FuriosaAI, a South Korean AI hardware startup known for power-efficient AI accelerators, has launched its RNGD AI inference cards in Equinix’s Lisbon datacenter, marking its expansion beyond domestic markets into Europe’s sovereign AI compute space.
- RNGD AI accelerators deployed at Equinix Lisbon datacenter
- Cards offer competitive AI performance with lower power consumption
- Next-gen chips with Broadcom collaboration planned for scalability
What happened
FuriosaAI, a South Korean startup specializing in AI inference chips, has started deploying its RNGD line of AI accelerators in the European market. The first known deployment is at Equinix's LS2 datacenter in Lisbon, Portugal, marking the company’s first significant push outside its home market. These accelerators feature 48 GB of HBM3 memory with 1.5 TB/s bandwidth and deliver up to 512 teraFLOPS of dense FP8 compute performance while maintaining a modest 180-watt power envelope.
Founded in 2017 by June Paik and Hanjoon Kim, FuriosaAI began with domestic successes including partnerships with LG Electronics. The company’s current RNGD systems are air-cooled, rack-compatible, and feature up to eight accelerator cards per server. FuriosaAI sees Europe as an opportunity to supply sovereign AI compute infrastructure that demands both efficiency and capacity.
Why it matters
FuriosaAI’s entry into European datacenters challenges established AI accelerator vendors by offering competitive performance with a much lower power footprint. While its current RNGD chips don’t match Nvidia or AMD in raw capability, the efficiency gains allow easier integration into existing datacenter racks without significant cooling upgrades. This positions FuriosaAI as a contender in a market increasingly sensitive to power and thermal costs.
Moreover, FuriosaAI's upcoming generation of accelerators, developed in partnership with Broadcom, promises substantial improvements. These future chips will employ cutting-edge HBM4/4e memory and Broadcom’s advanced networking tech, enabling larger scale-up clusters beyond the current eight-card systems. This roadmap aligns with the evolving needs of large AI models that require high compute density and connectivity.
What to watch next
Observe FuriosaAI’s progress in European datacenter deployments beyond Lisbon, including potential partnerships with cloud and colocation providers. How quickly the company gains traction in this competitive market will indicate the demand for power-efficient alternatives to dominant US-based chipmakers.
Further developments from FuriosaAI and Broadcom’s collaboration should be monitored closely. The launch timeframe and performance benchmarks of the third-generation AI accelerators will be important in determining FuriosaAI's ability to scale up performance and compete with incumbents on both efficiency and raw power.