FuriosaAI Brings Nvidia-Rival AI Chips to Europe
▼ Summary
– South Korean startup FuriosaAI has activated its RNGD AI accelerators at an Equinix datacentre in Lisbon, marketed as a cooler, cheaper alternative to Nvidia.
– Each RNGD accelerator uses a 5nm Tensor Contraction Processor, delivering 512 teraFLOPS of FP8 at a 180-watt thermal profile, requiring no liquid cooling.
– The deployment targets European enterprises seeking efficient, locally sourced AI compute amid rising energy costs and efforts to reduce reliance on American silicon.
– FuriosaAI is developing a third-generation accelerator with Broadcom using HBM4 memory, while RNGD is already in mass production with over $250 million in funding.
– The chip is not designed to outperform Nvidia but to offer a cost-effective, low-power option for buyers wanting sustainable inference outside the United States.
A South Korean chip startup is betting that European enterprises want a cooler, more cost-effective alternative to Nvidia. Its first RNGD AI accelerators, pronounced “renegade,” have just gone live in a Lisbon data center, marking a strategic push into the region.
FuriosaAI announced Tuesday that it has installed RNGD servers at Equinix’s LS2 datacentre in Lisbon. The deployment builds on the company’s existing presence there, which includes a compiler-focused R&D lab and a new flagship office. The timing is no accident. The reveal coincides with the RAISE Summit in Paris, as European businesses increasingly seek efficient AI compute that can be sourced locally, rather than relying on American hardware.
The core of Furiosa’s pitch is efficiency over raw speed. Each RNGD accelerator uses a 5nm Tensor Contraction Processor that delivers 512 teraFLOPS of FP8 while holding to a strict 180-watt thermal profile. Eight of these chips form the NXT RNGD Server, a 3kW system that fits into standard racks without requiring liquid cooling or retrofits. The company markets it as a dense, air-cooled inference engine.
The efficiency gap is the key differentiator. As The Register notes, Nvidia’s closest competitor, the RTX Pro 6000, offers double the memory and compute but draws more than three times the power. “We unlock the ability for enterprises to run inference sustainably and reliably,” said June Paik, co-founder and CEO of FuriosaAI.
Europe is the logical next step. The continent is racing to build its own AI infrastructure and reduce dependence on American silicon. With energy costs climbing, a chip that sips power and fits existing racks becomes an easier sell. The move is as much about mindshare as it is about sales.
Furiosa is also looking ahead. It is working with Broadcom on a third-generation accelerator designed for frontier models with a trillion or more parameters. That chip will use faster HBM4 memory for hyperscale inference. The current RNGD chip is already in mass production, built on TSMC’s process with SK hynix memory. Furiosa says it has raised more than $250 million to date.
Furiosa will not dethrone Nvidia. Its cards are smaller and slower, and its next chip depends on HBM4 memory that is only now reaching the market. But that is not the bet. Some European buyers simply want a chip they can run cheaply, cool quietly, and buy outside the United States. Furiosa joins a crowded field of Nvidia challengers, all trying to prove there is room for more than one name in AI compute.
(Source: The Next Web)