Investors in Moreh, a startup specializing in AI software tool development that optimizes and creates AI models, include Advanced Micro Devices (AMD) and Korean telecommunications company KT. The company, based in Santa Clara and Seoul, recently secured $22 million in a Series B funding round, bringing its total raised capital to $30 million.
Moreh’s flagship AI software, known as MoAI, is comparable to Nvidia’s CUDA but is compatible with popular machine learning frameworks such as Meta’s PyTorch and Google’s TensorFlow. It enables the execution of applications and AI models that were previously limited to Nvidia.
The startup contends that current AI software is suitable for small-scale AI models using only a few graphics processing units (GPUs) but falls short when dealing with more extensive AI infrastructure. As AI matures at an enterprise level, existing IT infrastructure and data architectures are often ill-suited for training AI models, according to recent reports.
Moreh’s AI solutions empower various users, including AI developers, data center operators, and AI chip manufacturers, to create more adaptable AI infrastructure, a crucial factor given the global shortage of GPUs. Gangwon Jo, co-founder and CEO of Moreh, emphasized the flexibility of their AI software, which allows GPUs and other AI chips, such as neural processing units (NPUs), to operate AI models, including large language models like GPT-3 and T5, without requiring any code modifications.
KT has been collaborating with Moreh since 2021 to build an affordable and scalable AI infrastructure powered by AMD GPUs and MoAI software. They claim that Moreh’s offering has demonstrated its effectiveness and has surpassed Nvidia’s DGX in terms of performance speed and GPU memory capacity when developing large language models. Moreh states that AMD’s MI250 Instinct accelerator, supported by the MoAI platform, showed a 116% higher GPU throughput compared to Nvidia’s A100. They also mentioned that their software could reduce the time required to initiate training for large AI models by one-tenth.
Recently, Moreh completed the training of a Korean-language-based large language model with 211 billion parameters, which they plan to release as open source later this year. The startup began generating revenue in 2021 and aims to reach approximately $30 million in revenue by the end of 2023.
Brad McCredie, corporate vice president of data center GPU and accelerated processing at AMD, noted the growing AI software ecosystem supporting AMD’s AI hardware, providing choices for data scientists and AI users. Moreh intends to utilize the funding for research and development, expanding its product offerings, and increasing its workforce, which currently comprises 70 employees. South Korean VC firms Smilegate Investment and Forest Partners, both existing investors in Moreh, also participated in the Series B funding round.