Baud

What it does

Baud is developing a novel AI chip and developer platform designed to dramatically lower the cost and energy required to train, fine-tune, and deploy frontier AI models. The company has created an arithmetic representation that eliminates multiplications during both forward and backward passes, compressing neural networks without loss. This is paired with an ASIC architecture that leverages this representation, trading multiplications for additions to pack more compute and memory into the same die area. The result is a chip that is faster, more power-efficient, and simpler to build than incumbents at the same process node and memory bandwidth.

Who it is for

Baud targets businesses of all sizes that want to own their artificial intelligence—training and serving their own models rather than relying on a handful of large providers. The platform is designed for workloads such as pretraining, fine-tuning, RL post-training, and inference. It is particularly suited for multimodal models (vision, language, audio), world models (learned simulators for planning and control), and media generation (image, video, audio synthesis).

Why it matters

Current AI compute is prohibitively expensive, concentrating frontier intelligence in the hands of a few companies. Baud claims that existing silicon and software operate orders of magnitude above the theoretical minimum required to create and serve the same intelligence. By approaching the efficiency of the human brain, Baud aims to democratize access to AI, enabling more organizations to participate in AI development and discovery.

Launch signal

Baud has demonstrated a small model trained on their ASIC emulated on an FPGA, running inference at 1000+ tokens per second on a single FPGA. They are opening early access to their first cluster, allowing partners to experience accelerated training and inference on a multiplier-free ASIC. Applications for early access are open via their website.

Brand and naming

The name 'Baud' is short and tech-forward, evoking data transmission rates (baud rate), which aligns with the chip's focus on efficiency. However, it lacks direct semantic connection to AI or chips, potentially confusing memorability. The domain baudlabs.ai reinforces the AI focus. The tagline 'AI Training Chips and Developer Platform' clearly positions the brand in the hardware+software AI infrastructure space, but the name itself doesn't convey this, relying heavily on the tagline for clarity.