Rebellions Takes On Nvidia in AI Inference Chips

Rebellions, a Korean chipmaker worth over $2.3 billion, is setting its sights on Nvidia in one particular area: AI inference, the phase where a trained model processes a request. Established in 2020, it’s received $850 million in funding, is valued at over $2.3 billion, and is going global. “We are here to create one of the rebellions against Nvidia, which is one of the leaders in the AI space,” said Rebellions’ chief business officer Marshall Choy on the Targeting AI podcast.

However, Rebellions doesn’t want to compete with Nvidia across the board. It started by only targeting inference, ignoring the training aspect of AI (which is where Nvidia’s GPUs shine). “By focusing on the inferencing side of things, to complement the existing GPUs that people have today doing training,” said Choy. This strategy seems to have paid off, given the rise of agentic AI, as these autonomous agents rely on inference to determine their actions, making the price and performance of deploying models in the real world paramount.

1A different chip design

Rebellions uses ‘chiplets’ instead of a single large chip. An AI chip is broken up into smaller pieces, each of which contains compute and memory and is highly configurable to fit a certain task (e.g., a call center or medical system), much like a neural processing unit optimized for specific tasks. Additionally, Rebellions optimizes for mixture-of-experts and multimodal models instead of just large language models, and it is closely aligned with the open-source PyTorch ecosystem. It’s a major player in South Korea’s ‘K-Nvidia,’ a project worth billions of dollars aimed at developing AI and chip sovereignty.

2Why this matters for you

Out of sight, out of mind. When a chatbot responds to your question or when an AI agent completes a task, that’s typically done through inference. Today, most of that inference happens on Nvidia hardware, which makes it expensive and creates a monopoly. Having a strong contender in the form of a cheaper, more efficient inference engine will force down API costs and reduce the risk of vendor lock-in for AI products. I believe that the true battle in AI is shifting from “who can train the largest model” to “who can deploy it cost-effectively,” and Rebellions has chosen to do battle there. Whether a startup worth $2.3 billion can make a dent in a much larger corporation remains to be seen, but for anyone who has a usage-based pricing model, having another player in the inference game is a win.