Meta Strikes Massive $27 Billion AI Infrastructure Pact with Nebius to Secure Next-Gen Compute PowerIn one of the largest external infrastructure deals in its history, Meta has announced a strategic partnership with Nebius, a leading AI cloud provider, worth up to $27 billion. This 5-year agreement is designed to rapidly scale Meta’s data center capacity to meet the skyrocketing demands of its global AI initiatives.
Building the Foundations of the Future
The partnership is structured into two primary phases:
Dedicated Capacity ($12 Billion): Nebius will provide Meta with exclusive, large-scale AI compute resources across multiple global regions. This includes the development of a flagship data center featuring one of the first large-scale deployments of the NVIDIA Vera Rubin platform. Deliveries are set to commence in early 2027.
Flexible Expansion ($15 Billion): Meta has committed to purchasing additional capacity from Nebius’s future data center expansions over the next five years. Notably, any surplus capacity not utilized by Meta can be sold by Nebius to other AI cloud customers, providing the Dutch-based firm with a diversified revenue stream.
The Multi-Pronged AI Strategy
This deal is the latest move in Meta’s aggressive "Infrastructure First" philosophy. By partnering with Nebius, Meta secures long-term access to specialized hardware while continuing to collaborate with industry giants like NVIDIA and AMD, as well as advancing its own custom silicon through the MTIA (Meta Training and Inference Accelerator) project.
Nebius (restructured from the former Yandex) has rapidly become a rising star in the Neocloud landscape. Just days before the Meta deal, NVIDIA announced a $2 billion investment in Nebius to build a 5-gigawatt "AI factory." Meta's choice of Nebius underscores the emergence of new AI-focused cloud providers challenging giants like AWS and Azure.
In the same week, Meta launched four new MTIA chips (300, 400, 450, and 500), specifically designed for inference workloads. Meta's continued $27 billion investment in Nebius demonstrates that, despite having its own chips, the demand for computing power for future Llama 4 and 5 is far greater than its existing chip manufacturing facilities can handle.
Access to the NVIDIA Vera Rubin platform through Nebius will allow Meta to run "Argentic AI" models with up to 10 times better performance per watt compared to the Blackwell era, resulting in significant long-term operating costs.
Analysts view this deal as a way to "lock in" future cloud rental prices, as the demand for GPUs is projected to remain in the market during 2026-2028 (supply shortage). The five-year contract with Nebius assures Meta of a guaranteed supply of resources at the agreed-upon price.
NVIDIA Unleashes Vera Rubin Groq 3 Integration Redefines AI Inference Speed.
Source: Nebius
Meta Strikes Massive $27 Billion AI Infrastructure Pact with Nebius to Secure Next-Gen Compute PowerIn one of the largest external infrastructure deals in its history, Meta has announced a strategic partnership with Nebius, a leading AI cloud provider, worth up to $27 billion. This 5-year agreement is designed to rapidly scale Meta’s data center capacity to meet the skyrocketing demands of its global AI initiatives.
Building the Foundations of the Future
The partnership is structured into two primary phases:
Dedicated Capacity ($12 Billion): Nebius will provide Meta with exclusive, large-scale AI compute resources across multiple global regions. This includes the development of a flagship data center featuring one of the first large-scale deployments of the NVIDIA Vera Rubin platform. Deliveries are set to commence in early 2027.
Flexible Expansion ($15 Billion): Meta has committed to purchasing additional capacity from Nebius’s future data center expansions over the next five years. Notably, any surplus capacity not utilized by Meta can be sold by Nebius to other AI cloud customers, providing the Dutch-based firm with a diversified revenue stream.
The Multi-Pronged AI Strategy
This deal is the latest move in Meta’s aggressive "Infrastructure First" philosophy. By partnering with Nebius, Meta secures long-term access to specialized hardware while continuing to collaborate with industry giants like NVIDIA and AMD, as well as advancing its own custom silicon through the MTIA (Meta Training and Inference Accelerator) project.
Nebius (restructured from the former Yandex) has rapidly become a rising star in the Neocloud landscape. Just days before the Meta deal, NVIDIA announced a $2 billion investment in Nebius to build a 5-gigawatt "AI factory." Meta's choice of Nebius underscores the emergence of new AI-focused cloud providers challenging giants like AWS and Azure.
In the same week, Meta launched four new MTIA chips (300, 400, 450, and 500), specifically designed for inference workloads. Meta's continued $27 billion investment in Nebius demonstrates that, despite having its own chips, the demand for computing power for future Llama 4 and 5 is far greater than its existing chip manufacturing facilities can handle.
Access to the NVIDIA Vera Rubin platform through Nebius will allow Meta to run "Argentic AI" models with up to 10 times better performance per watt compared to the Blackwell era, resulting in significant long-term operating costs.
Analysts view this deal as a way to "lock in" future cloud rental prices, as the demand for GPUs is projected to remain in the market during 2026-2028 (supply shortage). The five-year contract with Nebius assures Meta of a guaranteed supply of resources at the agreed-upon price.
NVIDIA Unleashes Vera Rubin Groq 3 Integration Redefines AI Inference Speed.
Source: Nebius
Comments
Post a Comment