Cantor Fitzgerald's Brett Knoblauch went into CoreWeave's second quarter expecting roughly $40 billion of new backlog, matching the pace the company set in the first quarter. What printed was $104.2 billion of revenue backlog at June 30, up from the first quarter's $99.4 billion at March 31, which works out to net additions of $4.8 billion and roughly $7.4 billion gross once you add back the $2.58 billion of revenue recognized during the quarter. The backlog-conversion case failed on the exact number its bulls had built the thesis around. CRWV then closed Wednesday, August 12 at $107.73, up 19.28% on the session.
That pairing is rare enough to be worth taking apart properly. A stock does not usually re-rate by nearly a fifth on the day it undershoots the single metric everyone agreed to watch, and understanding why this one did tells you something durable about how AI infrastructure names are being priced right now.
The Number Everyone Was Watching Came In Small
Backlog is the metric that carries the entire AI data center thesis, because it is the only line that claims to convert future demand into a contracted, countable figure. CoreWeave's grew 246% year over year, from $30.1 billion to $104.2 billion, which sounds like an unambiguous win until you look at the sequential step. CoreWeave's first quarter reportput backlog at $99.4 billion as of March 31, and the company itself called that its strongest bookings quarter ever.
Knoblauch's model had the second quarter reaching roughly $131 billion against a broader Wall Street figure near $104.4 billion. The Street number was almost exactly right. The Cantor number was off by about $27 billion, and the analyst who had been loudest about backlog acceleration was the one whose estimate the print embarrassed.
So the sequential math is what it is. Add $4.8 billion net, gross it up by the $2.58 billion the company burned off the balance as recognized revenue, and you get about $7.4 billion of new contracted work signed inside the three months ending June 30. Against a $40 billion expectation, that is the thesis metric arriving at roughly a fifth of what the most bullish covering analyst wanted. Our own preview from earlier in August framed the backlog math as the risk, and on the printed quarter the risk was real.
What the Printed Quarter Actually Delivered
Everything below the backlog line was fine, and in places genuinely strong. This is where the accounting basis matters, because CoreWeave reports two different loss figures and pairing the wrong ones manufactures a miss that did not happen.
|
Metric
|
Q2 2026 print
|
Expectation going in
|
|
Revenue
|
$2.58 billion (+112% y/y)
|
~$2.56 billion
|
|
Adjusted loss per share
|
$1.03
|
$1.20 (LSEG consensus)
|
|
Adjusted operating income
|
$128 million
|
~$66 million
|
|
Adjusted EBITDA
|
$1.51 billion, 59% margin
|
not separately polled
|
|
Revenue backlog at June 30
|
$104.2 billion
|
~$104.4 billion Street, ~$131 billion Cantor
|
On an adjusted basis CoreWeave lost $1.03 per share against an adjusted consensus of $1.20, a beat by 17 cents. The GAAP net loss of $626 million belongs in a different conversation, because it carries interest expense on a debt load that has grown fast to fund buildout, and setting a GAAP number against an adjusted estimate is the most common way traders talk themselves into a miss the tape never priced.
Capital intensity is the honest counterweight here. Second-quarter capex hit $9.4 billion, the largest quarterly figure in company history, and first-half spending reached $16.1 billion against $4.8 billion in the same period of 2025.
The $25 Billion That Was Not in the Backlog
Then comes the turn. CoreWeave's second quarter release states plainly that the $104.2 billion figure "does not include more than $25 billion of net new customer commitments added in early Q3."
Read that against the $7.4 billion gross figure from the quarter itself. In roughly six weeks after June 30, CoreWeave signed more than three times what it signed in the entire second quarter. The balance-sheet metric measured a period that had already stopped being representative by the time it was published, and the market simply refused to trade a stale snapshot when a live signing rate was sitting right next to it in the same document.
The physical build supports that reading. Active power reached 1.5 gigawatts across 51 data centers at quarter end, contracted power hit 3.7 gigawatts and then 4.2 gigawatts as of August 11, and management raised the year-end active power target to more than 1.85 gigawatts. Contracts longer than 48 months went from 10% to 21% of the mix, which is the duration story underneath the headline number.
Guidance did the rest. The third quarter is guided to $3.45 billion to $3.6 billion, full-year 2026 revenue to $12.4 billion to $13.2 billion, and exit annual recurring revenue to $18.5 billion to $19.5 billion.
The Options Market Underpriced It and the Move Held
Ahead of the print, the at-the-money straddle on the August 14 expiry implied a 15.5% move, priced off an $85 strike where the call was $6.85 and the put $6.35. That sat slightly below CoreWeave's average absolute post-earnings move of 16.76% across the prior four quarters, so the options market was pricing a normal CRWV reaction and got a 19.28% one. Straddle buyers were paid, which happens often enough after a large-cap AI print that it barely counts as a surprise on its own.
What does count is what happened next. Session closes verified against stockanalysis.com's CRWV history run $88.19 on Monday, August 10 for a 2.74% loss, $90.32 on Tuesday, August 11 for a 2.42% gain, then $107.73 on Wednesday, August 12. Thursday, August 13 closed at $106.29 for a 1.34% loss and Friday, August 14 at $105.26 for another 0.97%, so two sessions of profit-taking gave back a combined 2.3% of a 19.28% gain.
A gap that fills is a liquidity event. A gap that holds through two sessions of natural selling is a re-rating, and the follow-through separates the two more reliably than the size of the initial move ever does.
Cantor supplied the confirmation from the side you would least expect. On Thursday, August 13, Knoblauch raised his price target from $167 to $176 and held Overweight, one session after the quarter had contradicted his central estimate. SanDisk delivered a structurally similar re-rating the same week off its investor day, covered separately in our profile of its finance chief.
What This Changes About How to Read a Backlog Number
A backlog is a point-in-time measurement with a cutoff date, and in a market where individual deals run to tens of billions, the gap between that cutoff and the reporting date can hold more economic news than the reported period itself. Most traders who got the direction wrong treated the quarterly disclosure as the full information set. Anyone holding a short into that print on backlog-deceleration logic was reading a correct number about a window that had already closed.
But the same logic cuts the other way, and it should be said plainly. Forward commitments are commitments, not revenue, and they depend on counterparties funding them across multi-year horizons while CoreWeave keeps spending at a $9 billion quarterly clip to serve them. Customer concentration remains high, and the debt financing the buildout is what turns adjusted profitability into a GAAP loss. Traders who were right about the re-rating are not automatically right about the next twelve months.
That tension runs across the AI infrastructure trade, and it is the mirror image of what hit Tesla after a record-revenue quarter. The same separation between the print and the reaction shows up in Nvidia's 2026 setup, in our Micron earnings-outlook breakdown and in the Marvell and Broadcom AI chip comparison.
Frequently Asked Questions
Did CoreWeave beat or miss on earnings?
It beat on the reported financials and undershot on the metric the bulls cared about most. Revenue of $2.58 billion topped the roughly $2.56 billion consensus and the adjusted loss of $1.03 per share came in ahead of the $1.20 adjusted consensus, while sequential backlog additions of about $7.4 billion gross fell far short of the roughly $40 billion Cantor had modeled.
Why did CRWV rise 19.28% if the backlog number disappointed?
Because the release disclosed more than $25 billion of net new customer commitments signed in early Q3 that were excluded from the $104.2 billion. The market repriced off the forward signing rate rather than the June 30 snapshot.
Is CoreWeave profitable?
Not on a GAAP basis, where the company posted a $626 million net loss in the second quarter. It generated $1.51 billion of adjusted EBITDA at a 59% margin over the same period, and the difference between those two figures is dominated by interest expense on the debt funding its data center buildout.
What does an options-implied move actually tell you?
It tells you what the market is charging for earnings risk, not what direction the stock will go. CoreWeave's 15.5% implied move against a 19.28% realized move meant option sellers were underpaid, and the useful signal was the two sessions afterward, when the stock retained most of the gain.
Who is the CoreWeave CEO?
Michael Intrator co-founded the company and has run it through the transition from crypto mining infrastructure to AI cloud. Our full profile covers how Intrator rode the AI data center boom, and our CRWV index-inclusion explainertraces how the stock reached index-relevant scale.
Bottom Line
The trade from here rests on one question. Does the early-Q3 signing rate that re-rated the stock show up as a step change in the September-quarter backlog disclosure, or was the $25 billion a pull-forward that leaves the next print looking thin again. That number arrives with third-quarter results and it is the single line worth an alert.
Structurally, $105.26 from Friday, August 14 is the reference, and holding above the pre-print $90.32 base keeps the re-rating intact regardless of chop in between. A move back through the low $90s would say the market decided the forward commitments were worth less than it briefly paid for them. Carry the general version into the next AI infrastructure print. When a company discloses activity that falls outside the reporting window, that disclosure is the news and the reported metric is history.
This article is for informational purposes only and does not constitute financial or investment advice. Cryptocurrency trading involves substantial risk. Always conduct your own research before making trading decisions.
