← All stories
● Covered by 1 source · 1 reportMedium impact1 neutral

AI Clusters Exceed Single Data Center Limits, Driving Need for Scale-Across Interconnects

🔄 Updated 1d ago
New to BrevFeed? We gather this story from every outlet covering it into one summary — ranked by real-world impact, not just the latest headline — so you never miss what matters. What is BrevFeed? →

Key points

  • AI clusters with over 100,000 accelerators exceed single data center limits.
  • Power and cooling constraints necessitate distributing AI workloads across multiple facilities.
  • A new 'scale-across' interconnect domain is emerging for multi-site connectivity.
  • Efficient distributed computing requires more than just faster network connections.

AI Clusters Outgrow Single Data Centers

The rapid expansion of AI clusters, now commonly featuring over 100,000 AI accelerators and millions of CPU endpoints, is pushing the boundaries of traditional data center infrastructure. Physical space and power consumption limitations within individual facilities are making it impractical to house these massive systems in one location.

This necessitates distributing AI compute resources across multiple data centers, which can be in different buildings, campuses, or even distinct geographical regions. Researchers from The Ohio State University discussed this trend at Supercomputing 2026 (SC26), emphasizing the need for new networking solutions.

Challenges of Massive AI Systems

Modern AI clusters place significant demands on network bandwidth, latency, and congestion management. While accelerator performance continues to advance, communication capabilities have not kept pace, meaning faster GPUs do not always translate to proportional application performance gains. The efficiency of the network infrastructure directly impacts how effectively large-scale AI systems utilize their computational power.

Evolution of Interconnect Domains

Historically, high-performance computing has relied on two main interconnect types: scale-up and scale-out. Scale-up networks (e.g., AMD Infinity Fabric, Intel UPI, Nvidia NVLink) connect components within tightly integrated systems. Scale-out networks (e.g., Ethernet, InfiniBand, HPE Slingshot, Ultra Ethernet) link individual servers or racks into larger clusters, typically within a single data center.

Emergence of Scale-Across Interconnects

The scale and power requirements of contemporary AI clusters are driving the need for a third interconnect domain: scale-across. Unlike scale-out networks confined to a single data center, scale-across architectures extend high-performance connectivity between distinct physical locations. This approach allows operators to aggregate computational resources from various sites to meet the demands of large AI workloads.

✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →

The daily brief

One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.

One email a day. Unsubscribe in one click, any time.

Today's brief

Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.

~4 min · 3 stories · Nov 21

▶ Play today's brief Listen on Spotify

New every morning, and the back catalogue is archived by date.

Reporting from

The increasing size and power demands of AI clusters, often exceeding 100,000 accelerators, are forcing their distribution across multiple data centers and geographical regions. This shift creates a demand for a new networking architecture, termed 'scale-across' interconnects, to efficiently connect these distributed compute resources.