DeepX, which focuses on developing on-device AI semiconductors, has presented its vision for the future by declaring a challenge to implement ultra-large AI on on-device AI chips, following its global spotlight at CES 2024.

DeepX, an AI semiconductor fabless company, announced on the 6th that it is developing the concept of 'Federated Operation of LLM' as a technology that will enable the commercialization of generative AI, and that it plans to participate in MWC 2024 to be held at the end of February to expand partnerships with global partners.

It was reported that the company is currently working with domestic and European mobile carriers and global data center companies to promote compatibility and optimization with network and cloud systems.

DeepX is reported to have won three CES Innovation Awards for its first-generation on-device AI semiconductor technology at CES 2024, attracting over 5,000 visitors to its booth. In addition, the number of customers for EECP, our customer technology cooperation program, has doubled, reaching approximately 70 global cooperation companies.

In the future, DeepX announced that it will develop a technology to jointly drive the server's super-large AI and the on-device's large AI, and will present an on-device AI semiconductor that can drive the intelligence of a super-large AI in a personalized device with power of several watts in the second half of next year.

Against this backdrop, on-device AI is emerging as a fundamental solution that can popularize not only small-sized AI models but also ultra-large AI such as large language models (LLMs) and generative AI.

Recently, NVIDIA's GPGPU-based solutions have been attracting attention as the most cost-effective solution for LLM services such as ChatGPT and Gemini. However, as the commercialization of large AI models expands, the power energy consumed by GPUs worldwide has reached a level that exceeds the power energy of a single country.

Accordingly, Microsoft, Amazon, Meta, Google, OpenAI, and others are creating their own chips that replace GPUs and are optimized to perform their own AI algorithms.

DeepX emphasizes that its AI semiconductor core technology has the strength of optimizing high performance, low power, and economy when running AI on-device. With these strengths, DeepX defines 'Federated Operation of LLM' as a crucial technology for commercializing generative AI and plans to develop it as a core technology.

DeepX expects that if this technology can complete the collaboration between the server's super-large AI and the on-device's large AI model, it will be possible to reduce energy consumption, carbon emissions, and costs by 10 to 1,000 times compared to relying solely on data centers.

Kim Nok-won, CEO, said, “We will open the AI Everywhere era by fully advancing into the global market through our first-generation product consisting of four AI semiconductors that will be mass-produced starting in the second half of this year,” adding, “Following this, we will develop new technologies that enable AI services at the level of ultra-large AI at 5W or less, provide core technologies in the large-scale AI technology triggered by ChatGPT, and through these, become a comprehensive AI semiconductor company that leads the global market.”