So here’s what caught our eye this week. ARK Invest quietly sold nearly 100,000 shares of CoreWeave (NASDAQ:CRWV) on May 4. That’s roughly $12 million out the door. Two days before earnings. The stock has already surged 75% year-to-date. And May 7 is when the numbers land. The big question on every desk right now: is this a warning shot — or just classic ARK profit-taking ahead of a binary event? Here are four things you need to know before Thursday’s print.
The Cathie Wood Signal: Smart Money Move or Noise?
On May 4, ARK Invest sold 99,692 CoreWeave shares across multiple ETFs. The stock had ripped 56% in the prior month alone. This kind of trim ahead of earnings is classic ARK. Wood tends to build early in a growth story. Then, when valuations stretch and risk-reward narrows, she shaves. This doesn’t read as a bearish call on the business. It reads like disciplined portfolio management ahead of a high-volatility event.
Context matters here. ARK trimmed AMD on the same day. It also sold Teradyne. Meanwhile, it bought Shopify and L3Harris — very different risk profiles. This looks like a broader rebalancing of AI-linked high-flyers rather than a specific view on CoreWeave’s Q1 numbers. Wood isn’t exiting AI. She’s rotating within it. Trimming names that have already moved big and picking up others with more perceived runway. Whether that’s prescient or simply opportunistic is the debate.
Options markets are pricing an 18.71% move in either direction after Thursday’s print. That is significant implied volatility. In that environment, locking in gains after a 56% one-month run is rational portfolio management. The signal here is genuinely ambiguous. Read it carefully before reading too much into it. The real verdict lands Thursday evening.
The Backlog Bombshell: $95 Billion & Counting
CoreWeave exited Q4 2025 with a contracted revenue backlog of $66.8 billion. That number grew more than 4x in a single year. But the real headline coming into Thursday is what happened in April. The company announced a $21 billion deal with Meta and a $6 billion deal with Jane Street. Jefferies analyst Brent Thill now expects CoreWeave’s RPO to exceed $95 billion when Q1 numbers drop. That would be a staggering sequential jump from Q4.
The structure of this backlog matters. These are take-or-pay contracts. Customers pay whether they use the compute or not. At the Morgan Stanley TMT Conference in March, co-founder Brannin McBee noted the weighted average contract length has extended to five years, up from three years just 24 months ago. Customers aren’t just buying compute. They’re locking it in for longer — because they believe this infrastructure will remain essential to their business roadmap.
The demand profile has also diversified meaningfully. It started with AI labs. Then hyperscalers. Now enterprises — Cognition, Cursor, Mercado Libre, Midjourney, Runway — are signing reserved instance contracts. CoreWeave added roughly twice as many new reserved instance customers in Q4 as in any prior quarter. Demand is broadening, not narrowing. Thursday’s backlog update will likely be the single most-watched number in the entire print.
The Margin Story: Growth at What Cost?
This is the part that rattled investors after Q4. CoreWeave guided for Q1 2026 adjusted operating income of just $0–$40 million on $1.9–2 billion in revenue. Razor thin. CFO Nitin Agrawal explicitly called Q1 the “trough” of the annual margin trajectory. The structural reason is timing. When new data center capacity comes online, depreciation and lease costs kick in immediately. Revenue, however, ramps over the following months. That gap punishes near-term margins hard.
The scale of the ramp is what makes this so pronounced. CoreWeave added roughly 260 megawatts of active power in Q4 alone. That’s about one-third of its entire installed base in a single quarter. In 2026, it targets over 1.7 gigawatts of active power — double 2025 levels. With that expansion velocity, more capacity costs will run ahead of revenue throughout the year. The key question Thursday: did Q1 margin land within guidance? And is sequential expansion from Q2 onward still on track?
The longer-term picture is more constructive. Fully stabilized contracts generate contribution margins in the mid-20s. EBITDA margins on mature contracts run near 70%. Management targets 25–30% adjusted operating margins over the long term. Storage already crossed $100 million in ARR with an 80% attach rate among large customers. Software licensing adds another upside layer. But these are longer-dated drivers. Near-term, margins stay messy — and the market knows it.
The NVIDIA Partnership & The Software Layer Bet
NVIDIA and CoreWeave are not simply supplier and customer. In January, NVIDIA invested $2 billion in CoreWeave. That lifted its stake by nearly 50%, to 47.3 million shares. NVIDIA is now CoreWeave’s second-largest shareholder. As part of the expanded collaboration, NVIDIA plans to test and validate CoreWeave’s proprietary cloud stack — including software tools SUNK and Mission Control — toward incorporating them into NVIDIA’s global reference architecture for cloud, enterprise, and sovereign customers.
This software angle is important. It is not included in CoreWeave’s 2026 guidance. Any revenue from licensing the cloud stack to third parties is pure upside to current estimates. CoreWeave is already seeing select customers license SUNK as their default cluster management platform across multi-cloud environments. NVIDIA’s validation accelerates the go-to-market. It also unlocks sovereign AI customers — government entities building national AI infrastructure — as an entirely new demand category that didn’t previously exist for CoreWeave.
On hardware, CoreWeave was first to market with H100s and GB200s. It expects to be among the first to bring NVIDIA’s Rubin GPU platform to market in H2 2026. NVIDIA’s Vera CPU and BlueField storage are also being added to the product suite. These products directly support the agentic AI workloads enterprise customers are rapidly building. Agentic workloads demand far more from CPU, memory, and storage than traditional training runs. Each new product category is a new attach point — and a new revenue stream.
Final Thoughts: A Lot Of Growth Already In The Price
Heading into Thursday’s print, CoreWeave is a study in contrasts. The business fundamentals — backlog depth, demand breadth, contract duration, and infrastructure execution — are genuinely impressive. The financial model, however, is front-loading enormous cost today to capture tomorrow’s revenue. And the valuation already reflects a considerable amount of optimism.
On a trailing basis, as of May 5, CoreWeave trades at 18.73x LTM EV/Revenue and 39.91x LTM EV/EBITDA. Trailing price-to-sales sits at 13.53x. Price-to-book is 19.40x. Price-to-tangible book reaches 32.37x. On earnings or free cash flow, trailing multiples are not meaningful in the traditional sense — LTM diluted EPS was -$2.81 and LTM levered free cash flow was approximately -$4.6 billion. This is emphatically not a value stock by conventional measures.
The forward multiples compress considerably given the revenue ramp ahead. NTM EV/Revenue sits at 7.69x and NTM EV/EBITDA at 13.02x as of the same date — more digestible for a business growing revenue at approximately 140% year-over-year. That said, NTM levered free cash flow remains deeply negative at roughly -$24.9 billion, reflecting the sheer intensity of the CapEx build program.
The stock is up 75% year-to-date. Cathie Wood’s trim is not a verdict. It is one data point among many. What to watch Thursday: the RPO update, the Q1 margin print relative to guidance, and any color on the NVIDIA software licensing timeline. Those three things will likely determine which direction the implied 18.71% move goes — and whether the market decides the growth story is worth what it’s already pricing in.
Disclaimer: We do not hold any positions in the above stock(s). Read our full disclaimer here.
