Inference service
Collect requirements from TBD Labs, integrate relevant software, and move the service toward production with Mark Jen and infra owners.
Momento brings a compact systems team to build the primitives that make new infrastructure capacity deployable and operable by agents.
Momento is a Series A infrastructure startup building low-latency, multi-tenant cache and inference gateway primitives, powered by a hyper-optimized Rust proxy layer. Momento Cache and Inference Gateway sit in the critical path of large-scale, mission-critical workloads at demanding customers, including Capcom for Monster Hunter launch, Fox for Super Bowl and World Cup workloads, BBC for live stream caching and origin, CBS Sports for caching and personalization, Nubank for the critical path of its US launch, and WBD for anti-piracy enforcement.
The common thread is low-latency, highly available infrastructure that customers trust in production when failure is not an option.








Infrastructure deployed and operated by agents is no longer optional for Meta Core Infrastructure's 20x growth plan. Momento would sit inside Surupa's Core Infrastructure org with a clear charter: make that transition safe, fast, and predictable.
Before Momento, members of this team helped define new cloud primitives. They brought NASA and Intuit to public cloud, built, operated, and scaled DynamoDB, turned live broadcast video into an AWS business, and helped create Lightstep, the birthplace of OpenTelemetry.
At Momento, the team applied lessons from running DynamoDB's large multi-tenant fleet to the world of caching and inference primitives. The team built Cache, Inference Gateway, and a high-throughput Rust proxy for demanding production workloads, combining capacity planning, tenancy isolation, latency characterization, and distributed-systems operations in the critical path for large customers. Momento also served as a launch partner for Google Cloud's first Arm-based VM offering, Tau T2A.
The same pattern now extends into Valkey. Momento helps customers and cloud partners move to Valkey, organizes practitioner meetups with Amazon and Google, and contributes performance work across vector search, characterization, and RDMA.