Neural Magic builds a software platform for deep learning inference that delivers GPU‑class performance on commodity CPU hardware. The company develops algorithms and tools that place software, rather than specialized hardware, at the center of ML infrastructure. Spun out of MIT by founders Nir Shavit and Alexander Matveev and led by CEO Brian Stevens, Neural Magic open‑sourced its machine learning tools earlier this year and has seen growing adoption. The company intends to use new funds to expand operations and business reach and to advance its leadership in pure‑software machine learning acceleration. Neural Magic is based in Somerville, Massachusetts. After the Series A, the company has raised $50M to date. Neural Magic builds an inference engine that lets data scientists run deep learning models on commodity CPUs rather than specialized accelerators like GPUs or TPUs. The company announced early access to this first product, designed to run in containerized, microservices environments. Founders Nir Shavit and Alex Mateev developed the approach from lab work on neurobiology data at MIT. Neural Magic says its software can deliver GPU-like speeds on CPUs, reduce costs, and eliminate memory limitations imposed by accelerators. The startup raised a $15 million seed round led by Comcast Ventures, bringing total funding to $20 million after a prior $5 million pre-seed. Seed investors included NEA, Andreessen Horowitz, Pillar VC and Amdocs.

Total raised

$45M

Funding rounds

2

Latest round

Series A

Latest activity

Oct 2021