Korean chipmaker FuriosaAI ramps Renegade to take on Nvidia

AI chip startup FuriosaAI unveiled its second-generation Renegade chip for AI inference, and said it began mass production in January.

The chip has a thermal design power of around 180 watts, uses a 5-nanometer TSMC process with 40 billion transistors, pairs SK hynix HBM3 memory, and claimed benchmarking with overseas clients showed up to 7.4x more concurrent users than Nvidia’s RTX Pro 6000 at the same power.

It started development in 2022 and produced early samples in 2024, and has started mass production of about 4,000 units this year, plus plans to expand its software stack, including compilers and SDKs.

🔗 Source: Korea Herald

🧠 Food for thought

Implications, context, and why it matters.

A national AI strategy is backing this Nvidia challenger

FuriosaAI’s launch ties to South Korea’s broader “K-Moonshot” effort to align companies around national missions. Samsung and LG take part 1.
FuriosaAI is working with samsung sds. Samsung sds adds a cloud research team that builds custom AI networking and software to improve performance 2.
Samsung sds already supports Upstage (a South Korean AI startup). Upstage uses Samsung Cloud Platform’s “GPU as a service” offering for model training, which could create early demand for a future inference chip 3.

Targeting cost over peak performance creates a new market lane

FuriosaAI chose the standard PCIe 5.0 interface (a common server expansion connection) instead of a custom high-speed interconnect like Nvidia’s NVLink 4.
That decision can limit performance when scaling to very large models 4. A third-party publication expects the RNGD price near $10,000, though the company has not set pricing, which would sit far below widely reported pricing for an Nvidia H100 5.
The rollout targets inference workloads (running trained models in production) where cost and power use matter 4. It also offers competitors a route to enterprise customers while leaving the biggest training jobs to Nvidia.