{"changes":[],"country":"US","description":"MatX designs accelerator chips, marketed as MatX One, purpose-built for training and serving large language models, with weights in SRAM and HBM for KV cache. It targets frontier AI labs running training, RL, and inference at hundred-thousand-chip scale. The architecture pushes the highest FLOPS per square millimeter and clears 2,000 output tokens per second on huge MoE models.","domain":"matx.com","employees":null,"founded":2022,"fundingTotalQuarantined":false,"githubOrg":"matx","githubStars":null,"industry":"infra","investors":[],"lat":39.78373,"lng":-100.445882,"location":"USA","name":"MatX","rounds":[{"amountQuarantined":false,"amountUsd":80000000,"currency":"USD","date":"2024-11-22","investors":[],"leadInvestor":null,"source":{"fact":"round","publisher":"techcrunch","retrievedAt":"2026-06-02","status":"attributed","url":null},"status":"attributed","type":"series-a","url":null}],"slug":"matx-com","sources":[{"fact":"round","publisher":"techcrunch","retrievedAt":"2026-06-02","status":"attributed","url":null},{"fact":"existence","publisher":"matx.com","retrievedAt":"2026-06-02","status":"self-reported","url":"https://matx.com"},{"fact":"description","publisher":"matx.com","retrievedAt":"2026-06-02","status":"self-reported","url":"https://matx.com"}],"stage":"series-a","techStack":["hsts","vercel"],"totalRaisedUsd":105000000,"updatedAt":"2026-06-02"}