莱纳·波普Reiner Pope

Worked on large-model serving at Google, where the bottleneck was never abstract, and left in 2024 with Mike Gunter to found MatX on a narrow premise: build a chip that does nothing but run and train large language models, and give up everything a general-purpose GPU spends area on. Raised a Series A above $100M led by Spark with Jane Street, Nat Friedman and Daniel Gross, then a $500M Series B in February 2026 co-led by Leopold Aschenbrenner's Situational Awareness and Jane Street, taking MatX past $600M raised. The forthcoming MatX One is claimed to beat any other chip on throughput while holding the lowest latency.