[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"kb-article-furiosaai-rngd-inference-accelerator-lands-at-equinix-lisbon-power-efficient-ai-for-europe-en":3,"ArticleBody_7Bc03XVhF5jQw0Tyy0kc5Y70icWfKrz9FIIPJW9dlys":206},{"article":4,"relatedArticles":175,"locale":60},{"id":5,"title":6,"slug":7,"content":8,"htmlContent":9,"excerpt":10,"category":11,"tags":12,"metaDescription":10,"wordCount":13,"readingTime":14,"publishedAt":15,"sources":16,"sourceCoverage":52,"transparency":54,"seo":57,"language":60,"featuredImage":61,"featuredImageCredit":62,"isFreeGeneration":66,"trendSlug":67,"trendSnapshot":68,"niche":77,"geoTakeaways":80,"geoFaq":89,"entities":99},"6a52f5dc3b2138b8b5d0b4fa","FuriosaAI RNGD Inference Accelerator Lands at Equinix Lisbon: Power-Efficient AI for Europe","furiosaai-rngd-inference-accelerator-lands-at-equinix-lisbon-power-efficient-ai-for-europe","European AI teams need more inference capacity, but many grids, power envelopes, and legacy data centers cannot support megawatt‑scale GPU clusters without costly upgrades.[4] FuriosaAI, led by CEO [June Paik](\u002Fentities\u002F6a52f737b15b2ddcc32bfcde-june-paik), is deploying its RNGD inference accelerators at Equinix’s LS2 facility in [Lisbon](\u002Fentities\u002F69de2171dc9b12943744bdb9-lisbon) as an alternative—[frontier‑grade inference](\u002Farticle\u002Fnvidia-gtc-2026-inside-the-agentic-ai-and-inference-infrastructure-wave) inside a 3 kW power budget per server.[1][2][4]\n\nFor ML engineers, LS2 becomes a place to test and scale LLM and agent workloads on hardware tuned for tokens‑per‑watt, while staying within typical European rack and power limits.[4]\n\n**Key takeaway:** Lisbon’s LS2 gives EU enterprises local, low‑latency access to power‑efficient inference hardware without waiting for AI‑only data centers.[1][2]\n\n---\n\n## Why FuriosaAI Chose Equinix Lisbon for RNGD’s European Launch\n\nFuriosaAI is rolling out its NXT RNGD servers at Equinix LS2 as the first node in a broader European footprint.[1][3][4] LS2 operates as a neutral, carrier‑dense colocation hub where customers bring their own networks and data and tap into specialized AI compute.\n\nTo anchor this, FuriosaAI opened a Lisbon office that combines:[1][4]\n\n- Commercial operations  \n- Customer engineering  \n- R&D for compilers, chip design, and PCB design  \n\nHaving engineering teams close to production infrastructure lets firmware, kernel, and compiler changes be validated quickly against real workloads.\n\nEurope’s AI ambitions are constrained by:[4][5]\n\n- Limited grid capacity and power availability  \n- Difficulty adding 30–40 kW GPU racks without major upgrades  \n- Cooling limits in legacy facilities  \n\nRNGD addresses this gap by delivering high inference density in 3 kW‑class servers that fit into standard air‑cooled racks.[1][4][5]\n\nEquinix LS2 offers:[2]\n\n- 2,050 sqm of colocation space over three floors  \n- Initial capacity for 625 racks  \n- An efficiency‑focused design aligned with FuriosaAI’s emphasis on energy‑efficient inference[1][2]  \n\n**Real‑world angle:** SaaS and fintech firms exploring LLM‑based support or analytics often find power and rack constraints—not model cost—block GPU cluster deployments; LS2 plus RNGD allows scale within existing colocation contracts.[4][5]\n\n---\n\n## Inside the RNGD Architecture and NXT RNGD Server at LS2\n\nRNGD is built on FuriosaAI’s proprietary Tensor Contraction Processor (TCP) architecture, fabricated on a [5 nm](https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002F5_nm_process) process.[1][2][4][5] Each accelerator delivers:\n\n- 512 TFLOPS of FP8 performance  \n- 180 W thermal design power  \n\nThis is tuned for production inference, especially LLMs where FP8 throughput and low TDP maximize tokens‑per‑watt within tight rack power budgets.[1][2][4][5]\n\nAt the system level, up to eight RNGD accelerators integrate into the NXT RNGD Server, yielding a 3 kW‑class system with:[1][2][4][5]\n\n- Up to 384 GB of HBM across cards  \n- Capacity to run 100B‑plus parameter LLMs at high concurrency  \n- No need for exotic cooling or major facility upgrades[5]  \n\n**Key point:** NXT RNGD servers are air‑cooled and drop into standard racks, avoiding liquid‑cooling retrofits that can delay deployments for months.[4][5]\n\nCompared to more power‑hungry accelerators, RNGD’s 180 W profile is modest; Nvidia’s RTX Pro 6000‑class devices can draw ~3.33× more power for similar workloads.[5] In a European rack capped at ~5–8 kW, this can be the difference between:\n\n- One GPU‑heavy node, or  \n- A multi‑server inference cluster.[4][5]\n\nThe Lisbon deployment reflects a trend of placing AI compute where power is efficient and sustainable, similar to Hive Digital’s use of [Paraguay](\u002Fentities\u002F6a39d454add847c9a8510301-paraguay)’s renewable‑heavy grid for AI workloads.[7] AI hubs are now chosen for energy characteristics as much as latency.[1][7]\n\nWithin the inference ecosystem, RNGD sits alongside LLM‑optimized chips like OpenAI’s Jalapeño, which targets substantially better performance per watt than current state‑of‑the‑art devices.[8][9] Both approaches share a thesis:[4][8][9]\n\n- Workload‑specific, full‑stack design beats raw peak FLOPS for serving large models.  \n\n**Key takeaway:** RNGD’s advantage is not just “512 FP8 TFLOPS” but delivering that throughput inside 180 W per card, enabling meaningful cluster sizes in power‑constrained European racks.[1][4][5]\n\n---\n\n## Enterprise Benefits, Use Cases, and How to Engage in Lisbon\n\nFor EU enterprises, LS2 offers three primary benefits:[1][2][4]\n\n- Local, low‑latency RNGD access for EU‑hosted data and users  \n- A controlled environment to benchmark LLM and agentic workloads on specialized inference hardware  \n- Predictable power and rack planning via 3 kW‑class servers  \n\nTypical engagement starts with a proof‑of‑concept supported by FuriosaAI’s Lisbon technical team, where customers port a few representative services—such as a retrieval‑augmented LLM API—to NXT RNGD servers.[1][4] They then benchmark:\n\n- Tokens per second and latency at target concurrency  \n- Tokens per watt and per‑request energy cost  \n- Total cost of ownership vs. existing GPU setups over 1–3 years[1][4][5]  \n\nFuriosaAI positions RNGD for LLMs and agentic AI, with a software stack that reduces the need for hand‑tuned kernels and minimizes migration overhead.[4]\n\nIllustrative use cases include:[1][4][5]\n\n- **Financial services:** Risk models and multilingual chatbots that must stay within regulated data‑center power ceilings.[4]  \n- **Customer support automation:** High‑throughput ticket triage and conversational agents with fixed rack footprints.[1][4]  \n- **Public‑sector and sovereign AI:** EU‑based, power‑efficient infrastructure that aligns with data residency and sustainability mandates.[4][5]\n\nLisbon is a template for additional European rollouts, likely reusing the “local office + colocation + power‑efficient inference” pattern across more hubs.[1][3][4] PoCs proven in Lisbon are intended to be portable to future Furiosa‑enabled regions.\n\n**Key point:** Understanding your workload on low‑TDP inference hardware early makes later scaling easier without re‑architecting around grid or cooling constraints.[1][4][5]\n\n---\n\n## Conclusion: A Practical Path to Efficient European Inference\n\nFuriosaAI’s RNGD deployment at [Equinix LS2](\u002Farticle\u002Fmistral-ai-s-vibe-industrial-engineering-stack-and-data-center-bet) combines a 5 nm, FP8‑optimized inference accelerator with a scalable, power‑aware colocation environment, offering European enterprises a practical way to grow modern AI workloads without waiting for entirely new data centers.[1][2][4][5]","\u003Cp>European AI teams need more inference capacity, but many grids, power envelopes, and legacy data centers cannot support megawatt‑scale GPU clusters without costly upgrades.\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa> FuriosaAI, led by CEO \u003Ca href=\"\u002Fentities\u002F6a52f737b15b2ddcc32bfcde-june-paik\">June Paik\u003C\u002Fa>, is deploying its RNGD inference accelerators at Equinix’s LS2 facility in \u003Ca href=\"\u002Fentities\u002F69de2171dc9b12943744bdb9-lisbon\">Lisbon\u003C\u002Fa> as an alternative—\u003Ca href=\"\u002Farticle\u002Fnvidia-gtc-2026-inside-the-agentic-ai-and-inference-infrastructure-wave\" class=\"internal-link\">frontier‑grade inference\u003C\u002Fa> inside a 3 kW power budget per server.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>For ML engineers, LS2 becomes a place to test and scale LLM and agent workloads on hardware tuned for tokens‑per‑watt, while staying within typical European rack and power limits.\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Key takeaway:\u003C\u002Fstrong> Lisbon’s LS2 gives EU enterprises local, low‑latency access to power‑efficient inference hardware without waiting for AI‑only data centers.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003C\u002Fp>\n\u003Chr>\n\u003Ch2>Why FuriosaAI Chose Equinix Lisbon for RNGD’s European Launch\u003C\u002Fh2>\n\u003Cp>FuriosaAI is rolling out its NXT RNGD servers at Equinix LS2 as the first node in a broader European footprint.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa> LS2 operates as a neutral, carrier‑dense colocation hub where customers bring their own networks and data and tap into specialized AI compute.\u003C\u002Fp>\n\u003Cp>To anchor this, FuriosaAI opened a Lisbon office that combines:\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Commercial operations\u003C\u002Fli>\n\u003Cli>Customer engineering\u003C\u002Fli>\n\u003Cli>R&D for compilers, chip design, and PCB design\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Having engineering teams close to production infrastructure lets firmware, kernel, and compiler changes be validated quickly against real workloads.\u003C\u002Fp>\n\u003Cp>Europe’s AI ambitions are constrained by:\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Limited grid capacity and power availability\u003C\u002Fli>\n\u003Cli>Difficulty adding 30–40 kW GPU racks without major upgrades\u003C\u002Fli>\n\u003Cli>Cooling limits in legacy facilities\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>RNGD addresses this gap by delivering high inference density in 3 kW‑class servers that fit into standard air‑cooled racks.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>Equinix LS2 offers:\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>2,050 sqm of colocation space over three floors\u003C\u002Fli>\n\u003Cli>Initial capacity for 625 racks\u003C\u002Fli>\n\u003Cli>An efficiency‑focused design aligned with FuriosaAI’s emphasis on energy‑efficient inference\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cstrong>Real‑world angle:\u003C\u002Fstrong> SaaS and fintech firms exploring LLM‑based support or analytics often find power and rack constraints—not model cost—block GPU cluster deployments; LS2 plus RNGD allows scale within existing colocation contracts.\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Chr>\n\u003Ch2>Inside the RNGD Architecture and NXT RNGD Server at LS2\u003C\u002Fh2>\n\u003Cp>RNGD is built on FuriosaAI’s proprietary Tensor Contraction Processor (TCP) architecture, fabricated on a \u003Ca href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002F5_nm_process\" class=\"wiki-link\" target=\"_blank\" rel=\"noopener\">5 nm\u003C\u002Fa> process.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa> Each accelerator delivers:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>512 TFLOPS of FP8 performance\u003C\u002Fli>\n\u003Cli>180 W thermal design power\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>This is tuned for production inference, especially LLMs where FP8 throughput and low TDP maximize tokens‑per‑watt within tight rack power budgets.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>At the system level, up to eight RNGD accelerators integrate into the NXT RNGD Server, yielding a 3 kW‑class system with:\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Up to 384 GB of HBM across cards\u003C\u002Fli>\n\u003Cli>Capacity to run 100B‑plus parameter LLMs at high concurrency\u003C\u002Fli>\n\u003Cli>No need for exotic cooling or major facility upgrades\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cstrong>Key point:\u003C\u002Fstrong> NXT RNGD servers are air‑cooled and drop into standard racks, avoiding liquid‑cooling retrofits that can delay deployments for months.\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>Compared to more power‑hungry accelerators, RNGD’s 180 W profile is modest; Nvidia’s RTX Pro 6000‑class devices can draw ~3.33× more power for similar workloads.\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa> In a European rack capped at ~5–8 kW, this can be the difference between:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>One GPU‑heavy node, or\u003C\u002Fli>\n\u003Cli>A multi‑server inference cluster.\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>The Lisbon deployment reflects a trend of placing AI compute where power is efficient and sustainable, similar to Hive Digital’s use of \u003Ca href=\"\u002Fentities\u002F6a39d454add847c9a8510301-paraguay\">Paraguay\u003C\u002Fa>’s renewable‑heavy grid for AI workloads.\u003Ca href=\"#source-7\" class=\"citation-link\" title=\"View source [7]\">[7]\u003C\u002Fa> AI hubs are now chosen for energy characteristics as much as latency.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-7\" class=\"citation-link\" title=\"View source [7]\">[7]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>Within the inference ecosystem, RNGD sits alongside LLM‑optimized chips like OpenAI’s Jalapeño, which targets substantially better performance per watt than current state‑of‑the‑art devices.\u003Ca href=\"#source-8\" class=\"citation-link\" title=\"View source [8]\">[8]\u003C\u002Fa>\u003Ca href=\"#source-9\" class=\"citation-link\" title=\"View source [9]\">[9]\u003C\u002Fa> Both approaches share a thesis:\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-8\" class=\"citation-link\" title=\"View source [8]\">[8]\u003C\u002Fa>\u003Ca href=\"#source-9\" class=\"citation-link\" title=\"View source [9]\">[9]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Workload‑specific, full‑stack design beats raw peak FLOPS for serving large models.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cstrong>Key takeaway:\u003C\u002Fstrong> RNGD’s advantage is not just “512 FP8 TFLOPS” but delivering that throughput inside 180 W per card, enabling meaningful cluster sizes in power‑constrained European racks.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Chr>\n\u003Ch2>Enterprise Benefits, Use Cases, and How to Engage in Lisbon\u003C\u002Fh2>\n\u003Cp>For EU enterprises, LS2 offers three primary benefits:\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Local, low‑latency RNGD access for EU‑hosted data and users\u003C\u002Fli>\n\u003Cli>A controlled environment to benchmark LLM and agentic workloads on specialized inference hardware\u003C\u002Fli>\n\u003Cli>Predictable power and rack planning via 3 kW‑class servers\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Typical engagement starts with a proof‑of‑concept supported by FuriosaAI’s Lisbon technical team, where customers port a few representative services—such as a retrieval‑augmented LLM API—to NXT RNGD servers.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa> They then benchmark:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>Tokens per second and latency at target concurrency\u003C\u002Fli>\n\u003Cli>Tokens per watt and per‑request energy cost\u003C\u002Fli>\n\u003Cli>Total cost of ownership vs. existing GPU setups over 1–3 years\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>FuriosaAI positions RNGD for LLMs and agentic AI, with a software stack that reduces the need for hand‑tuned kernels and minimizes migration overhead.\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>Illustrative use cases include:\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Financial services:\u003C\u002Fstrong> Risk models and multilingual chatbots that must stay within regulated data‑center power ceilings.\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Customer support automation:\u003C\u002Fstrong> High‑throughput ticket triage and conversational agents with fixed rack footprints.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Public‑sector and sovereign AI:\u003C\u002Fstrong> EU‑based, power‑efficient infrastructure that aligns with data residency and sustainability mandates.\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Lisbon is a template for additional European rollouts, likely reusing the “local office + colocation + power‑efficient inference” pattern across more hubs.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-3\" class=\"citation-link\" title=\"View source [3]\">[3]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa> PoCs proven in Lisbon are intended to be portable to future Furiosa‑enabled regions.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Key point:\u003C\u002Fstrong> Understanding your workload on low‑TDP inference hardware early makes later scaling easier without re‑architecting around grid or cooling constraints.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n\u003Chr>\n\u003Ch2>Conclusion: A Practical Path to Efficient European Inference\u003C\u002Fh2>\n\u003Cp>FuriosaAI’s RNGD deployment at \u003Ca href=\"\u002Farticle\u002Fmistral-ai-s-vibe-industrial-engineering-stack-and-data-center-bet\" class=\"internal-link\">Equinix LS2\u003C\u002Fa> combines a 5 nm, FP8‑optimized inference accelerator with a scalable, power‑aware colocation environment, offering European enterprises a practical way to grow modern AI workloads without waiting for entirely new data centers.\u003Ca href=\"#source-1\" class=\"citation-link\" title=\"View source [1]\">[1]\u003C\u002Fa>\u003Ca href=\"#source-2\" class=\"citation-link\" title=\"View source [2]\">[2]\u003C\u002Fa>\u003Ca href=\"#source-4\" class=\"citation-link\" title=\"View source [4]\">[4]\u003C\u002Fa>\u003Ca href=\"#source-5\" class=\"citation-link\" title=\"View source [5]\">[5]\u003C\u002Fa>\u003C\u002Fp>\n","European AI teams need more inference capacity, but many grids, power envelopes, and legacy data centers cannot support megawatt‑scale GPU clusters without costly upgrades.[4] FuriosaAI, led by CEO Ju...","trend-radar",[],888,4,"2026-07-12T02:12:12.238Z",[17,22,26,30,32,36,40,44,48],{"title":18,"url":19,"summary":20,"type":21},"FuriosaAI Expands European AI Infrastructure with RNGD Deployment at Equinix’s Lisbon Data Center","https:\u002F\u002Ffuriosa.ai\u002Fblog\u002Ffuriosaai-equinixs-lisbon-data-center-press-release","LISBON - JULY 7, 2026 -\n\nFuriosaAI today announced the expanded availability of its RNGD AI inference accelerator across Europe. As an initial step, Furiosa is currently installing RNGD servers at Equ...","kb",{"title":23,"url":24,"summary":25,"type":21},"FuriosaAI deploys its RNGD servers at Equinix data center in Lisbon, Portugal","https:\u002F\u002Fwww.datacenterdynamics.com\u002Fen\u002Fnews\u002Ffuriosaai-deploys-its-rngd-servers-at-equinix-data-center-in-lisbon-portugal\u002F","South Korean AI chip firm FuriosaAI has expanded its footprint in Europe with a deployment of its custom AI inference accelerators.\n\nThe company has deployed RNGD servers at Equinix's LS2 data center ...",{"title":27,"url":28,"summary":29,"type":21},"FuriosaAI finds a European beachhead for efficient inference","https:\u002F\u002Fwww.facebook.com\u002Fjonpeddieresearch\u002Fposts\u002Ffuriosaai-is-expanding-access-to-its-rngd-inference-accelerator-in-europe-throug\u002F1615323130599812\u002F","FuriosaAI is expanding access to its RNGD inference accelerator in Europe through a deployment at Equinix’s Lisbon LS2 data center. The move gives European enterprises a local environment to evaluate ...",{"title":27,"url":31,"summary":29,"type":21},"https:\u002F\u002Fwww.jonpeddie.com\u002Fnews\u002Ffuriosaai-finds-a-european-beachhead-for-efficient-inference\u002F",{"title":33,"url":34,"summary":35,"type":21},"South Korean chip startup FuriosaAI invades European datacenters","https:\u002F\u002Fwww.theregister.com\u002Fai-and-ml\u002F2026\u002F07\u002F07\u002Fsouth-korean-chip-startup-furiosaai-invades-european-datacenters\u002F5267884","RNGD accelerators land in Equinix's Lisbon DCs\n\nPower-efficient South Korean AI chip startup FuriosaAI has landed on European shores.\n\nOn Tuesday, the chip biz revealed that it had begun fielding its ...",{"title":37,"url":38,"summary":39,"type":21},"HIVE's Paraguay AI infrastructure performance validated in Columbia University study","https:\u002F\u002Fwww.bnamericas.com\u002Fen\u002Fnews\u002Fhives-paraguay-ai-infrastructure-performance-validated-in-columbia-university-study","Press Release\n\nHIVE's Paraguay AI infrastructure performance validated in Columbia University study\n\nBnamericas Published: Wednesday, June 24, 2026",{"title":41,"url":42,"summary":43,"type":21},"Hive Digital Technologies launches AI cloud platform in Asunción, Paraguay","https:\u002F\u002Fnewenergyevents.com\u002Fhive-launches-ai-cloud-platform-in-paraguay-advancing-hpc-and-llm-capabilities\u002F","Hive Digital Technologies has launched an AI cloud platform from its data center in Asunción, Paraguay, supporting high-performance computing and large language model research, including projects with...",{"title":45,"url":46,"summary":47,"type":21},"OpenAI and Broadcom unveil LLM-optimized inference chip","https:\u002F\u002Fwww.reddit.com\u002Fr\u002FLocalLLaMA\u002Fcomments\u002F1ueexbr\u002Fopenai_and_broadcom_unveil_llmoptimized_inference\u002F","Quoted from the start of the blog post:\n\n- Early testing shows that the first-generation accelerator will deliver performance per watt substantially better than current state-of-the-art\n- Built from t...",{"title":49,"url":50,"summary":51,"type":21},"OpenAI's Jalapeño: AI Designed Inference Chip for LLMs","https:\u002F\u002Fwww.linkedin.com\u002Fposts\u002Frichard-ho-chips_openai-and-broadcom-unveil-llm-optimized-activity-7475540055822901248-_988","Richard Ho\n2w\n\nWhen we started Jalapeño, the question was not “how do we build another AI accelerator?” It was: what should an inference chip look like if it is designed around the way modern LLMs act...",{"totalSources":53},9,{"generationDuration":55,"kbQueriesCount":53,"confidenceScore":56,"sourcesCount":53},316479,100,{"metaTitle":58,"metaDescription":59},"FuriosaAI RNGD Inference: Power-Efficient AI in Lisbon","EU teams face power limits. FuriosaAI deploys RNGD at Equinix Lisbon to run tokens-per-watt inference in 3 kW racks — learn how it boosts local AI capacity","en","https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1591696331111-ef9586a5b17a?ixid=M3w4OTczNDl8MHwxfHNlYXJjaHwxfHxmdXJpb3NhYWklMjBybmdkJTIwaW5mZXJlbmNlJTIwYWNjZWxlcmF0b3J8ZW58MXwwfHx8MTc4MzgyMTc4OHww&ixlib=rb-4.1.0&w=1200&h=630&fit=crop&crop=entropy&auto=format,compress&q=60",{"photographerName":63,"photographerUrl":64,"unsplashUrl":65},"Markus Winkler","https:\u002F\u002Funsplash.com\u002F@markuswinkler?utm_source=coreprose&utm_medium=referral","https:\u002F\u002Funsplash.com\u002Fphotos\u002Fwhite-and-black-typewriter-with-white-printer-paper-tGBXiHcPKrM?utm_source=coreprose&utm_medium=referral",true,"furiosaai-rngd-inference-accelerator-deployment-at-equinix-lisbon",{"score":56,"type":69,"sourceCount":70,"topSourceDomains":71,"detectedAt":75,"mentionsLast7Days":76},"spiking",8,[72,73,74],"techtimes.com","datacenterdynamics.com","startupfortune.com","2026-07-12T00:07:49.320Z",2,{"key":78,"name":79,"nameEn":79},"ai-engineering","AI Engineering & LLM Ops",[81,83,85,87],{"text":82},"Equinix LS2 in Lisbon provides local, low‑latency access to FuriosaAI’s RNGD inference hardware and offers 2,050 sqm of colocation space with initial capacity for 625 racks.",{"text":84},"The NXT RNGD server integrates up to eight RNGD accelerators delivering 512 TFLOPS FP8 per card and 180 W TDP, enabling a 3 kW‑class server that can host 100B+ parameter LLMs with high concurrency.",{"text":86},"RNGD’s 180 W per card profile and air‑cooled design allow deployment in standard European racks without liquid‑cooling retrofits, enabling multi‑server inference clusters inside typical 5–8 kW rack envelopes.",{"text":88},"Lisbon deployment reduces the need for megawatt‑scale upgrades: enterprises can benchmark tokens‑per‑watt, latency, and total cost of ownership on-site via FuriosaAI’s local engineering team and PoC support.",[90,93,96],{"question":91,"answer":92},"What makes RNGD and the Lisbon deployment suitable for European AI inference workloads?","RNGD is purpose-built for inference with a workload‑specific Tensor Contraction Processor fabricated on a 5 nm node, delivering 512 TFLOPS FP8 at 180 W per accelerator, which maximizes tokens‑per‑watt. Lisbon’s Equinix LS2 provides a neutral, carrier‑dense colocation with 2,050 sqm and room for 625 racks, and the NXT RNGD servers are air‑cooled 3 kW‑class systems that fit standard racks. That combination lets EU teams run large LLMs locally with low latency and within existing power and cooling limits—avoiding megawatt upgrades or liquid‑cooling retrofits.",{"question":94,"answer":95},"How do enterprises engage with FuriosaAI in Lisbon and run a proof‑of‑concept?","Start with a PoC supported by FuriosaAI’s Lisbon commercial and engineering teams who help port representative services—such as a retrieval‑augmented LLM API—onto NXT RNGD servers. The PoC typically benchmarks tokens\u002Fsec and latency at target concurrency, tokens-per-watt, per-request energy cost, and 1–3 year total cost of ownership versus current GPU setups. FuriosaAI’s local office enables rapid firmware, kernel, and compiler validation against real workloads, so customers can iterate quickly and determine whether to scale within existing colocation contracts.",{"question":97,"answer":98},"How does RNGD compare to conventional GPUs like Nvidia for inference efficiency and scaling?","RNGD targets inference efficiency rather than raw peak FLOPS: each card provides 512 FP8 TFLOPS at a 180 W TDP, which is roughly 3× lower power than many high‑end Nvidia devices for comparable inference tasks. This lower per‑card power and air‑cooled form factor allow multiple NXT RNGD servers to fit into a single 5–8 kW rack footprint, enabling meaningful cluster sizes where traditional GPUs would force single high‑power nodes or facility upgrades. In practice, RNGD yields higher tokens‑per‑watt and simpler facility integration for EU sites constrained by grid, rack, and cooling limits.",[100,108,114,119,124,130,134,141,147,152,158,163,168],{"id":101,"name":102,"type":103,"confidence":104,"wikipediaUrl":105,"slug":106,"mentionCount":107},"6960720d19d266277e14ff98","LLMs","concept",0.99,null,"6960720d19d266277e14ff98-llms",489,{"id":109,"name":110,"type":103,"confidence":111,"wikipediaUrl":105,"slug":112,"mentionCount":113},"699841d19aa9beba177c6f78","FP8",0.93,"699841d19aa9beba177c6f78-fp8",3,{"id":115,"name":116,"type":103,"confidence":117,"wikipediaUrl":105,"slug":118,"mentionCount":76},"6a52f77db15b2ddcc32bfd1e","Tensor Contraction Processor",0.92,"6a52f77db15b2ddcc32bfd1e-tensor-contraction-processor",{"id":120,"name":121,"type":103,"confidence":117,"wikipediaUrl":105,"slug":122,"mentionCount":123},"6a52f825b15b2ddcc32bfd58","180 W","6a52f825b15b2ddcc32bfd58-180-w",1,{"id":125,"name":126,"type":103,"confidence":127,"wikipediaUrl":128,"slug":129,"mentionCount":123},"6a52f824b15b2ddcc32bfd55","5 nm",0.9,"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002F5_nm_process","6a52f824b15b2ddcc32bfd55-5-nm",{"id":131,"name":132,"type":103,"confidence":127,"wikipediaUrl":105,"slug":133,"mentionCount":123},"6a52f825b15b2ddcc32bfd57","512 TFLOPS","6a52f825b15b2ddcc32bfd57-512-tflops",{"id":135,"name":136,"type":137,"confidence":104,"wikipediaUrl":138,"slug":139,"mentionCount":140},"6a39d454add847c9a8510301","Paraguay","location","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FParaguay","6a39d454add847c9a8510301-paraguay",25,{"id":142,"name":143,"type":137,"confidence":104,"wikipediaUrl":144,"slug":145,"mentionCount":146},"69de2171dc9b12943744bdb9","Lisbon","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FLisbon","69de2171dc9b12943744bdb9-lisbon",6,{"id":148,"name":149,"type":137,"confidence":150,"wikipediaUrl":105,"slug":151,"mentionCount":123},"6a52f824b15b2ddcc32bfd54","Equinix LS2",0.97,"6a52f824b15b2ddcc32bfd54-equinix-ls2",{"id":153,"name":154,"type":155,"confidence":104,"wikipediaUrl":105,"slug":156,"mentionCount":157},"6a52f737b15b2ddcc32bfcda","FuriosaAI","organization","6a52f737b15b2ddcc32bfcda-furiosaai",5,{"id":159,"name":160,"type":155,"confidence":104,"wikipediaUrl":161,"slug":162,"mentionCount":76},"6a4a90ff0a066c693a415763","HIVE Digital","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FHive_(ransomware)","6a4a90ff0a066c693a415763-hive-digital",{"id":164,"name":165,"type":166,"confidence":127,"wikipediaUrl":105,"slug":167,"mentionCount":123},"6a52f826b15b2ddcc32bfd5c","EU enterprises","other","6a52f826b15b2ddcc32bfd5c-eu-enterprises",{"id":169,"name":170,"type":171,"confidence":172,"wikipediaUrl":173,"slug":174,"mentionCount":14},"6a52f737b15b2ddcc32bfcde","June Paik","person",0.96,"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FNam_June_Paik","6a52f737b15b2ddcc32bfcde-june-paik",[176,184,191,199],{"id":177,"title":178,"slug":179,"excerpt":180,"category":181,"featuredImage":182,"publishedAt":183},"6a5320ec3b2138b8b5d0b83c","EU AI Act Enforcement from August 2, 2026: What ML and AI Teams Must Change Now","eu-ai-act-enforcement-from-august-2-2026-what-ml-and-ai-teams-must-change-now","From August 2, 2026, high‑risk AI systems in the EU move from soft guidance to hard enforcement, with penalties up to €35 million or 7% of global annual turnover for serious violations.[1][2] Complian...","safety","https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1603815210222-16aed3c10dfc?ixid=M3w4OTczNDl8MHwxfHNlYXJjaHwxfHxhY3QlMjBlbmZvcmNlbWVudCUyMGF1Z3VzdCUyMDIwMjZ8ZW58MXwwfHx8MTc4MzgzMzE4MXww&ixlib=rb-4.1.0&w=1200&h=630&fit=crop&crop=entropy&auto=format,compress&q=60","2026-07-12T05:13:00.151Z",{"id":185,"title":186,"slug":187,"excerpt":188,"category":181,"featuredImage":189,"publishedAt":190},"6a51cf1487450904396743a8","GPT-5.6 in the Wild: How OpenAI’s New Model and Custom Silicon Will Reshape Production LLM Systems","gpt-5-6-in-the-wild-how-openai-s-new-model-and-custom-silicon-will-reshape-production-llm-systems","GPT-5.6 is landing in a different world than GPT-4 or 5.4. OpenAI now owns not just models and products but also a custom “Intelligence Processor” ASIC, Jalapeño, designed specifically for LLM inferen...","https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1782414963066-2aab3094fd43?ixid=M3w4OTczNDl8MHwxfHNlYXJjaHwxfHxncHQlMjB3aWxkJTIwb3BlbmFpJTIwbmV3fGVufDF8MHx8fDE3ODM3NDY1NTV8MA&ixlib=rb-4.1.0&w=1200&h=630&fit=crop&crop=entropy&auto=format,compress&q=60","2026-07-11T05:09:14.613Z",{"id":192,"title":193,"slug":194,"excerpt":195,"category":196,"featuredImage":197,"publishedAt":198},"6a51186587450904396739fc","JadePuffer: Engineering the First Fully LLM‑Driven Ransomware Kill Chain","jadepuffer-engineering-the-first-fully-llm-driven-ransomware-kill-chain","1. From LLM Hallucinations to Operational Malware: Why JadePuffer Is Plausible\n\nBrowser-only ransomware was once dismissed as “LLM hallucination,” until researchers showed a fully browser-native ranso...","hallucinations","https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1581092335397-9583eb92d232?ixid=M3w4OTczNDl8MHwxfHNlYXJjaHwxfHxqYWRlcHVmZmVyJTIwZW5naW5lZXJpbmclMjBmaXJzdCUyMGZ1bGx5fGVufDF8MHx8fDE3ODM3MTU1MTl8MA&ixlib=rb-4.1.0&w=1200&h=630&fit=crop&crop=entropy&auto=format,compress&q=60","2026-07-10T16:10:32.272Z",{"id":200,"title":201,"slug":202,"excerpt":203,"category":196,"featuredImage":204,"publishedAt":205},"6a50eede874509043967394c","JadePuffer: Inside the First Fully LLM‑Driven Ransomware Attack and How Langflow Agents Were Weaponized","jadepuffer-inside-the-first-fully-llm-driven-ransomware-attack-and-how-langflow-agents-were-weaponized","JadePuffer shows what happens when autonomous LLM agents, wired into real tools and data, are given ransomware objectives.\n\n- 75% of organizations were hit by ransomware in the last year; average brea...","https:\u002F\u002Fimages.unsplash.com\u002Fphoto-1678957949479-b1e876bee3f1?ixid=M3w4OTczNDl8MHwxfHNlYXJjaHw2MXx8YXJ0aWZpY2lhbCUyMGludGVsbGlnZW5jZSUyMHRlY2hub2xvZ3l8ZW58MXwwfHx8MTc4MzY5MzkyMHww&ixlib=rb-4.1.0&w=1200&h=630&fit=crop&crop=entropy&auto=format,compress&q=60","2026-07-10T13:14:22.299Z",["Island",207],{"key":208,"params":209,"result":211},"ArticleBody_7Bc03XVhF5jQw0Tyy0kc5Y70icWfKrz9FIIPJW9dlys",{"props":210},"{\"articleId\":\"6a52f5dc3b2138b8b5d0b4fa\",\"linkColor\":\"red\"}",{"head":212},{}]