Overview
- Modal has begun integrating Nvidia’s Arm‑based Vera servers after internal benchmarks that the company says roughly doubled agent throughput in its agentic sandbox tests.
- The switch from x86 Intel and AMD systems required major software and image‑build changes at Modal to run an Arm‑based stack on its infrastructure.
- Modal will operate Vera in a mixed fleet alongside Intel and AMD hardware so jobs are routed to the processor that runs them best rather than abandoning existing suppliers.
- Independent technical reporting highlighted that Vera sustains throughput under high concurrency by using higher memory bandwidth, producing larger gains as container contention grows in these specific agent workloads.
- The move ties to big commercial stakes: Modal recently raised $750 million and is growing rapidly while Nvidia projects at least $20 billion in Vera sales and lists other major AI customers planning to adopt the chips.