Meta to spend up to $27 billion on Nebius Group AI infrastructure over the next five years

Meta is ready to spend up to $27 billion over five years to gain access to AI infrastructure from Nebius Group. This is one of the most notable signals that the largest tech companies are already competing not only on models, but also on computing power, without which these models cannot be quickly trained and deployed to production.

Scale of the Agreement

Essentially, this is about long-term access to external AI infrastructure. Meta can direct up to $27 billion toward this over five years. The wording is important: the company is not buying Nebius Group outright and is not announcing an acquisition, but is reserving a large volume of capacity for itself. For the market, this signals that even big tech companies no longer find it sufficient to simply build everything internally, when the demand for training and servicing AI systems is growing faster than new capacity can be deployed.

The amount itself sets the scale. If the commitments are fully utilized, this is one of the largest known infrastructure contracts in the AI segment in recent times. Such deals are typically not made as experiments, but for specific product and research loads: training new generations of models, servicing users, testing features, and creating a buffer for demand spikes. In other words, Meta appears to be hedging against computing deficits at a time when the race for model quality is only accelerating.

It is also important that the deal's horizon stretches over five years. Over such a timeframe, companies typically purchase not one-time capacity, but supply predictability. This helps balance between proprietary data centers and external partners: keeping some load in-house and quickly transferring the rest to where the needed amount of accelerators is available. For Meta, this approach reduces the risk that new AI releases or model updates will have to be postponed due to simple infrastructure deficits.

Why This Matters for Meta

For Meta, infrastructure is no longer just a support layer, but a direct part of product strategy. The more actively the company develops AI services, the more it needs GPUs, networks, storage, and supply chains that can scale without a long pause for building proprietary data centers. Access to capacity years in advance provides not only speed, but also predictability: teams can plan model training and feature launches without checking hardware availability every quarter.

Practically, such a contract can give Meta several advantages at once:

• more resources for training large models;
• stable capacity for deploying AI features to mass-market products;
• reserved bandwidth for periods of sharp load growth;
• less dependence on a scenario where all infrastructure must be built entirely in-house.

For Nebius Group, this is also a strong signal. If the company receives a contract of this scale from Meta, the market perceives it not as a niche contractor, but as an infrastructure partner capable of handling one of the most demanding AI workloads in the world. In the AI market, this is no less important than money: such agreements increase the trust of other large clients who need not just access to GPUs, but confidence in reliability, timelines, and delivery scale.

What This Means

The AI race is increasingly shifting from the level of demos and releases to the level of energy, servers, and long-term computing contracts. The news of Meta's plans to spend up to $27 billion shows a simple shift: winners will include not only creators of strong models, but also those who secured infrastructure capacity in advance. For the market, this means further growth in demand for independent AI providers and an even higher price for planning errors in capacity.