CoreWeave Vera Rubin NVL72 Cluster Deployment, September 2026

CoreWeave brought multi-rack NVIDIA Vera Rubin NVL72 clusters online at its Livingston facility this month — the first commercial deployment of the next-generation architecture — against a contract backlog that grew to more than $129 billion in Q2 2026.

CoreWeave announced this month that it has brought multi-rack NVIDIA Vera Rubin NVL72 clusters online at its Livingston facility, marking the first commercial deployment of the next-generation GPU architecture. According to reports, Dell Technologies delivered the world's first Vera Rubin NVL72 server rack to CoreWeave. The clusters are designed for agentic AI workloads, and CoreWeave simultaneously announced new capabilities in its AI Object Storage platform.

The rollout arrives against the strongest quarterly numbers CoreWeave has yet reported. In Q2 2026, revenue reached $2.58 billion, up 112% year on year, while the contract backlog grew from roughly $104 billion to more than $129 billion during the quarter; full-year guidance was raised to $12.4–$13.2 billion. All four of the major AI research laboratories are now signed customers, with the Meta contract reportedly expanded to $21 billion. The customer base has been broadening beyond the frontier model training core: Hudson River Trading selected CoreWeave to host its quantitative research platform, and Parallel Works is running a DARPA-commissioned managed AI cloud environment on CoreWeave infrastructure.

That commercial momentum rests on a capital programme of unusual scale. Slicast's compilation of SEC XBRL data shows capital expenditure reaching $10.31 billion in FY2025, up 18.5% from $8.70 billion in FY2024, at an intensity of 201% of the year's $5.13 billion in revenue — the highest absolute capex of any of the seven neoclouds Slicast tracks, though sixth by intensity ratio among that peer group. The single largest quarter on record was $7.70 billion in the period ended March 2026. The build is still running: Blockfusion signed a 15-year lease at its Niagara Falls data centre campus for CoreWeave capacity this month, and Core Scientific reportedly delivered 200 MW of power capacity ahead of schedule as part of CoreWeave's stated priority to fast-track power delivery to infrastructure partners.

The stock's path has been more volatile than the top-line figures might suggest. Shares closed at $82.98 on September 14, down 6.8% on the session; no published report identifies the specific cause of that day's move. The decline follows a pattern of sector-wide pressure: a rate-driven sell-off in August sent CoreWeave down roughly 7% alongside Nebius and TeraWulf, and reports of warnings about AI training demand softening have introduced recurring uncertainty into neocloud valuations. The lock-up period on post-IPO shares expired on August 14, adding a supply overhang following the Q2 earnings rally. At $83.35 on the day of writing, up 3%, some analysts — as reported by Crypto Briefing — argue the current discount does not adequately reflect CoreWeave's lead in deployed capacity; others have flagged technical warning signals despite the revenue growth.

The broader competitive picture is evolving in ways that complicate the long-run outlook. Meta's announced plans to expand its own compute capacity, alongside SpaceX's reported entry into AI infrastructure, have focused investor attention on the risk that hyperscaler self-sufficiency could eventually constrain neocloud demand growth. Spot pricing competition is already visible: Lambda reportedly offered B200 instances at $6.69 per GPU-hour as of August 2026. Nebius, a direct rival, has reportedly sold out its GPU capacity through 2027 but faces a substantial funding challenge to sustain its expansion. CoreWeave's contracted backlog and early access to NVIDIA's latest hardware provide some insulation from spot-market pricing pressure, but they do not eliminate the financing burden of operating at 201% capex intensity.

Three signals will define the near-term narrative. First, how quickly the Vera Rubin NVL72 clusters convert into incremental signed revenue: the first commercial deployment is a proof point, but the pace of follow-on customer commitments will reveal whether demand for next-generation hardware runs as deep as the backlog implies. Second, whether the Q3 backlog continues its expansion beyond $129 billion without signs of customer concentration or cancellation risk — all four major AI labs are signed customers, and any shift in their infrastructure plans would have a disproportionate effect. Third, the capex trajectory: with the March 2026 quarter setting a single-period record of $7.70 billion, the question of when CoreWeave generates meaningful free cash flow will remain the central valuation debate through the remainder of the year.