Nebius Stock: A 50-Megawatt Saudi Option, a Prepayment Cushion, and Piper Sandler's Lower Bar

Piper Sandler trimmed its price target on Nebius from $237 to $185 yesterday while keeping a "Neutral" rating, and the reasoning behind the cut says as much about the company's financing model as it does about its expansion pipeline. Analyst James Fish pointed to investor worries over AI demand, returns on invested capital and the funding of AI services firms, and the firm also marked down long-term valuation multiples across the sector. The cautious stance lands at an awkward moment: Nebius has just unveiled a fresh set of growth options, none of which yet generates revenue.

A framework agreement, not capacity

At the NVIDIA GTC conference in Berlin, HUMAIN and DataVolt signed a preliminary framework agreement with Nebius covering up to 50 megawatts of data center capacity at the planned Oxagon AI campus in Saudi Arabia, according to NEOM. The capacity is expected to come online in stages starting in the fourth quarter of 2027. Nebius would take half of the first 100 megawatts currently under construction, a tranche that sits inside an initial phase planned at 360 megawatts overall.

Every qualifier in that announcement matters. The agreement is provisional, the figure is a ceiling and the start date is an expectation. Treating the deal as computing power already in hand would be premature; its value for now lies in the optionality it creates. The same test applies to the Maasvlakte AI data center in the Netherlands, where Nebius signed a letter of intent with Volt on October 12 for a "substantial portion" of a planned 600-megawatt site, with construction reportedly still uncertain.

Both projects belong in the category of strategic options. That is not a downgrade — a company needs such openings to prepare its expansion. But investors should distinguish between Nebius lining up access to a possible location and actually turning that access into usable infrastructure. The size of a project alone does not settle the question.

Prepayments carry part of the capex load

How Nebius intends to fund larger builds is already visible in signed customer contracts. In the second quarter of 2026 the company reported four AI cloud deals with an average total contract value above $1 billion each. Customers are shouldering a meaningful share of the upfront financing, which means Nebius does not have to carry every associated investment cost on its own balance sheet first.

Roughly 70% of the deals closed in that quarter included prepayments, covering 50–60% of the related investment costs. Many of the contracted capacities are due to become available at the end of 2026, while the bulk of the resulting contract revenue is expected to land only in the following year. That gap between signing, building and booking revenue keeps the financing of the interim phase at the center of the growth story. None of this says anything about the terms of the planned Oxagon project.

Flexible access models and a measurable customer benefit

Alongside the build-out, Nebius is working on how its compute gets used. Yesterday it announced publicly visible live spot prices for GPU capacity, plus reservation auctions and usage-based access. It is also piloting a rebate program for customers who release unused GPUs. Different access models can make it easier for customers to buy compute that fits their needs, and the release pilot is notable because it addresses resources already reserved rather than simply adding more capacity. The announcement on its own does not prove commercial success.

The Physical AI Workbench unveiled with NVIDIA fits the same pattern. The open-source development tool supports robotics workflows from data generation through evaluation, and customers have reported cutting development time by up to 40%, according to the company. Such customer effects carry more weight than a list of planned sites because they describe concrete utility. The reach of the claim stays limited, though: reported time savings are neither a guarantee of broad success nor a measure of Nebius's profitability.

What the cautious case rests on

Piper Sandler's stance is that the group could stay under pressure until free cash flow improves or the financing picture clears. The firm continues to recommend building a position gradually, but prefers CoreWeave. That framing ties the growth and financing questions together: more expansion options do not by themselves answer the question of economic quality.

The constructive reading is that Nebius is combining build-out plans with more flexible access models and tangible applications, which speaks to strategic breadth. What it does not justify is an unconditional vote of confidence. The measure that counts is whether provisional agreements turn into usable capacity and whether technical customer benefits turn into durable earnings — not the announced megawatt figures.