Advanced Micro Devices (AMD) and Korean telco KT are among the investors of Moreh, which builds an AI software tool that optimizes and creates AI models. The Santa Clara- and Seoul-based startup said it has raised a $22 million Series B round, bringing its total raised to $30 million.
Moreh says its flagship AI software, called the MoAI, is similar to Nvidia’s CUDA but is compatible with existing machine learning frameworks like Meta’s PyTorch, Google’s TensorFlow, and applications and AI models that previously could only run on Nvidia.
The startup believes existing AI software is suitable for small-scale AI models using only a few GPUs (graphics processing units) but is inadequate for more massive AI infrastructure. According to a recent report, as “AI is rapidly maturing at enterprise scale,” companies’ IT infrastructure and data architectures are “unfit” for training AI models.
Moreh’s AI solutions allow users, including AI developers, data center operators and AI chip makers, to build more flexible AI infrastructure, important given the general shortage of GPUs globally, Gangwon Jo, co-founder and CEO at Moreh, told TechCrunch.
The three-year-old startup’s AI software enables GPUs and other AI chips, such as neural processing units (NPUs), to operate AI models, including large language models (LLMs) like GPT-3 and T5, without any code changes, Jo explained. (NPU is a specialized chip designed to accelerate machine learning algorithms, while GPU is used for supercomputing, AI, machine learning, and extensive data analysis where numerical precision is required.)
KT has been working with Moreh since 2021 to build a cost-effective, scaleable AI infrastructure powered by AMD GPUs and MoAI software.
KT claims that Moreh’s offering proved its technology works and that its performance in developing LLM and other AI models surpassed Nvidia’s DGX in terms of performance speed and GPU memory capacity. AMD’s MI250 Instinct accelerator powered by the MoAI platform showed 116% higher GPU throughput than Nvidia’s A100, according to Moreh, which also said AI developers could reduce a tenth of the total time required to initiate training for large AI models via its software.
Most recently, the startup completed the training of Korean-language-based LLM with 211 billion parameters and will release the model as open source later this year.
Moreh started generating revenue in 2021 and aims to reach approximately $30 million by the end of 2023, Jo told TechCrunch.
“The AI software ecosystem supporting AMD AI hardware continues to grow, providing choice for data scientists and other users of AI as they build the AI models and solutions that will drive the continued growth of this industry,” Brad McCredie, corporate vice president of data center GPU and accelerated processing at AMD, said in a statement.
Moreh will use the money for research and development, product expansion and hiring additional staff, which now has 70 employees. South Korean VC firms Smilegate Investment and Forest Partners, Moreh’s existing investor, also participated in the Series B round.