Moffett AI closed a second funding round of nearly RMB 1 billion within five months, backing its dual weight-and-activation sparsity algorithm and dedicated inference chip line. The piece argues that MoE-style conditional computation, exemplified by a 1.6-trillion-parameter model activating only ~49 billion parameters per token, is pushing sparse computing to the center of AI chip design where general-purpose GPUs cannot deeply co-optimize for sparsity patterns.
Termsslug: activation-sparsity
Activation Sparsity
激活稀疏化
Identifying and skipping channels or tokens whose activations contribute negligibly during inference, reducing the data actually
Curated Intelligence
量子位官网T2·Infra
58 pts
量子位 (微信公众号)T2·Infra
62 pts
Chinese chip startup Moffett AI closed a nearly RMB 1 billion Series C+ round, just five months after a similar-sized round. It bets on sparse inference silicon, pairing dual weight/activation sparsity algorithms with algorithm-software-hardware co-design; its S30 card has topped select MLPerf Inference categories three times running.