Meta acqui‑hires Dreamer team
Meta acqui‑hires the founders and team behind agentic AI startup Dreamer into Meta Superintelligence Labs while reportedly leaving the startup’s IP behind—adding talent for agent orchestration work. At the same time Meta is doubling down on custom AI chips, pushing designs that tightly couple software and hardware for higher throughput. (bloomberg.com) (businessday.ng)
Dreamer’s founding team includes David Singleton (former CTO of Stripe and ex‑VP of engineering for Google Android), Hugo Barra (ex‑Google Android product VP, former head of VR at Meta, and a former Xiaomi executive) and Nicholas Jitkoff (former lead designer on Google Chrome OS). (bloomberg.com) The startup had raised about $56 million at a roughly $500 million valuation in late 2024, according to earlier coverage of the company’s seed/early rounds. (siliconangle.com) Reporting diverges on deal terms: Bloomberg says the arrangement “did not include Dreamer’s technology,” while other outlets say Dreamer remains a separate legal entity that granted Meta a non‑exclusive license and that investors were repaid more than their original investment. (bloomberg.com) (economictimes.indiatimes.com) Dreamer co‑founder David Singleton publicly acknowledged the arrangement in a post on LinkedIn/X, saying he had shown the product to Mark Zuckerberg earlier this year and thanking Alexandr Wang for early support. (finance.yahoo.com) Meta’s internal messaging from Chief AI Officer Alexandr Wang emphasized a renewed “conviction in agents,” describing work on agents that are “personalized and always‑on” and able to integrate across surfaces and wearables, positioning the Dreamer team against that roadmap. (bloomberg.com) Separately, Meta publicly accelerated its MTIA custom‑silicon program — naming MTIA 300, 400, 450 and 500, noting MTIA 300 is already in production and that the company plans four new chip generations within two years while partnering closely with Broadcom on the designs. (about.fb.com) (theregister.com) Meta’s MTIA strategy emphasizes an “inference‑first” stack, a six‑month‑cadence for iterative chip releases, deployment of hundreds of thousands of MTIA devices for recommendations and ads today, and plans to scale to “multiple gigawatts” of capacity as production ramps into 2027. (about.fb.com) (businessday.ng)
