CoreWeave Q3 2026 Earnings: Revenue Jumps 168% to $3.66 Billion

CoreWeave posted third-quarter 2026 revenue of $3.66 billion, up 168% from $1.37 billion a year earlier, and told investors the quarter's revenue grew by more than $1 billion sequentially. The AI cloud provider also raised its full-year 2026 revenue guidance to $13.0 billion-$13.3 billion and said its revenue backlog had reached roughly $132.5 billion as of September 30, even as its GAAP net loss widened to $709 million on higher interest expense and depreciation.

CoreWeave's Q3 2026 numbers, in one place

CoreWeave (Nasdaq: CRWV) reported the results on November 10, 2026, calling it a "breakout quarter." Revenue, profit and margin all improved from both the prior-year quarter and the growth the company had guided to in August. The company said adjusted operating margin reached double digits "a quarter ahead of plan."

CoreWeave was founded in 2017 and completed its initial public offering on Nasdaq in March 2025. It describes itself as "The Essential Cloud for AI," a cloud provider built specifically around renting out large clusters of Nvidia GPUs and the networking, power and software needed to keep them running at scale, rather than offering the broader general-purpose cloud computing menu of Amazon Web Services, Microsoft Azure or Google Cloud. In its own words, the company says it is trusted by 9 of the 10 leading foundation model providers, which is why its quarterly results are often read as a proxy for how much AI infrastructure demand is actually translating into paid contracts, rather than just announcements.

Why the net loss widened even as margins improved

CoreWeave's GAAP net loss grew to $709 million from $110 million a year earlier, and the net loss margin widened to 19% of revenue from 8%. The company's own SEC filing and reconciliation tables show the gap is driven mostly by two line items that scale with its data center buildout: interest expense, net, which rose to $846 million from $310 million, and depreciation and amortization, which rose to $1.89 billion from $630 million. Both are consequences of financing and depreciating a much larger fleet of GPU infrastructure, not a sign that the underlying cloud business is losing money on each dollar of revenue — adjusted EBITDA margin actually held steady at 62%, barely changed from 61% a year ago.

CoreWeave's cash flow statement tells a similar story: the company generated $1.98 billion in cash from operating activities during the quarter, up from $1.69 billion a year earlier, so the core cloud business is a net cash generator even while it posts a GAAP accounting loss. The gap is on the investing side, where CoreWeave spent $9.44 billion on property, equipment and capitalized internal-use software in the quarter alone, pushing nine-month purchases of property and equipment to $23.56 billion. That spending is financed mostly through debt: CoreWeave raised $10.3 billion of new debt in the quarter and repaid $2.34 billion, leaving total liabilities at $87.5 billion against total assets of $91.5 billion as of September 30, 2026.

A $132.5 billion backlog, and guidance raised again

CoreWeave said its revenue backlog — remaining performance obligations plus other committed-contract revenue it expects to recognize later, subject to delivery and availability — stood at approximately $132.5 billion as of September 30, 2026. That backlog is the company's preferred way of showing demand that hasn't yet converted into quarterly revenue.

Management used the quarter to raise guidance across the board. Full-year 2026 revenue guidance moved up to a range of $13.0 billion to $13.3 billion. Full-year adjusted operating income guidance rose to $1.275 billion-$1.375 billion, which CoreWeave said was a 26% increase at the midpoint versus its prior guidance. The company also raised the low end of its full-year 2026 capital expenditure guidance, from a prior range of $35 billion-$39 billion to a new range of $37 billion-$39 billion, citing plans to deploy infrastructure and deliver it to customers sooner than originally planned. Annualized run-rate revenue (ARR) — the final month of the quarter's revenue multiplied by 12 — is now guided to exit 2026 at $20 billion-$21 billion.

Doubling its active power footprint

CoreWeave frames its growth in gigawatts as much as dollars. The company added more than 200 megawatts of active power in the quarter to pass 1.7 gigawatts of active capacity, which it said doubles its active footprint since the start of 2026. Total contracted power — power secured for future buildout, not yet active — grew by more than 900 megawatts to over 4.6 gigawatts, and the company said it is diversifying its portfolio of power providers and expanding its powered land holdings. CoreWeave's international footprint also grew, to more than 1.3 gigawatts of contracted power across India, Indonesia and Spain. For the full year, CoreWeave still expects to end 2026 with more than 1.85 gigawatts of active power.

The distinction between "active" and "contracted" power matters for reading these numbers correctly. Active power is capacity that is live today, energizing racks of GPUs that customers can actually rent and run workloads on right now. Contracted power is secured for the future — a site CoreWeave has locked in a deal for, but which may still need construction, utility interconnection or cooling infrastructure before it comes online. A growing gap between the two, which CoreWeave now has at roughly 4.6 gigawatts contracted versus 1.7 gigawatts active, is effectively a forward-looking measure of how much more revenue-generating capacity is already locked in but not yet delivering revenue.

That buildout pace mirrors — at a different scale — the AI data center expansion Google Cloud has been running with AMD's EPYC processors, and it depends on the same category of sovereign and hyperscale power deals that have been reshaping chip supply chains this year.

Betting on Nvidia's next-generation Vera Rubin platform

On the hardware side, CoreWeave said it brought online multi-rack deployments of Nvidia's Vera Rubin NVL72 systems during the quarter, and that the first production workloads running on the platform — for AI coding startup Cognition — achieved 4.8 times greater token throughput than Nvidia's prior-generation GB200 NVL72 racks. CoreWeave also said it had announced support for Nvidia Vera, which Nvidia designed as the first CPU built specifically for AI agent workloads, to support what CoreWeave called rapid growth in post-training and agentic AI workloads.

CoreWeave also pointed to two independent, third-party performance benchmarks from the quarter: it said it became the only cloud provider to earn SemiAnalysis' Platinum ClusterMAX rating three consecutive times, and that it delivered the highest server throughput among cloud providers in the MLPerf Inference v6.1 benchmark round using an Nvidia GB300 NVL72 rack.

Who signed with CoreWeave this quarter

CoreWeave said it added new customers across AI labs, hyperscalers and enterprises, naming Canva, Mirendil, Parallel Web Systems, Serve Robotics, Substack and Wayve as new wins, and saying it expanded existing relationships with the Chan Zuckerberg Initiative, Cohere, Hippocratic AI, Jane Street, OpenAI and Perplexity. The company also said it launched CoreWeave Forge, a development environment meant to combine training, inference, evaluation and agent-development workflows for enterprise AI teams, and introduced a "Physical AI Field Engineering" program that embeds CoreWeave engineers directly with customer teams.

Beyond its commercial customer base, CoreWeave said it expanded into sovereign and public-sector AI work through a collaboration with defense contractor Leidos, and was selected alongside Parallel Works to deliver a managed AI cloud environment supporting biological research for DARPA. It also struck a multi-year agreement with Solidigm for priority access to enterprise SSD capacity — a supply-chain move that matters given how tight NAND and SSD capacity has been across the industry this year, and a reminder that AI infrastructure providers are now competing for storage allocation the same way they compete for GPUs and power.

CoreWeave also used the quarter to broaden what it sells beyond raw compute rental. The CoreWeave Partner Network, a curated set of production-validated integrations spanning observability, security and data infrastructure, launched with partners including Grafana Labs, CrowdStrike and VAST Data. Combined with CoreWeave Forge and the new Physical AI Field Engineering program, the company is positioning itself less as a bare-metal GPU landlord and more as a full AI development platform — a positioning shift that mirrors how other AI infrastructure providers have been expanding their own software layers this year.

How CoreWeave is funding a $37-39 billion capex plan

Funding that scale of spending is its own story. CoreWeave said it raised approximately $4.2 billion through an upsized offering of convertible senior notes during the quarter, established its first at-the-market equity program for up to 35 million shares of Class A common stock, and closed a $2.6 billion "DDTL 5.5" facility, which the company described as its first financing facility designed specifically to support shorter-duration customer contracts. CEO Michael Intrator said the company's growth is now supported by "multiple sources of funding, including financing backed by customer contracts, operating cash flow, and approximately $9 billion in customer prepayments and deposits expected in the next three months alone."

Taken together, CoreWeave's balance sheet nearly doubled in size in nine months: total assets stood at $91.5 billion as of September 30, 2026, versus $49.3 billion at the end of 2025. Property and equipment, net — largely GPUs, servers and data center build-out — makes up $56.8 billion of that, up from $30.6 billion at the start of the year.

What the results mean for CoreWeave and its AI-cloud rivals

CoreWeave's results land in the same week that rival AI infrastructure provider Nebius posted its own swing to adjusted profitability, and they follow a string of sovereign and hyperscale AI buildout deals — including Nvidia's NVLink Fusion partnership with Rebellions and SK Telecom in Korea — that underline how much of the current AI capex cycle is still being driven by compute scarcity rather than slowing demand. CoreWeave's widening GAAP loss shows that scarcity isn't free: financing a gigawatt-scale buildout means carrying tens of billions of dollars in debt and depreciation years before all of that capacity is fully contracted and generating revenue. The backlog and raised guidance are CoreWeave's argument that the demand underneath that spending is real and still growing.

What's next for CoreWeave

CoreWeave's next scheduled checkpoint is its fourth-quarter and full-year 2026 report, expected in early 2027, where investors will be watching whether the company hits the high end of its newly raised revenue, ARR and capex guidance, and whether its active power build keeps pace with the 1.85-gigawatt year-end target. CoreWeave's quarterly SEC filings, investor webcasts and press releases are published at investors.coreweave.com.

Written byMara Lindqvist

Mara Lindqvist edits the hardware desk. She covers graphics cards, processors, memory and storage, the foundries and chip designers behind them, and what the numbers on a spec sheet mean for people choosing a PC. Specifications in her stories come from manufacturer spec pages and datasheets.

More from Mara LindqvistAll Hardware coverage