Section C · Operations

Infrastructure

Lambda's infrastructure footprint is meaningful but smaller than CoreWeave's. Multiple regions, current-generation GPUs, InfiniBand fabric for clusters. Operationally sound rather than headline-grabbing.

Datacenter footprint

Lambda operates from multiple datacenter regions in North America. Capacity has grown substantially through 2023-2026 as funding has supported aggressive buildouts. New regional sites continue to come online.

Total capacity is meaningful but doesn't approach CoreWeave's scale. Lambda is comfortable with this — the customer base doesn't require multi-hundred-MW campus capacity that the largest reservations need.

GPU mix

Lambda's GPU mix:

  • A100 80GB (substantial legacy fleet from earlier deployments).
  • H100 80GB (the bulk of current-generation deployment).
  • H200 (growing share).
  • B200 (coming online with allocation availability).
  • Limited L40S for inference workloads.

Consumer cards aren't part of Lambda Cloud — the cloud product is datacenter-grade only. Workstation hardware (separate business line) covers the consumer-card market for Lambda customers who want that form factor.

Network

Within clusters:

  • NVLink + NVSwitch for in-node GPU communication.
  • InfiniBand HDR / NDR for inter-node communication in cluster configurations.
  • Standard datacenter Ethernet for control plane.

The fabric is sufficient for cluster sizes Lambda targets — tens to hundreds of GPUs in a training cluster. Very large clusters (thousands of GPUs) go to CoreWeave or Crusoe.

Operations

Lambda's operational team includes:

  • Datacenter operations across sites.
  • Network engineering.
  • Customer success / technical account management for larger deployments.
  • Engineering teams maintaining the platform software.

The operational quality is enterprise-grade. Multi-year customer relationships demonstrate this.

Scaling

Lambda's capacity scaling is driven by customer demand and capital availability. The pace through 2024-2026 has been aggressive, with new regions and capacity additions in step with funding rounds. Not as fast as CoreWeave's growth but meaningful.

Takeaway

Lambda's infrastructure is mid-tier enterprise — competent, sufficient for the customer base, not built for the multi-billion-dollar marquee reservations. The next chapter examines who actually uses it.