Mobilint Kicks Off Global Launch of MLA100, Its Next-Gen AI Accelerator, at MWC Barcelona 2025

BARCELONA, Spain - Feb. 28, 2025 - PRLog -- Mobilint, a leading South Korean fabless AI chipmaker, is introducing a breakthrough AI inference solution at Mobile World Congress (MWC) 2025.

The company will showcase its flagship AI chips: ARIES, a state-of-the-art AI accelerator designed for on-premises applications, and REGULUS, a cutting-edge AI system-on-chip (SoC). Live demonstrations will showcase wide-used AI applications, such as live image captioning, large language models (LLMs), real-time object detection, and face recognition.

Higher AI performance with a fraction of the power used by GPUs

As a cost-efficient alternative to traditional GPUs for on-premises AI inference, MLA100 delivers twice the performance at half the cost, achieving a 4× better cost-performance ratio. In addition, with 80 TOPS of compute power at just 25W, MLA100 consumes only 10% of the power required by a GPU with similar performance.

With a universal PCI Express interface, the low-profile card can run AI models in place of GPU across diverse real-world applications, including on-premises servers, chatbots, smart factories, smart cities, healthcare, and robotics.

ARIES: The high-performance AI chip at MLA100's core

MLA100 is powered by ARIES, Mobilint's ASIC-based neural processing unit (NPU).

ARIES has been validated with 300+ models, ensuring compatibility with the latest model architectures, such as transformers, convolutional neural networks, recurrent neural networks, and long short-term memory models. A recent software upgrade enabled support for large language models and multimodal AI.

Powered by ARIES, MLA100 incorporates advanced techniques:

• Reduced off-chip memory access to minimize latency and power consumption.
• Network-optimized dynamic fusion layers for faster AI inference.
• Proprietary dynamic fixed-point technology for computational efficiency.
• Optimized chip architecture for cost-effective manufacturing.

These features make MLA100 an ideal choice for AI applications that require scalable performance.

Universal accessibility with market-ready deployment

MLA100 features a PCIe interface for universal accessibility and is fully compatible with Linux and Windows.

Following extensive proof-of-concept (PoC) testing, Mobilint successfully shipped its first mass-production batch in late 2024.

"MLA100 sets a new standard for local AI inferencing, combining industry-leading performance, energy efficiency, and cost-effectiveness," said Sanghyun Yun, CSO of Mobilint. "We want to prove that AI doesn't have to carry the high power and cost burdens of GPUs."

Attendees can experience the live demo of MLA100 at Booth 7A62-14, Hall 7 of Fira Gran Via, MWC Barcelona 2025.

About Mobilint

Mobilint brings AI closer to everyday life with power-efficient, high-performance AI chip solutions by vertically integrating co-optimized algorithms, software, and hardware.

For more information, please visit https://mobilint.com.