Nebius ditches owned datacenters for partner-led compute model

The more pointed question the summary sidesteps is who Nebius is actually partnering with and on what terms, because the risk profile of an asset-light model depends almost entirely on whether those colocation agreements give Nebius pricing stability or leave it exposed to capacity squeezes when GPU demand spikes.

This is largely disconnected from the EU-versus-Google story covering Android and Search interoperability that ran the same day on The Verge. The Nebius move belongs to a separate thread: the ongoing shakeout among neocloud and GPU-cloud providers trying to find durable unit economics as hyperscaler competition intensifies. Asset-light infrastructure is a familiar playbook from earlier cloud generations, where companies like Rackspace eventually found that decoupling from owned hardware reduced capex but also compressed margins when landlord costs rose. For Nebius, the strategic logic is defensible in the short term, but it trades one risk (overbuilding) for another (dependency on third-party operators who may prioritize larger anchor tenants).

Watch whether Nebius discloses named colocation partners or contract durations within the next two quarters. Named, long-term agreements would suggest genuine cost predictability; vague partner language would confirm this is more positioning than structural commitment.