Momento / The team
Momento runs a serverless, multi-tenant data platform across 30+ cells in 15+ regions at over 2 million transactions per second, behind BBC streaming, HBO time-to-first-frame, CBS personalized feeds, and Fox’s Super Bowl metrics, where millions are watching and failure is never an option. The same team now builds the systems around inference, including agent harnesses, evals, gateways, and KV caching.
Teams that run on Momento’s infrastructure
The team has built and operated large-scale, multi-tenant systems across DynamoDB, AWS Elemental, and Momento, from low-latency data services and Rust runtime tuning to Kubernetes operators and control planes.
Our control plane continuously streams topology, health, capacity, and tenant limits to every router across 30+ cells, and the routers act on each update in real time. Traffic shifts on its own the moment capacity saturates, a node degrades, or a zone goes down.
Our reconcilers rediscover the fleet on every tick, make one change at a time, and confirm it took effect before the next. They add capacity before removing it, take destructive steps only when a majority of the fleet agrees, and hold steady for a human when no safe move exists.
Every change moves through cells in waves, from low-traffic to high-traffic to dedicated customer cells, with bake periods gated on live alerts and one-command pipeline blocks. Destructive changes in production require a second approver.
Momento serves over 2 million transactions per second. The team is building KV caching over RDMA on EC2 with AWS for its re:Invent launch, alongside semantic caching and an LLM gateway whose session routing raised prompt cache hit rates from about 30% to about 100%.
Engineers and operators from DynamoDB, AWS Elemental, Meta, and NASA, now building across AI systems, caching, and inference.