Rebellions Raises $400M to Challenge Nvidia Agent Inference Monopoly
Every time an agent thinks, plans, or acts, it runs inference. And right now, that inference runs almost exclusively on Nvidia hardware. Rebellions, a Samsung-backed Korean startup, just raised $400M in a pre-IPO round at $2.34B valuation to change that.
The money brings total funding to $850M. Lead investors include Mirae Asset Financial Group and the Korea National Growth Fund. The timing is deliberate β Rebellions is targeting the inference market specifically, not training. CEO Sunghyun Park named Meta and xAI as target customers, not hyperscalers like AWS or Azure. He is going after the labs that are building agents, not the clouds that host them.
Alongside the funding, Rebellions launched two products: RebelRack, a production-ready unit of inference compute, and RebelPOD, which clusters multiple racks into scalable inference infrastructure. They have also established entities in the US, Japan, Saudi Arabia, and Taiwan for global expansion, with an IPO planned for second half of 2026.
The agent economy has a hardware dependency problem. When Claude Code thinks about your codebase, when Codex plans its next edit, when any autonomous agent decides what to do next β it all runs on inference chips. Right now that is essentially an Nvidia monopoly. Rebellions bet is that as agent workloads explode, the market needs alternatives, especially ones optimized specifically for inference rather than training.
$400M at $2.34B for an inference-only chip company. That is how big the market expects the agent compute layer to get.
https://rebellions.ai/
← Back to all articles
The money brings total funding to $850M. Lead investors include Mirae Asset Financial Group and the Korea National Growth Fund. The timing is deliberate β Rebellions is targeting the inference market specifically, not training. CEO Sunghyun Park named Meta and xAI as target customers, not hyperscalers like AWS or Azure. He is going after the labs that are building agents, not the clouds that host them.
Alongside the funding, Rebellions launched two products: RebelRack, a production-ready unit of inference compute, and RebelPOD, which clusters multiple racks into scalable inference infrastructure. They have also established entities in the US, Japan, Saudi Arabia, and Taiwan for global expansion, with an IPO planned for second half of 2026.
The agent economy has a hardware dependency problem. When Claude Code thinks about your codebase, when Codex plans its next edit, when any autonomous agent decides what to do next β it all runs on inference chips. Right now that is essentially an Nvidia monopoly. Rebellions bet is that as agent workloads explode, the market needs alternatives, especially ones optimized specifically for inference rather than training.
$400M at $2.34B for an inference-only chip company. That is how big the market expects the agent compute layer to get.
https://rebellions.ai/
Comments