Moreh, a startup based in Santa Clara and Seoul, has secured $22 million in its Series B funding round, with investors including Advanced Micro Devices (AMD) and the Korean telecommunications company, KT. This brings the total funds raised by Moreh to $30 million. The company specializes in developing an AI software tool known as MoAI, which optimizes and creates AI models. MoAI is designed to work seamlessly with existing machine learning frameworks, such as Meta’s PyTorch and Google’s TensorFlow, enabling applications and AI models that were previously limited to Nvidia’s platform to run on it.
The startup recognizes that current AI software is suitable for smaller AI models using only a few GPUs but is insufficient for larger AI infrastructure. As AI continues to mature on an enterprise scale, the IT infrastructure and data architectures of companies are often ill-suited for training AI models. Moreh’s solutions cater to various users, including AI developers, data center operators, and AI chip manufacturers, offering them more flexible AI infrastructure. This is particularly important given the global shortage of GPUs.
Moreh’s AI software is compatible with GPUs and other AI chips like neural processing units (NPUs) and allows for the operation of AI models, including large language models (LLMs) such as GPT-3 and T5, without requiring any code changes. NPUs are specialized chips designed for accelerating machine learning algorithms, while GPUs are commonly used for supercomputing, AI, machine learning, and extensive data analysis requiring high numerical precision.
KT has been collaborating with Moreh since 2021 to create a cost-effective and scalable AI infrastructure using AMD GPUs and MoAI software. KT claims that Moreh’s technology outperforms Nvidia’s DGX in terms of performance speed and GPU memory capacity when developing LLM and other AI models. AMD’s MI250 Instinct accelerator, powered by the MoAI platform, has demonstrated a 116% higher GPU throughput than Nvidia’s A100, according to Moreh. Additionally, the startup claims that its software can reduce the time needed to initiate training for large AI models by a significant margin.
Moreh has recently completed the training of a Korean-language-based LLM with 211 billion parameters, and it plans to release this model as open source later in the year. The startup began generating revenue in 2021 and aims to reach approximately $30 million in revenue by the end of 2023.
AMD’s corporate vice president of data center GPU and accelerated processing, Brad McCredie, noted that the AI software ecosystem supporting AMD’s AI hardware is growing, providing options for data scientists and other AI users to build models and solutions that drive the growth of the industry.
The newly raised funds will be used by Moreh for research and development, product expansion, and the recruitment of additional staff, as the company currently has 70 employees. South Korean venture capital firms, Smilegate Investment and Forest Partners, who are existing investors, also participated in the Series B funding round.
