Energy efficiency has become the new battlefield for enterprise artificial intelligence. While giants like NVIDIA dominate the market with high-performance but also high-consumption chips, a new generation of hardware is proving that it is possible to run complex inference models with a fraction of the required energy. FuriosaAI's landing in European data centers marks an important milestone in this transition. The South Korean startup has begun deploying its line of RNGD accelerators (pronounced "renegade") at Equinix facilities in Lisbon, a move that responds to the growing demand for sovereign computing on the continent.
FuriosaAI, founded in 2017 by June Paik and Hanjoon Kim, has opted for a tensor contraction processor (TCP) architecture manufactured on TSMC's 5 nm process. Each PCIe card offers 48 GB of HBM3, 1.5 TB/s of bandwidth, and 512 teraFLOPS in FP8, with a consumption of just 180 W. This provides a decisive competitive advantage: compared to a competing GPU that triples consumption, Furiosa's chip allows building servers with eight accelerators totaling 3 kW, with 384 GB of HBM, enough for large models like LG Exaone 236B or Qwen 3-30B-A3B, all in air-cooled systems that integrate without modifications into any traditional rack.
Beyond the numbers, the context is what matters. The European Union is promoting data sovereignty and local computing capacity, especially for artificial intelligence applications that require low latency and regulatory control. This opens opportunities for efficient startups, but also demands a mature software ecosystem. Furiosa is working with Broadcom on a third generation of accelerators that will use HBM4 and Ethernet switching to scale beyond eight-node systems. However, the real challenge is not just the silicon, but how companies integrate these capabilities into their actual workflows.
This is where knowledge in artificial intelligence for businesses becomes indispensable. Having efficient hardware is not enough; it is necessary to orchestrate models, manage data pipelines, and ensure that each inference is performed with adequate levels of cybersecurity and regulatory compliance. At Q2BSTUDIO, we accompany organizations on this path, developing custom applications and custom software that leverage both cloud infrastructure (whether with AWS and Azure cloud services) and specialized accelerators. Additionally, we implement AI agents that automate decision processes, and business intelligence services with Power BI to turn inference results into actionable dashboards. The energy efficiency of hardware only translates into real savings if the software is optimized to exploit it, and that is precisely the value we provide from design to production deployment.
The arrival of FuriosaAI in Europe is not a technological anecdote, but a sign that the inference market is maturing towards more sustainable solutions adapted to local sovereignty. For companies that want to get ahead, combining efficient hardware with a well-defined AI strategy for businesses makes the difference between a simple pilot and productive large-scale adoption.
