Nebius Group N.V. (NBIS)

News Summary

• Nebius closed 4 landmark AI cloud deals averaging over $1B each at $20M–$25M per megawatt with 50–60% upfront CapEx coverage, and management is deliberately withholding 2027 capacity from long-term contracts, betting on further price appreciation; a capacity auction for Blackwell-generation GPUs cleared 15% above prior highs, while short-term scale training deals are pricing at $40M–$50M per megawatt, signaling a new pricing regime for AI compute infrastructure.
• Vera Rubin GPUs are already in Nebius labs producing expected results with deployment targeted for late 2026 or early 2027, and the agentic layer is gaining measurable traction: Tavily's developer community grew from 1M to 2.5M users in roughly one quarter, Token Factory achieved a 100% quality score on GLM 5.2, and Nebius is adding ARM and CPU deployments to handle the CPU-heavy demands of agentic orchestration and tool calling.
• Nebius raised its year-end contracted power target to 5 gigawatts and plans to deploy more than 1 gigawatt of new capacity in 2027 alone, backed by $40B in contracted backlog, $9B in customer prepayments, and a $775M asset-backed facility at SOFR+250bps; total contract value grew nearly fourfold QoQ, with ~70% of deals including upfront payments that offset half to three-fifths of related CapEx.
• Q2 revenue hit $582.3M (+454% YoY, +46% QoQ), with the AI cloud segment generating $574.9M at a 50% adjusted EBITDA margin; annualized run-rate revenue reached $3B at quarter-end (+56% from Q1's $1.9B), adjusted EBITDA swung to +$236.2M from a -$21M loss a year ago, and CapEx surged to $5.66B in the quarter vs. $510.6M a year earlier, with cash of $8.04B providing runway.
• NBIS rose ~9% on the close of August 14 and was up ~12–15% in premarket trading post-earnings; shares are up ~131% YTD vs. the S&P 500's ~13%, and Zacks noted an 82% EPS beat (adjusted loss of $0.12 vs.

Financial Highlights

Earnings Call Highlights

Nebius delivered a blowout Q2 with 454% revenue growth and a rapid acceleration in deal quality, signaling that the independent AI cloud market is entering a new pricing regime that management is deliberately exploiting by withholding capacity from long-term contracts. The tone from Arkady and the commercial team was unusually confident and strategic, framing Nebius not as a GPU reseller but as a full-stack platform company with pricing power, auction capability, and an asset-light model that could scale without proportional balance sheet risk. Analysts from Morgan Stanley, Goldman Sachs, Citi, and Baird all probed the sustainability of the capacity and pricing ramp, and management's answers were consistently bullish and specific, citing $40M-$50M per megawatt short-term deals, a capacity auction clearing 15% above prior highs, and $40 billion in contracted backlog. The dominant theme was optionality: Nebius is deliberately not selling out 2027 capacity because it believes prices will continue to rise, a posture that implies strong conviction in sustained AI infrastructure demand. The call functioned less as a backward-looking earnings report and more as a forward-looking investor pitch for 2027 and beyond.

• Nebius closed 4 landmark deals averaging over $1 billion each at $20M-$25M per megawatt with 50-60% upfront CapEx coverage, and management stated they could sell their entire 2027 capacity today at these terms but are deliberately holding back, betting on further price appreciation.
• A capacity auction cleared at 15% above the highest price Nebius had ever charged for Blackwell-generation GPUs, and short-term scale training deals are pricing at $40M-$50M per megawatt, representing a dramatic premium over midterm contract rates and providing real-time price discovery in a market where 'the price can change literally as we move through a sales cycle,' per Marc Boroditsky.
• Vera Rubin GPUs are already in Nebius labs producing expected results, with deployment targeted for late 2026 or early 2027; management noted the technical step from GB300 to Vera Rubin is easier than the prior generational transition, suggesting a smoother ramp.
• The agentic and inference layer is gaining traction: Tavily's developer community grew from 1 million to 2.5 million in roughly one quarter, Token Factory achieved a 100% quality score on GLM 5.2 with top independent benchmark rankings, and Nebius is adding ARM and CPU deployments alongside GPUs to handle the CPU-heavy demands of agentic orchestration and tool calling.
• Nebius raised its year-end contracted power target to 5 gigawatts and plans to deploy more than 1 gigawatt of new capacity in 2027 alone, funded by a combination of $9 billion in customer prepayments, a $775 million asset-backed facility at SOFR+250bps, and $40 billion in committed backlog available for further asset-backed financing.

Sources

• Nebius Group (NBIS) Reports Q2 Loss, Tops Revenue ...
• Nebius reports second quarter 2026 financial results - Business Wire
• Nebius Q2 2026 earnings beat revenue estimates on AI cloud ...