# FuriosaAI Brings Its Nvidia Challenger Chip to a Lisbon Data Center > Source: > Published: 2026-07-11 14:51:43+00:00 *FuriosaAI has put its Renegade inference chips inside Equinix's LS2 data center in Lisbon, and the useful part isn't the symbolism. It's that Europe now has another working option in a market Nvidia still dominates.* South Korea's FuriosaAI has installed its RNGD accelerators, nicknamed Renegade, at Equinix's LS2 data center in Lisbon. Data Center Dynamics reported this week that the deployment is the company's first commercial rollout in Europe. It also comes with a local base. FuriosaAI is setting up a Lisbon office for commercial operations, research, compiler work, chip design, and PCB design, building on a smaller engineering team it already had running there. Renegade is built on a 5-nanometer Tensor Contraction Processor architecture, a design that treats tensor contraction as a native operation instead of breaking it into a chain of matrix multiplications. Each card delivers 512 teraflops of FP8 compute, carries 48GB of HBM3 memory, and moves data at 1.5TB per second. A single server holds eight cards and draws 3 kilowatts. That number matters. FuriosaAI says the server can run in a standard air-cooled data center, without a liquid-cooling retrofit. For an operator like Equinix, that's the difference between putting a machine into service and turning the building into a plumbing job. You don't retrofit a building for Renegade. You plug it in. ## Renegade already has a customer test FuriosaAI isn't a new name pitching a slide deck. LG AI Research adopted RNGD in July 2025 to run its EXAONE foundation models, and KED Global reported that LG saw 2.25 times better inference performance per watt than the GPUs it replaced. That's the kind of validation a small Seoul chip company needs before it can ask a data center operator in Europe to take it seriously. It also explains why the company didn't simply sell itself when Meta came calling. Maeil Business Newspaper reported, in coverage later picked up by TechRadar, that FuriosaAI walked away from an $800 million acquisition offer from Meta in 2025 because the two sides could not agree on the company's future direction after a deal. Frankly, that was the harder choice. Cashing out is easy when Nvidia has most of the market and every rival has to prove not only the chip, but the software around it. June Paik founded FuriosaAI in 2017 with engineers from AMD, Qualcomm, Samsung, and other chip companies. Its first chip, Warboy, was built on Samsung's 14-nanometer process and didn't draw this kind of attention. Renegade changed that. Outlook Business and KED Global have reported that FuriosaAI is working with Morgan Stanley and Mirae Asset Securities on a pre-IPO round of up to $500 million, at a valuation near 3 trillion won, roughly $2.3 billion, with a possible IPO in 2027 or 2028. In May, FuriosaAI announced a partnership with Broadcom to build a third-generation accelerator. That's the big one, but it isn't here yet. The plan pairs a 2-nanometer compute die with a dedicated IO die and HBM4 or HBM4e memory, using Broadcom's advanced packaging and its Ethernet and PCIe interconnects to link accelerators at rack scale. Sampling is expected in the first half of 2028. So you should read Lisbon for what it is: today's deployment of the second-generation chip, not proof that the next one has arrived. ## Europe wants compute it can control Lisbon didn't happen by accident. FuriosaAI showed up at the RAISE Summit in Paris the same week, and Europe has been louder than any other region about wanting AI compute it doesn't have to import from the same narrow set of suppliers. You can see the appeal. A continent trying to talk seriously about AI sovereignty cannot depend forever on whichever Nvidia cards hyperscalers happen to have spare. CNBC reported that European chip startups including the Netherlands' Euclyd and Axelera and the UK's Fractile and Optalysys have been chasing nine-figure rounds this year. The money still tells a harsher story. Dealroom figures cited by CNBC put European AI chip funding at about $800 million in 2026, against $4.7 billion for American counterparts. Sovereignty is popular. Capital hasn't caught up. That gap is where FuriosaAI wants to stand: an Asian chipmaker with a working customer deployment, a Broadcom roadmap, and a rack already running in Lisbon. Nvidia isn't losing Europe because one Renegade server went live in Portugal. Don't pretend that. But if you're a cloud operator, a regulated company, or a public-sector buyer trying to avoid a single-vendor future, the interesting phrase here is not Nvidia challenger. It's available now. **Also read:** [Elon Musk Orders Tesla Staff Off Claude Despite Admitting Grok Is Worse](https://startupfortune.com/elon-musk-orders-tesla-staff-off-claude-despite-admitting-grok-is-worse/) • [Meta pulls its Muse Image AI tool from Instagram after three days of backlash over consent](https://startupfortune.com/meta-pulls-its-muse-image-ai-tool-from-instagram-after-three-days-of-backlash-over-consent/) • [OpenAI's safety chief exits as the company folds oversight into research](https://startupfortune.com/openais-safety-chief-exits-as-the-company-folds-oversight-into-research/)