Infrastructure for teams that cannot afford to be slow.

Momento was founded by the people who ran DynamoDB and built the AWS media vertical. We operate caching and data-plane infrastructure in production for some of the most demanding workloads on the internet, and we are pointing that experience at LLM inference.

DynamoDBRunning all engineering for one of the world's largest multi-tenanted fleets, plus chief of staff to Charlie Bell

AWS ElementalBuilt the media vertical, all seven AWS Media Services

LightstepCofounded OpenTelemetry, launched the Metrics product from scratch into a multi-million-dollar revenue stream

Database caching, already in production.

These teams run their caching on Momento today, across media, gaming, fintech, and commerce. Our forward-deployed engineers sit alongside our customers' engineering teams during critical events, which is where we continuously gain operational experience scaling massive production fleets and optimizing utilization.

Our KV cache product

Rasa. KV cache in 56µs.

An early preview, built in collaboration with AWS: an ultra-low-latency KV cache for LLM inference. Blocks live in registered memory on EFA data nodes and clients read them directly with one-sided RDMA, so prefixes you have already served come back over the wire instead of being recomputed on a GPU.

56µsp99 to read a 1 MiB chunk

Tbps+Across a fleet, at a measured 201 Gbps EFA line rate per data node

Near idle CPUThe NIC serves the reads.

Rasa deep dive Engineering blog

Where this goes next

How we work together.

We want to partner with Fireworks on the KV cache performance problem. You have already built most of it: disaggregated KV caching, prompt-aware routing, plus prefill and decode on separate pools. The gap we fill with Rasa sits underneath all that, transporting KV cache chunks in sub-100µs from remote DRAM, extending cache retention from minutes to days.

Our engineers would do all the integration work alongside your team, implementing directly against FireAttention and your KV cache manager. The same engineers are also distributed systems experts, and are useful beyond Rasa. Keeping caches fast for BBC, Fox, and Nubank through critical events exercises the same muscle a multi-tenanted fleet needs for optimizing utilization, capacity planning, and debugging incidents.

Next steps

Amazon and Momento are working together on a Valkey KV cache launch at re:Invent. The project will be open sourced under an Apache license, and we are seeking one additional partner to highlight at the launch, and we would love to work with Fireworks.

Team highlights.

Founders and key engineers from AWS, DynamoDB, Elemental, and Lightstep.

Founders

Khawaja Shams

Co-founder / CEO

As VP of Engineering at AWS Elemental, built the AWS media vertical, owned product and engineering for all seven AWS Media Services, scaled the organization after the Elemental acquisition, and helped establish Portland as a major AWS engineering site. Previously led DynamoDB at AWS, served as chief of staff to Charlie Bell, then AWS's senior infrastructure leader, and received NASA's Early Career Medal for contributions to the Mars Rovers. Recognized across the Valkey community for rigorous systems work and open-source leadership.

Daniela Miao

Co-founder / CTO

As one of Lightstep's first ten employees, served as Director of Engineering and launched Metrics support, improving revenue by 20% and helping create the path to a successful exit. At AWS DynamoDB, owned developer ecosystem work across SDKs, integrations, and common software paths, and helped release cross-region replication. Brings deep systems experience across performance, observability, security, and engineering execution.

Core leadership

Jason LaPier

VP Engineering

Led engineering for Elemental MediaStore, the high-performance object store built for media companies, then worked on AWS MediaTailor for personalized ad insertion. At Momento, works closely with strategic customers including Paramount on personalization engines built on Momento. Drives engineering execution, quality, architecture, and delivery across core product teams.

Ellery Addington-White

Founding Engineer

Ran AWS integration for TurboTax at Intuit and managed thousands of accounts across Intuit's AWS organization. As a founding engineer at Momento, has worked closely with each of the company's largest customers, rapidly prototyped new product ideas, and helped turn customer needs into production capabilities. Currently working on Momento's inference stack.

Kenny Chamberlin

Principal Engineer

Specialist in large-scale mission-critical systems, from DynamoDB Streams to Elemental MediaTailor, where he was lead engineer for personalized video ad insertion in live video events. At Momento, has been the primary architect for core services and production performance work. Currently leading RDMA support for Valkey in collaboration with AWS.

Michael Landis

Principal AI Scientist

Owns Momento's AI and inference-adjacent product work, including semantic caching, embedding-based retrieval, and the data-path patterns that connect low-latency infrastructure to production AI workloads. Previously at Parity Computing, building ML entity resolution for scientific content before its 2019 acquisition by Elsevier.