[Edaily Reporter Yun Junghoon ] #Nota Inc., a company specializing in optimizing AI inference, announced on the 12th that it has signed a follow-up contract with semiconductor startup HyperExcel to supply large language model (LLM) optimization technology. This marks the second consecutive year the company has secured an order since the initial contract in September 2025.

Under this contract, Nota Inc. will supply model and computational optimization technologies for major LLMs to be integrated into HyperExcel’s “Bertha LPU (LLM Processing Unit),” a semiconductor chip dedicated to LLM inference. By expanding its optimization scope—which had previously focused on GPUs, CPUs, and NPUs—to include LPU architecture, Nota Inc. is strengthening its software business to address the diversification of chip architectures.

The two companies plan to combine Nota Inc.’s optimization technology with HyperAccel’s hardware and development environment to significantly expand the range of LLMs that can be run on Bertha. In the data center and enterprise AI markets, key factors driving adoption decisions now include not only a chip’s computational performance but also whether it supports specific models required by enterprises and its execution efficiency.

Nota Inc. will perform optimizations based on HyperAccel’s SDK (HyperAccel SDK) and the embedded domain-specific language (eDSL) “Legato” through its AI optimization platform, “NetsPresso.” In particular, the company plans to modify the kernel—which handles the core computations of large language models (LLMs)—to align with the LPU architecture and apply quantization techniques tailored to specific model characteristics to maximize memory efficiency.

This collaboration demonstrates that Nota Inc.’s business scope is expanding beyond simple AI model lightweighting into the areas of semiconductor-specific execution environments and computational kernel optimization.

Kim Ju-young, CEO of HyperExcel, stated, “As the LLM inference market grows rapidly, the flexibility to run various models in real-world service environments has become a core competitive advantage for semiconductors,” adding, “Through our collaboration with Nota Inc., we will broaden the range of models supported by the Verda LPU and enhance our responsiveness to market demands.”

Chae Myeong-su, CEO of Nota Inc., remarked, “Success in the AI semiconductor market depends not only on the hardware itself but also on the software ecosystem that enables the rapid execution of models desired by customers.” He added, “This follow-up contract is an example of how technical trust with existing customers has led to additional orders, and we will continue to expand our collaboration with global AI semiconductor companies.”