Nebius Sees AI Demand Visibility Stretching to 2028 as Auction Pricing Delivers 15-20% Premiums
Goldman Sachs Communacopia + Technology Conference, September 8, 2026
Nebius Group offered investors a notably more bullish demand picture than it had previously disclosed, with CEO Arkady Volozh telling Goldman Sachs analyst Alexander Duval that customer order visibility has extended well beyond what the company was signaling just months ago. Where Nebius previously spoke of an 18-month demand runway, Volozh said that window has stretched to roughly 24 months or more, with the company now fielding orders for first and second quarter 2028 delivery, including requests for tens of thousands of Vera Rubin GPUs. "We see demand today as unlimited," Volozh said. "Nothing is unlimited in this world. But for now, the demand in all the visible perspective is much higher than anybody can serve."
The comments matter because durability of the AI infrastructure cycle has been the central investor debate around neocloud names, and Nebius is now on record saying that if it wanted to, it could presell all of its 2027 capacity today. Management said it is deliberately not doing so, choosing instead to hold back free capacity for pricing power reasons.
Auction Pricing Mechanism Yields 15-20% Premium
Perhaps the most concrete new data point from the session was color on Nebius's capacity auction experiment for Blackwell processors, run several weeks before the conference. Volozh said the auction produced prices 15% to 20% higher than the company's prior list prices, even after multiple manual price increases had already failed to clear the market. "We ran our first experimental auction for the Blackwell processors a month ago. We achieved results where we got 15%, 20% higher price than we ever sold before, after all our price raises," he said. Management plans to expand the auction mechanism further down the customer base, including to its long tail of self-served clients who currently operate under preemptive capacity arrangements. Volozh framed the auction as the cleanest price-discovery tool available in a structurally undersupplied market, with future discounting or upselling to be layered on top of auction-discovered prices depending on the customer. This is a meaningful signal for margin trajectory, since it suggests Nebius has both the demand density and the mechanism to keep extracting price increases even as the industry adds capacity.
Palantir Sovereign Partnership Opens Enterprise Channel
Nebius also used the appearance to detail a newly announced partnership under which it becomes a "sovereign" infrastructure partner to Palantir. Volozh explained the mechanics: enterprises increasingly want to avoid feeding proprietary data into commercial frontier models, since doing so improves competitors' access to the same intelligence. Instead, the preferred path is training open-weight models repeatedly on a company's own data in a closed loop, a lower starting point in raw quality but one that can surpass commercial models within a narrow domain. Volozh cited Shopify's CEO publicly stating that a repeatedly fine-tuned open-weight model outperformed GPT-5.6 within Shopify's specific use case. Palantir supplies the tooling for this loop but needed a controlled infrastructure stack, from GPU to data center, that does not route client data through a hyperscaler. Nebius provides that stack up through its "token factory" of open models, with Palantir's enterprise relationships acting as a distribution channel. Chief Revenue Officer Marc Boroditsky added that the tie-up should accelerate customer diversification, calling it a bridge from Nebius's AI-native roots into Palantir's established commercial enterprise base.
Hyperscaler Capacity Resale Not a Structural Threat
Asked about Meta's exploration of a cloud-like offering and SpaceX's decision to resell excess compute to the likes of Anthropic, Google and Azure, Volozh downplayed the competitive threat, framing it instead as incremental raw material for Nebius's own resale business. He noted that hyperscalers offloading unused capacity typically do so on a bare-metal basis, without the cloud and services layer Nebius adds. Sizing the opportunity, he estimated the market added roughly 5 to 10 gigawatts of capacity a year ago and could reach 15 to 20 gigawatts in 2028, versus the several gigawatts SpaceX or Meta might bring to market. "It doesn't change the market radically... It's another 20% here, it's a 30% there," he said, arguing additional bare-metal supply is net positive for Nebius since the company remains capacity-constrained.
Contract Structure and Recent $1 Billion-Plus Wins
Boroditsky laid out Nebius's three-tier contract framework: short-term deals of three to six months tied to scale training runs that command a premium multiple of ARR per megawatt; medium-term one-to-three-year agreements that represent "the lion's share of opportunity" and span AI natives, digital natives and early enterprise adopters; and five-year-plus long-term agreements historically reserved for hyperscaler relationships, used primarily to unlock capitalization for expanding the core business. On the four landmark deals disclosed last quarter, each averaging roughly $1 billion, Boroditsky said the wins were decided in proof-of-concept testing rather than on price, with one customer's technical team reportedly calling it "the best technical experience they have ever had with any supplier in the industry." All four customers are already in discussions for additional GB300 capacity and Vera Rubin requirements, and Nebius said it has a pipeline of similar prospects behind them.
Asset-Light Build-Out Model to Accelerate Capacity Growth
Volozh detailed a newly announced asset-light partnership structure aimed at scaling capacity faster than Nebius's balance sheet alone would allow. The model targets a class of counterparties, largely former electricity and data center companies, that control physical plots, power access and cheap financing but lack the technical know-how to operate AI infrastructure. Nebius would supply data center design expertise, racks, its software stack and its go-to-market organization in exchange for a share of the economics. Volozh said the company already has a long list of interested partners and several projects underway targeting capacity coming online in 2027 and beyond, describing it as "a non-linear way for us to grow" alongside organic build-out.
Power Sourcing Gets More Creative, Including Bloom Hydrogen Deal
On power procurement, Volozh acknowledged that physical build constraints, not customer demand, remain the binding limit on growth. Nebius is now pursuing grid connections, on-site generators, and a new partnership with Bloom Energy for hydrogen and co-generation power. "Whatever we can build as fast as we can build, everything will be sold," he said, reinforcing the message that supply, not demand, is the operative constraint on the business heading into 2027 and 2028.