup left no room. Entering a print up 80% year to date against an index up 7%, with the options market pricing an 18.7% implied move, is a setup in which good news is required rather than rewarded. The realized move landed comfortably inside what was priced.

Our read is that the reaction was proportionate this time. The near-term guide genuinely was soft, and the structural disclosures (bookings, pricing across generations, the investment-grade facility, the contracted 2027 framework) were substantially already in the price after a 51% one-month move. That is the difference between this quarter and the last two, and it is the reason our rating changes rather than our thesis.

Street Perspective

Debate: Is the Second-Half Ramp Underwritable or an Act of Faith?

Bull view: The bull case on the Street holds that the ramp is mechanical rather than aspirational. The fit-out disclosure gives a one-to-two-month cost lead and mid-20s contribution margins by month three, so capacity energized in the first half carries second-half margins by arithmetic. Capacity is sold out, contracts are signed, components are procured, and management reaffirmed the full year while raising the exit run-rate floor.

Bear view: The bear camp notes that 92% of the year's adjusted operating income sits in two quarters that have not happened, that this company has missed its profit guidance twice in four quarters for deployment-timing reasons, and that revenue must grow 71.6% half over half for the arithmetic to work. Reaffirming a full-year number in May is a low-cost commitment.

Our take: The mechanism is sound and the concentration of risk is real. We model the second half slightly below the guide midpoint, which is why our FY2026 adjusted operating income estimate sits at the lower half of the range. The decisive datapoint arrives in August: if second-quarter adjusted operating margin expands sequentially as guided, the framework holds and the second half becomes a scheduling question.

Debate: Does the Investment-Grade Facility Change the Equity Story or Only the Debt Story?

Bull view: A growing view holds that DDTL 4.0 is a step change rather than an increment. An A- rated, non-recourse, HPC-backed facility at under 6% establishes that the market will price CoreWeave's contracted cash flows as investment grade, which lowers the discount rate on every future contract with an investment-grade counterparty and compresses the cost of the entire capital stack over time.

Bear view: The skeptical framing is that cheaper debt on a rapidly growing balance is still more absolute interest, that $536M of quarterly interest against $21M of operating profit is the only number that matters, and that an investment-grade rating on the collateral tells you the lenders are safe rather than that the equity is.

Our take: Both are right and they are complementary rather than opposed. Investment-grade pricing on the assets raises enterprise value; it does not change where the equity sits in the stack. The reason it matters to equity holders is second-order but real: a lower cost of capital raises the return on each incremental megawatt, which is the spread that eventually accrues to equity. That is a multi-year effect and it does not help the 2026 income statement at all.

Debate: Is the Stock Cheap on 2027 or Expensive on 2026?

Bull view: The bull framing is that at roughly 2.7 times an exit-2027 run rate that is more than 75% contracted, with a business compounding revenue at over 100%, the stock is inexpensive on any forward measure and the 2026 margin trough is a distraction. Buy the year of investment, own the years of harvest.

Bear view: The bear framing is that enterprise value per unit of run-rate revenue is the wrong lens for a company that will add more net debt in 2026 than it will add revenue. Every dollar of enterprise value created by growth is matched by roughly a dollar of claim ahead of the equity, so the per-share outcome depends on the spread rather than on the growth rate, and that spread is not yet demonstrable.

Our take: This is why we downgrade. On our base case, which uses the exit-2026 run rate and an exit-2026 net debt assumption of roughly $37B, fair value brackets the current price rather than sitting above it. The bull framing requires valuing on 2027 while ignoring the 2027 balance sheet, and the balance sheet is precisely what the leverage makes non-ignorable. We are not bearish; we simply no longer find the asymmetry that justified an Outperform three months ago.

Model Update & Valuation Framework

Valuation base unchanged in method. We continue to value on management's exit-2026 annualized run-rate revenue, because reported annual revenue systematically understates earning power when capacity energizes continuously through the year, and we continue to state the net debt assumption explicitly because it does most of the work.

At the May 8 close of $114.15 and 527M shares, equity value is approximately $60.2B. Adding total debt of $24.86B and subtracting cash, restricted cash and marketable securities of $3.34B gives an enterprise value of roughly $81.7B, which is 6.5 times the midpoint of the FY2026 revenue guide, 4.4 times the midpoint of the exit-2026 run rate, and 2.7 times the exit-2027 target.

Twelve-month price target: $123 (range $105–140). The target rises 24% from $99 because the run-rate floor was raised, the net debt assumption tightened on equity issuance and stronger operating cash flow, and the multiple deserves to be higher after an A- rated financing and a 2027 framework that is more than three-quarters contracted. The rating falls because the share price rose 43% over the same period. A $123 target against $114.15 implies roughly 8% over twelve months, which is a market-like return and therefore a Hold under our framework.

What differentiates the scenarios. The bear case is not a demand failure; it is a second half in which the margin ramp arrives late, the exit run rate lands near $16B, and the market applies an infrastructure multiple to an equity behind $40B of net debt. The bull case requires no acceleration beyond guidance, only that the run rate reaches the top of the guided range and that the investment-grade financing plus the contracted 2027 framework earn a growth multiple. As at every prior report, the tails are wide and turn on the same variable: whether contracted revenue outlives the debt that funded the assets producing it.

Thesis Scorecard Post-Earnings

Grading Last Quarter's Commitments

Overall: Eight of ten commitments met or beaten, with the two incomplete items (the margin ramp and the licensing revenue) both forward-dated rather than failed. Across four quarters of coverage the pattern has been consistent and it held again: CoreWeave delivers on capacity, bookings, financing cost and asset-value evidence, and it defers profitability. This quarter it also delivered on the near-term profit guidance it had set, which it had not done before.

Action: Downgrade to Hold from Outperform with a $123 twelve-month target. This is a valuation decision. None of the four downgrade conditions we set in February was triggered: no second provider delay, no contrary residual-value evidence, no reduction to the exit run rate (the floor was raised), and the capex increase came with a run-rate increase attached. The stock simply rose 43% in ten weeks against an index up roughly 6%, and our base case no longer sits above it.

What would take us back to Outperform: a second-quarter adjusted operating margin that expands as guided, confirming the second-half framework, particularly if paired with a raise to the full-year profit guide; a meaningful pullback that restores the asymmetry, with the fundamental trajectory intact; disclosure of spot or Flex clearing prices that confirms marginal compute pricing is holding; or a first quantified contribution from CoreWeave Omni or the wider cloud-stack licensing, which is currently excluded from guidance and from our model.

What would take us to Underperform: a second-quarter adjusted operating margin that fails to expand from the first-quarter trough, which would break the mechanism management has now put on the record; a counterparty credit event among the ten customers committed to $1B or more; a debt raise at materially wider spreads or a reversal of the S&P outlook, either of which would signal the credit market has repriced the collateral; or the first quarter in which prior-generation GPU pricing declines.