AI chip and developer platform using multiplication-free arithmetic to dramatically reduce training and inference costs.
Baud has designed an ASIC and corresponding arithmetic representation that eliminates multiplications from neural network forward and backward passes, replacing them with additions to pack an order of magnitude more compute and memory into the same die area. A PyTorch-compatible compiler converts exportable models (Qwen, DeepSeek, GLM, Flux, Wan, and others) to the Baud format with bit-exact results in most cases. The company is offering early cluster access for pretraining, fine-tuning, RL post-training, and inference workloads while the platform matures. Baud is a product of Baud.
For people
For agents
Nothing listed yet.