Moffett AI raises $150M Series C extension for sparse AI chips
What's the deal? Moffett AI (墨芯)Dealroom has a profile for this one. Try Dealroom →, a Chinese sparse-computing chip developer, has raised nearly US$137.5M in a Series C extension announced in November 2026. Investors include Renai CapitalDealroom has a profile for this one. Try Dealroom →, Haitang FundDealroom has a profile for this one. Try Dealroom →, Dongheng Petroleum, Biqiu Capital, Paila Capital, and Zhongcai CapitalDealroom has a profile for this one. Try Dealroom →.
What's the endgame? The company builds sparse inference chips through a combined "algorithm-software-hardware" design, aiming to supply domestic computing power across industries. It will use the funds to upgrade its chip stack, fund next-generation research, and build out a sparse-computing industry-academia alliance.
Why now? The round lands just five months after Moffett closed a previous raise of nearly US$137.5M, taking its two-round total to close to US$275M. Demand for sparse computing is rising as mixture-of-experts (MoE) architectures spread, pushing inference workloads toward conditional, sparser computation.
By the numbers: Moffett reported several-fold revenue growth in the third quarter of 2026, though it did not disclose figures. In one test with Peking University's Shanghai research institute, its sparse technology lifted inference throughput by up to 2.88 times and compressed models by up to 7.64 times with little accuracy loss.
What else? Moffett says its inference chips are deployed across four regions in China and that it has partnered with a leading telecom operator to fold its sparse inference into that operator's compute network. A new board, the SP200, is due before year-end, alongside an upgraded software stack called SECA.
The signal: The raise is one of several large rounds flowing into Chinese inference-chip startups, as investors bet on growing AI inference demand and domestic alternatives to foreign silicon. The open question is whether sparse computing can deliver stable performance and cost advantages under real customer loads — and convert that into durable revenue.
Read more: sohu.com
Image credit: .RGB.
