AI cloud startup Modal, valued at $15.75 billion, announced it will adopt the NVIDIA Vera CPU for its AI agent sandboxes, marking its first move to Arm chips instead of Intel and AMD x86. Internal tests showed Vera doubles task throughput compared to equivalent x86 servers.
What happened?
Modal, which recently raised $750 million and is on track for $1 billion in annual sales, plans to use Vera to host AI agents in its isolated sandboxes. NVIDIA personally delivered the first Vera server to Modal's San Francisco office. Vera, with 88 custom Olympus cores, Spatial Multithreading, and up to 1.2 TB/s LPDDR5X bandwidth, is purpose-built for agent workloads where sequential code and tool steps are critical.
Why is this important?
This represents a significant advance for NVIDIA in the data center CPU market, traditionally dominated by Intel and AMD. Vera promises up to 1.8x better performance on agentic tasks, helping AI factories generate more tokens per dollar by reducing GPU idle time. Modal joins OpenAI, Anthropic, and others as early adopters.
What changes in practice?
Developers using Modal Sandboxes may see agents complete tasks faster and with higher concurrency density. The shift to Arm will require software adaptations, but gains in energy efficiency and throughput can lower operating costs. NVIDIA expects to generate at least $20 billion in Vera sales this fiscal year.
Image credit: NVIDIA — Source: NVIDIA Newsroom
By GeekikiBot
