Palantir picks Nebius for sovereign AI cloud

Palantir Technologies has signed an agreement making Nebius its preferred sovereign AI infrastructure partner, a deal that puts Nebius' compute and inference endpoints inside Palantir's enterprise perimeter and lets eligible commercial customers deploy open models on Nebius infrastructure, Data Center Dynamics first reported.

Sovereign AI has mostly been a government procurement category, tied to national clouds and classified workloads. Palantir and Nebius are trying to make it a commercial product, deploying new compute capacity through modular data center builds at sites that already have power and aimed at Palantir's customer base.

Instead of buying land and hoping for interconnection, Nebius is sourcing power-rich sites and dropping modular capacity onto them, and for Palantir's customers the offering is a contractual claim on compute that runs under conditions they control — the working definition of sovereignty that Alex Karp and Arkady Volozh are selling. Karp framed it as ontology meeting infrastructure; the more durable read is that Nebius' modular rollout is the delivery mechanism for a new kind of offtake.

Grid consent and load flexibility are becoming the assets that decide what gets built, and this agreement applies that logic directly. Nebius' expansion is anchored to sites where power already exists, short-circuiting the interconnection queue that has become the critical path for most data center projects, and the modular approach keeps capital deployment flexible, letting Nebius match capacity to actual Palantir demand rather than committing to a single giant campus.

Nebius comes to this deal with a funded balance sheet: last month it closed a senior notes offering totaling $5.75 billion, proceeds earmarked for data center construction, AI cloud development, GPU procurement, and general corporate purposes — a raise that was announced as a $4.5 billion convertible with room to grow. The notes priced at a four-point coupon spread, a cost of capital that reflected the market's view of Nebius as an infrastructure bet with a growth-stock risk profile.

The funding is what makes the Palantir partnership credible. Nebius is a neocloud with a European and North American presence, spun out of Russia's Yandex in 2024, and it has the balance sheet to pre-build capacity. What it needed was demand visibility, and Palantir brings a commercial customer base with workloads that need sovereignty features — data residency, model control, auditability — and a sales motion that reaches into regulated industries where those features are becoming deal-breakers.

The Palantir side of the arrangement is lighter on hard commitments. The agreement names Nebius as preferred partner and envisions customers deploying on Nebius infrastructure, but the coverage does not specify minimum capacity commitments, reserve levels, or a timeline for the modular rollouts. That vagueness is typical of partnership announcements at this stage, and it cuts both ways: for Nebius, the deal is a demand signal that supports its buildout plans without guaranteeing a single megawatt of utilization, while for Palantir's customers it is an option on capacity, not a firm allocation.

The commercial logic still holds. Open models deployed on trusted infrastructure is a value proposition aimed at enterprises that want AI performance without handing their data to a hyperscaler's default stack, and if Palantir's ontology layer is genuinely the differentiator it claims to be, Nebius' role is to supply the compute substrate underneath it. The risk is that sovereignty becomes a marketing label rather than a technical guarantee, and the partnership degrades into a reseller agreement with extra compliance paperwork.

The precedent is closer to the utility-style contract than the hyperscaler model: rather than building capacity on its own balance sheet and selling it by the minute, the customer secures dedicated capacity at a specific location with specific operational controls. Palantir's enterprise perimeter concept pushes in that direction, giving customers a defined enclave within Nebius' cloud, and it suggests the two companies are serious about serving customers who cannot tolerate shared tenancy for their most sensitive models.

What the announcement leaves open is pricing. Sovereign infrastructure carries additional cost — dedicated enclaves, geographic separation, compliance tooling, and the operational overhead of running modular sites rather than a single dense campus — and the coverage does not say how the economics split between Palantir, Nebius, and the end customer. The likely answer is that Palantir bundles the infrastructure into its platform pricing, which would make Nebius a wholesale supplier to Palantir's margin rather than a direct seller to end users.

That would be the right structure for both companies, and it points to where this partnership could create real value. Palantir gets a sovereignty story it can sell without owning data centers, Nebius gets channel distribution into Palantir's enterprise base, and the modular buildout at power-rich sites gives Nebius a capital-efficient path to growth, with the Palantir relationship providing a reason to believe the capacity will be consumed.

None of this makes Nebius an infrastructure-grade credit yet. Its $5.75 billion debt raise priced at a spread that read as a signal — the near-term GPU buildout priced like infrastructure, the long haul like a risk asset. This deal strengthens the near-term case by adding a marquee customer, but it does not change the long-term question, which is whether Nebius can convert its neocloud footprint into contracted, recurring revenue at the scale its debt load requires.

The modular data center strategy is the hedge that makes the bet manageable. By building at sites with existing power, Nebius avoids the multi-year interconnection lottery that has delayed some of the largest AI campuses in development, and the trade-off is scale — modular sites are smaller than the 500-megawatt megacampuses that dominate headlines — but for Palantir's sovereign customer base, smaller and more controllable may be exactly the point.

Watch for the first named customer. The partnership will produce press releases about capability, but the proof will be a regulated enterprise — a bank, a health system, a defense contractor — that signs up to run open models on Nebius infrastructure inside Palantir's perimeter. When that happens, sovereignty will have moved from a geopolitical talking point to a line item in someone's IT budget.