Nebius Group and CoreWeave Surge: Nvidia Endorsement, Blowout Q2 Results, and Unprecedented Blackwell Pricing Power
Despite structural concerns regarding hyperscaler insourcing, specialized "neocloud" providers have solidified their position as indispensable partners in the artificial intelligence buildout. In September 2026, Nvidia CEO Jensen Huang delivered a resounding endorsement of Nebius Group (NBIS) and CoreWeave (CRWV), highlighting their unique ability to secure critical physical infrastructure—land, power, and facilities—that traditional hyperscalers have exhausted.
Jensen Huang's Endorsement at Goldman Sachs
Speaking at the Goldman Sachs Communacopia + Technology Conference in September 2026, Jensen Huang addressed the core bottleneck of the AI capex cycle: physical infrastructure. He spotlighted CoreWeave and Nebius Group as key enablers that are bypassing traditional utility grid bottlenecks:
“They secure land, power, and shell for us that the CSPs have already exhausted1,” Huang explained, describing these neocloud operators.
With Nvidia's demand exceeding 100% growth but supply constrained to approximately 70% growth for fiscal 2028, these specialized clouds act as a vital bridge to deploy Nvidia's reference architectures at scale. Nvidia has backed both companies with separate $2 billion strategic investments, targeting the deployment of over 5 gigawatts of AI infrastructure by 2030.
CoreWeave: $104 Billion Backlog and Massive Capex Expansion
CoreWeave's Q2 2026 financial disclosures reveal the sheer scale of its operations:
- Revenue: Q2 revenues reached $2.58 billion, up 112% from $1.21 billion in the prior year.
- Backlog: CoreWeave's revenue backlog exploded to a massive $104 billion, representing a 246% annual increase.
- Capital Expenditure: The company increased its 2026 capex guidance to a range of $35 billion to $39 billion to fund its aggressive expansion.
- Debt and Net Income: The rapid build remains capital-intensive, with CoreWeave carrying $33.8 billion in total debt and posting a Q2 net loss of $626 million.
Nebius Group: 454% Revenue Growth and Blackwell Auction Premiums
Nebius Group has emerged as a leader in capital efficiency and pricing power. Its Q2 2026 earnings report highlighted an extraordinary inflection:
- Revenue: Q2 revenue surged 454% year-over-year to $582.3 million.
- EBITDA & Profitability: AI-cloud adjusted EBITDA margins reached 50% (Group adjusted EBITDA of $236 million at a 41% margin), while its adjusted net loss was slashed by 64% to $33.2 million.
- Contracted Power & RPO: Contracted power capacity was raised to 5 gigawatts by year-end, and Remaining Performance Obligations (RPO) stand at $37.5 billion.
- Anchor Deals: Nebius has secured a potential $27 billion five-year infrastructure agreement with Meta Platforms and up to $17.4 billion in GPU-cloud commitments from Microsoft through 2031.
Blackwell Pricing Power and Capacity Auctions
Nebius's pricing power is demonstrating extreme demand tightness. On its Q2 earnings call, CEO Arkady Volozh noted that Nebius "could sell today our entire 2027 capacity on these terms if we wanted to2," but is deliberately holding back capacity for premium, short-duration deals.
Nebius's first capacity auction for Nvidia's Blackwell GPUs cleared 15% above the highest price the company had ever charged. Short-duration contracts are currently being negotiated at a premium of $40 million to $50 million per megawatt, compared to $20 million to $25 million for mid-term deals. This extreme pricing power indicates that despite concerns over an "AI ROI gap," buyers are willing to pay significant premiums to secure immediate Blackwell compute.
-
An instance of Hardware deployment is no longer gated by silicon fabrication but by gigawatt-scale power allocations. — Huang names the scarce inputs as land, power, and shell — the hyperscalers are exhausted on sites and grid, not on chips. ↩︎
-
An instance of Generational compute capacity is fully monetized and locked in before the silicon even exists. — CoreWeave's $104B backlog and Nebius's sellable 2027 capacity show next-generation compute fully pre-sold at premium prices before it is deployed. ↩︎