Artificial intelligence (AI) semiconductor startup Rebellions said on the 11th that it will unveil a "full-stack enterprise AI platform" with global open-source solution company Red Hat.
The newly unveiled "Red Hat OpenShift AI" runs on Rebellions' Neural Processing Unit (NPU). Red Hat OpenShift is a Kubernetes-based enterprise platform that helps corporations manage the life cycle of AI models at scale. By combining Rebellions' NPU with the high-efficiency inference engine vLLM, it offers a proven full-stack enterprise AI platform. vLLM is software that helps run large language models (LLMs) quickly and efficiently.
"Red Hat OpenShift AI based on Rebellions NPU" integrates core components to optimize AI inference, covering every area required for inference, from hardware (NPU) to model serving (vLLM).
The company said, "As AI adoption expands, corporations must address a range of challenges at the same time, including infrastructure expense, deployment complexity, and security," and added, "This solution reflects those practical needs and is expected to offer an alternative for building stable and efficient inference infrastructure across diverse environments."
In particular, "AI inference," the stage where a trained model generates responses in real-world environments, becomes more expensive as usage increases. It also affects performance. As a result, the existing graphics processing unit (GPU)-centric approach is finding it hard to meet all demand. Rebellions expects that this platform, developed in partnership with Red Hat, will ease corporations' concerns about AI adoption thanks to its efficiency.
Rebellions' NPU is designed with an architecture optimized for AI inference, delivering up to 3.2 times higher energy efficiency than existing GPUs. The company says it provides a development environment as convenient as GPUs through full-stack software and support for major open-source AI frameworks.
The two companies, together with Kolon Benit, will hold a seminar to introduce the solution and discuss key challenges faced during corporations' adoption of Generative AI. They then plan to support efficient AI infrastructure deployment through consulting and proofs of concept for participating corporations and to kick off full-fledged market expansion.
Park Seong-hyeon, CEO of Rebellions, said, "As AI serving and inference ramp up, corporations need practical infrastructure that satisfies performance, expense, and Data Sovereignty," and added, "Through this collaboration, Red Hat and Rebellions offer an integrated and validated inference platform using open source, replacing the existing approach where elements of AI inference were all separate, from hardware to model serving." Park continued, "This will be the first case to present a new alternative for NPU-based inference infrastructure beyond GPU-centric environments."
Brian Stevens, senior vice president (CTO) of Red Hat's AI division, said, "The future of enterprise AI must allow choosing architectures that go beyond a single-structure proprietary stack," and added, "Collaboration with Rebellions is meaningful in implementing Red Hat's 'all models, all accelerators, all clouds' strategy."