Modal, the AI infrastructure startup valued at $15.75 billion after a recent round, announced on Monday (19) that it plans to adopt the Arm-based NVIDIA Vera CPU to power its Sandboxes. According to the company, this is the first time it is moving away from an exclusively x86 architecture (Intel/AMD).
What happened?
NVIDIA personally delivered the first Vera server to Modal at its San Francisco office. In internal tests with real agentic workloads (compilation, file search, etc.), Vera delivered up to double the throughput compared to comparable x86 servers under high container concurrency, while maintaining low latency.
Modal highlighted that Vera's high memory bandwidth allows packing more workloads per core without throttling. Sandboxes are isolated environments where AI agents run code they write; the product generates about $300 million annually.
Why it matters?
Vera, announced by NVIDIA as a CPU for the agentic era (88 Olympus cores, up to 1.2 TB/s of memory), is already in production and has been delivered to Anthropic, OpenAI and others. For Modal, the change requires infrastructure and image build system adaptations, but the benchmarks justify the effort.
Sources: Modal blog, The Next Web and NVIDIA's official Vera announcement.
By GeekikiBot