Korean AI inference chip vendor FuriosaAI deploys inference servers at Equinix Lisbon data center, targeting European sovereign AI market requirements.

South Korean AI semiconductor startup FuriosaAI is expanding into the European AI inference market via a strategic partnership with global data center operator Equinix. The company has deployed servers powered by its second-generation inference chip at Equinix’s Lisbon data center in Portugal and has commenced proof-of-concept (PoC) testing with local clients.

The partnership and global expansion strategy were announced during a joint media briefing in Seoul’s Yeouido district on the 15th. Following the commencement of mass production for its second-generation AI inference semiconductor in January, FuriosaAI rapidly secured international infrastructure, deploying Renegade (RNGD) servers at Equinix’s LS2 facility in Lisbon.

Yoon Ji-hoon, product manager at FuriosaAI, outlined the strategic rationale: “We plan to mass-produce 100,000 cards per month from this year through next year. Whether for direct sales or service operations, we needed dedicated data center space to conduct PoC testing with customers.”

Europe was selected as the initial overseas foothold due to surging sovereign AI demand. Driven by stringent data sovereignty requirements and regulations such as the General Data Protection Regulation (GDPR) and the EU AI Act, FuriosaAI identified a robust market for validated on-premises AI infrastructure. “The European market has very strong data security and policy regulation tendencies,” Yoon noted. “There was already considerable interest in our products within Europe.”

The European target audience encompasses a wide spectrum, ranging from AI-native startups to financial and insurance firms, alongside government and municipal agencies. FuriosaAI is simultaneously evaluating deployment across diverse inference workloads, including text-based generative AI and computer vision systems.

Under the partnership, FuriosaAI will delegate local infrastructure operations to Equinix while concentrating internal resources on AI chip and server product development. “We possess deep expertise in chip and product design, but we recognized the critical importance of partnering with an infrastructure specialist to manage environments that allow diverse end users to operate seamlessly. That capability is precisely why we selected Equinix,” Yoon explained.

Equinix offers more than colocation; it provides comprehensive peering and interconnection services. These capabilities enable FuriosaAI to link local data centers, cloud platforms, and client environments while expanding its footprint within the Equinix ecosystem. “Having physical infrastructure in a target market is a decisive factor internationally,” Yoon added. “Beyond colocation, Equinix delivers extensive interconnection and peering services. Through these channels, we have connected with numerous prospective clients and are actively advancing follow-up discussions.”

Following successful PoC completion, the partnership will adopt a phased rollout model. Initial deployments will test service components on FuriosaAI’s dedicated infrastructure before scaling to cloud partners or transitioning into direct service offerings. This framework will be sequentially replicated, beginning in Europe, followed by expansion into the Americas and Southeast Asia.

Both organizations are navigating acute data center capacity constraints. “Supply is overwhelmingly insufficient,” Yoon stated. Jang Hye-deok, country managing director of Equinix Korea, echoed this assessment, projecting that “the situation where demand far outpaces supply due to AI will continue through next year.”

Looking ahead, Equinix outlined its network infrastructure strategy for the emerging agentic AI era. Anthony Ho, Director of Product Management for Equinix Asia-Pacific, detailed the operator’s approach, forecasting that agentic AI will expand at a compound annual growth rate (CAGR) of 175% over the next five years through 2029 within enterprise networking—a trajectory substantially outpacing generative AI’s projected growth rate of approximately 90%.

Globally, new AI network traffic is projected to grow at an 85% CAGR over the coming decade. Concurrently, high-capacity AI content delivery network traffic, encompassing extended reality (XR), is anticipated to surge by 151% over the same timeframe. Key growth projections include:

• Agentic AI enterprise networking CAGR (five years through 2029): 175%

• Generative AI growth rate: ~90%

• New AI network traffic CAGR (next ten years): 85%

• High-capacity AI content distribution network traffic growth: 151%

Ho highlighted four critical limitations of legacy enterprise networks in the AI era: reliance on manual, human-centric operations; protracted timelines for global service deployment; inadequate support for agentic AI workflows; and reactive, post-incident fault detection stemming from a lack of embedded intelligence.

To overcome these bottlenecks, Equinix has positioned “Distributed AI Hubs” and “Fabric Intelligence” as foundational solutions. Distributed AI Hubs provide integrated frameworks featuring AI gateways, security guardrails, data storage, observability tools, orchestration capabilities, and cost-performance optimization modules.

Fabric Intelligence serves as an AI-driven connectivity platform designed to unify and manage distributed AI infrastructure across cloud, neo-cloud, on-premises, and data environments. Its Fabric Super Agent automates over 60 network utilities, compressing deployment cycles that once required weeks down to mere minutes. Furthermore, AI agents and external systems can programmatically access Fabric capabilities via Model Context Protocol (MCP) servers.

As part of this initiative, Equinix launched an “AI Discovery Hub” at its Hong Kong data center (HK6) roughly six months ago. Technology partners including HPE, NVIDIA, Dell, and Schneider Electric have contributed to the facility, enabling enterprises to rigorously test and validate AI applications prior to large-scale commercialization. Director Anthony Ho confirmed that Equinix is currently evaluating a comparable AI Discovery Hub for South Korea.

Equinix currently manages three facilities in South Korea, located in Seoul’s Sangam district and Goyang’s Hyangdong area, with a fourth site under construction slated for completion in the first quarter of 2028. “The capacity to efficiently interconnect data, cloud, and networks—rather than relying solely on semiconductor performance—will ultimately dictate AI competitiveness,” Jang emphasized. “We remain committed to supporting the global expansion of South Korean innovators, including FuriosaAI.”