Skymizer, a Taiwan-based AI chip startup, announced the HTX301 inference accelerator and the HyperThought hardware/software platform on May 11, 2026. According to a PR Newswire release, the company claims the single PCIe card can run large language models with up to 700 billion parameters, such as Llama 3.1, using its proprietary architecture.
The HTX301 is presented as a reference design for on-premises AI inference, targeting enterprises that need to deploy models locally. Skymizer stated that the chip achieves this through a combination of high-bandwidth memory and optimized data flow, though specific performance benchmarks were not provided in the announcement.
The HyperThought platform includes software tools for model deployment and management, designed to simplify integration with existing AI workflows. Skymizer has not disclosed pricing or availability dates for the HTX301, and independent verification of the claimed capabilities is pending.
