AI infrastructure startup Modal says it plans to start using Nvidia's Vera CPUs for the first time for hosting agents, alongside its current Intel and AMD chips
Context & Ripple Effects
Vera’s AI-data-center push has progressed from early benchmarks against Intel and AMD CPUs to deployments: Nvidia said Anthropic, OpenAI and SpaceX were among its first major users, while AM Intelligence placed a large system order. Modal adds a cloud-platform use case centered on agent hosting rather than a single buyer’s internal infrastructure.
Nvidia’s Vera team has argued that agentic AI is renewing the importance of single-threaded CPU performance. Modal’s mixed-vendor approach puts that design claim into a commercial infrastructure setting where Intel and AMD remain part of the fleet.
First-order effects
- Modal adds Nvidia Vera to the CPU options behind its agent-hosting service while retaining Intel and AMD chips, broadening the hardware configurations it can offer customers.
- Nvidia gains Modal as a platform customer for Vera, extending adoption beyond the named early users cited by Nvidia.
Second-order effects
- Modal can allocate agent workloads across Nvidia, Intel and AMD systems based on observed performance, making CPU selection a more explicit service-capacity decision.
- Intel and AMD face a new competitive comparison point inside Modal’s infrastructure, where agent-hosting performance—not CPU presence alone—can influence future capacity allocation.
Third-order effects
- If more AI infrastructure operators maintain mixed CPU fleets, CPU vendors will increasingly compete on workload-specific agent throughput and platform integration rather than on general-purpose server positioning.
- The move supports a heterogeneous AI-compute model in which cloud operators assemble specialized CPU, GPU and software combinations for distinct AI workloads.
The trend: AI infrastructure providers are turning agent hosting into a heterogeneous-compute service, using specialized CPUs alongside established x86 fleets to tune capacity for workload behavior.