Supermicro reveals Arm-based AI infrastructure

Supermicro, a provider of application-optimised IT systems, has announced a new portfolio of rack-scale infrastructure platforms based on Arm AGI CPUs, targeting enterprise AI and agentic AI workloads. The company says the systems have been designed to address increasing demand for compute capacity while improving energy efficiency and rack density within existing data centre environments. The new platforms combine Arm's Neoverse CSS V3-based CPU architecture with Supermicro's Data Center Building Block Solutions (DCBBS) approach, which integrates servers, storage, networking, cooling, and rack infrastructure. Charles Liang, President and CEO of Supermicro, says, "Supermicro continues to lead the industry when it comes to deploying new and innovative rack-scale solutions that maximise performance and efficiency. "Our DCBBS technology stack delivers end-to-end data centre solutions of any size, which, combined with the new density and efficient, performance-optimised Arm AGI CPU microarchitecture, helps enterprises realise significant TCO savings on their agentic AI infrastructure investments." The launch includes air-cooled and liquid-cooled server platforms designed for AI inference, AI training, cloud computing, and high-density enterprise workloads. Among the systems announced are a dual-socket 2U server for compute-intensive applications, a 5U GPU server supporting up to eight double-width GPUs, a liquid-cooled multi-node platform for rack-scale deployments, and a single-socket edge-focused server design. Focus on rack density and energy efficiency According to Supermicro and Arm, the infrastructure has been developed to maximise performance per watt and increase compute density for AI environments. Arm says its AGI CPU architecture features up to 136 cores per processor and is designed to support large-scale AI orchestration workloads through increased memory bandwidth, expanded memory capacity, and scalable I/O capabilities. The companies state that deployments can exceed 6,000 CPU cores within a single air-cooled rack, while larger Open Compute Project-based configurations can support significantly higher densities. Mohamed Awad, Executive Vice President, Cloud AI Business Unit at Arm, says, "Agentic AI is driving a fundamental shift in infrastructure requirements, where efficiency, scalability, and orchestration performance are becoming just as critical as raw compute. "By combining Arm AGI CPUs with Supermicro's rack-scale system expertise, we're enabling infrastructure designed to deliver higher AI throughput, maximum compute density, and improved data centre economics at scale." Supermicro says the platforms are intended to help organisations deploy AI infrastructure while making more efficient use of available data centre space, power, and cooling resources. The announcement expands Supermicro's portfolio of AI-focused infrastructure as demand continues to grow for high-density computing environments capable of supporting increasingly complex AI workloads. For more from Supermicro, click here.

Vertiv unveils high-capacity rack platform

Vertiv, a global provider of critical digital infrastructure, has introduced the Rack Extreme, a rack platform designed to support high-density computing, AI workloads, and next-generation IT deployments. The new rack has been developed to accommodate increasingly large and heavy computing equipment while supporting airflow management, cable organisation, and deployment flexibility within data centre environments. According to Vertiv, the platform is intended to address growing infrastructure requirements driven by higher compute densities and the adoption of AI applications. Giuseppe Leto, Senior Director, IT Systems at Vertiv, says, "The Vertiv Rack Extreme reflects our expanded capabilities in rack and enclosure designs for high-density and AI-driven deployments. "The platform also draws on Vertiv’s long-standing rack engineering expertise, including solutions historically developed under the Knürr brand, to support scalable next-generation IT infrastructure." The Rack Extreme is available in multiple sizes and configurations, allowing operators to tailor deployments to specific application requirements. The units are shipped fully assembled and are designed to integrate with a range of compatible cable management and airflow optimisation accessories. Designed for high-density deployments Vertiv says the Rack Extreme offers both static and dynamic load ratings of up to 2,045kg, enabling it to support high-density equipment installations while maintaining the same load capacity when being moved or when stationary. The company states that this provides greater flexibility during deployment and infrastructure changes, particularly in environments where heavy equipment must be repositioned after installation. The rack features a welded frame construction, integrated cable management options, high open-area mesh doors, flexible mounting rails, vertical cable bars, and corner mounting bars for rack power distribution units. Vertiv has also incorporated shipping features designed to simplify installation, including shock-absorbing pallets and reusable ramps intended to reduce the risk of equipment damage during transportation and deployment. The Rack Extreme has been designed to integrate with Vertiv's wider portfolio of data centre infrastructure products, including uninterruptible power supplies (UPS), rack PDUs, rear-door heat exchanger systems, coolant distribution units, and KVM management platforms. For more from Vertiv, click here.

GNM expands network presence into North America

GNM, a Dutch internet exchange (IX) and backbone operator, has expanded its network into North America with the launch of new points of presence (PoPs) in Miami and Ashburn, Virginia. The company has deployed its first US infrastructure at Equinix MI1 in Miami and Equinix DC1–DC15 and DC21 in Ashburn, creating a dedicated transatlantic platform linking North America with GNM's European backbone network. According to the company, the new sites are configured as a protected East Coast ring and are fully integrated with its existing infrastructure, allowing customers to exchange traffic across a single operational environment. The expansion marks GNM's first infrastructure deployment in the United States and is intended to support organisations seeking connectivity between North American and European markets. New PoPs strengthen transatlantic connectivity The Miami PoP is located within Equinix MI1, a major connectivity hub for subsea cable systems linking North America, Latin America, and Europe. Meanwhile, the Ashburn deployment places GNM within Northern Virginia, one of the world's largest data centre and interconnection markets. Both facilities connect directly to GNM-IX, the company's IX platform, which supports more than 700 connected networks and peak traffic exceeding 10.95Tbps. GNM says the new locations will allow network operators, carriers, cloud providers, and content platforms to access its interconnection and transport services through a single provider, while simplifying traffic exchange between continents. For North American organisations, the expansion provides direct access to European connectivity opportunities without requiring infrastructure deployments in Europe. European operators, meanwhile, gain direct access to two major US interconnection markets. Alex Surkov, Head of Business Development at GNM, says, "Launching in Miami and Ashburn is a defining milestone for GNM. "We have created our first North American platform and directly connected it to our European backbone. This gives customers on both sides of the Atlantic something genuinely valuable: simpler interconnection, direct access to new traffic flows, and the ability to grow internationally through one integrated network ecosystem." GNM now operates more than 90 PoPs globally, with infrastructure spanning Europe, Asia, and North America. For more from GNM, click here.

EUDCA welcomes European cloud and AI plans

The European Data Centre Association (EUDCA) has welcomed the publication of the proposed Cloud and AI Development Act (CADA), introduced by the European Commission as part of its Tech Sovereignty Package. The association says the proposal is an important step towards expanding Europe's compute capacity and creating conditions that support continued investment in digital infrastructure across the EU. According to EUDCA, access to scalable digital infrastructure will play a key role in Europe's ability to compete in artificial intelligence and other emerging technology sectors. The organisation argues that meeting growing demand for AI and cloud services will require further deployment of compute capacity alongside investment in skills and technology adoption. Permitting, investment, and infrastructure EUDCA highlighted several elements of the proposed legislation that it believes could help support future data centre development. These include requirements for EU Member States to develop national cloud and AI strategies, the creation of designated development zones with access to essential resources, and streamlined permitting processes intended to reduce barriers to infrastructure deployment. The association also welcomed proposals for a strategic project designation for data centres, which could help attract investment and support projects that demonstrate strong integration with energy systems and digital infrastructure. In addition, EUDCA noted the proposed alignment between the legislation and the Energy Efficiency Directive, arguing that existing reporting frameworks could help identify projects that meet high environmental and efficiency standards while avoiding additional administrative burdens. Michael Winterson, Secretary General of EUDCA, says, "The Cloud and AI Development Act marks an important step for Europe’s digital infrastructure ambitions. Our industry stands ready to support the growth of Europe’s compute capacity and is committed to contributing to the development of sustainable and resilient data centre capacity in Europe. "This initiative reflects several of the key enabling conditions we have long advocated for, including more efficient permitting, access to resources, and clearer strategic direction. We look forward to engaging with policymakers and Member States to support effective implementation." Calls for further action Whilst welcoming the proposal, EUDCA says several areas will remain important as the legislation progresses. The association is calling for a consistent application of definitions across EU Member States, continued investment in electricity grid infrastructure, measures to address skills shortages, and greater access to sustainable resources, including renewable energy and non-potable water for cooling. EUDCA says these factors will be essential in ensuring that Europe can develop the digital infrastructure required to support future AI growth while meeting sustainability objectives. For more from the EUDCA, click here.

AFL: Why AI infrastructure planning is changing

In recent years, AI infrastructure discussions centred predominantly on training clusters. Industry attention focused on larger models, sizeable GPU estates, dense scale-out fabrics, and the synchronisation demands created by collective communication across thousands of accelerators. In 2026, however, deployment patterns point to inference as the dominant operational AI workload. This transition introduces infrastructure behaviours that extend beyond the assumptions of traditional training environments. While much of the industry conversation still focuses on accelerators and compute scale, less attention is given to the implications for network architecture, optical connectivity, and physical infrastructure design. In response, AFL, a manufacturer of fibre optic cables and connectivity equipment, has developed a whitepaper series to help address that gap. The first paper, Architecting AI at Scale: From Training Clusters to Inference-Driven Infrastructure, introduces six workload categories representing the evolving AI deployment landscape. These include synchronous training fabrics, throughput inference systems, disaggregated reasoning architectures, heterogeneous decode environments, context-centric infrastructure, and workflow orchestration platforms. The paper provides practical insight into evolving network behaviours, optical requirements, and multi-domain infrastructure planning. Future instalments will examine the engineering implications in greater depth. Click here to register to receive email notifications as soon as each paper in the series becomes available. For more from AFL, click here.

Experts urge sustainable requirements for new developments

Industry experts are calling on the UK Government to introduce stronger sustainability requirements for new data centre developments, following reports that the sector's future energy demand could exceed previous forecasts. The comments follow a report by The Times suggesting that growing demand from data centres could significantly increase national electricity consumption, driven in part by the rapid expansion of AI infrastructure. Concerns have also been raised over proposals for some large-scale developments to build dedicated gas-fired power generation to overcome grid capacity constraints. David Woon, Head of Net Zero Engineering and Operations at Ennovus Solutions, says, “The figures for the expected energy demand of data centres are staggering, but immediately pivoting to new gas power stations is incredibly disappointing. "The space and resources required to build a new gas power plant could almost certainly be used instead for significant renewable generation development - ideally utilising wind turbines to better match the consistent, 24/7 energy demand of these facilities. “While on-site renewables may not provide 100% of the baseline power required by these data centres, a forward-thinking country aiming for energy independence and climate mitigation should jump at the chance to integrate green generation directly into planning permissions. "We already mandate solar panels on new-build homes; why are we not implementing similar, strict sustainable development mandates for industrial-scale data centres? “Furthermore, as the Government considers mandating grid ‘flexibility’ from operators, we must look beyond standard battery technologies like lithium-ion. Long-standing, energy-hungry data centres need a technology that matches their requirements, like Vanadium Flow batteries. They are suited to large energy demand projects, provide up to double the lifespan of lithium-ion, experience no degradation, and avoid environmentally hazardous, scarce materials like cobalt and lithium. "If battery storage is on the table to support the National Grid, it is nonsensical not to bring on-site renewable generation into the exact same conversation.” Grid constraints remain a major challenge Lee Ackerman, Utilities General Manager at Connectus Utilities, adds that while infrastructure upgrades are possible, they are likely to require significant investment and lengthy delivery timescales. He says, “The reality is that, while resolving these infrastructure bottlenecks is physically possible, it carries massive cost and time implications. "The National Energy System Operator (NESO) and Ofgem have transitioned the grid from a ‘first come, first served’ model to a ‘first ready and needed, first connected’ approach. "While major structural upgrades, new on-shore power lines, and smart grid sensors are scheduled for rollout between 2026 and 2028, there is an immense amount of work to do before we see true grid relief by 2030. “Every utility connection faces identical hurdles: cable lengths, land access, and complex legal processes. It’s not just an electricity problem either; data centres require major water capacity for cooling. “Developers could and should be targeting the Environmental Discounts offered in Water Charging Statements, aiming for Tier 2 or Tier 3 water neutrality incentives through advanced rainwater harvesting and greywater recycling. "Some might argue that technology will naturally become more efficient over time, but history shows that as components shrink, developers simply pack more technology into the same footprint. The energy demand isn’t going to drop on its own; we must build [in] sustainability from day one.” The comments highlight growing debate around how future AI and data centre infrastructure should be powered, with industry figures calling for greater emphasis on renewable energy generation, long-duration energy storage, water efficiency, and sustainable design principles during the planning process.

Arista launches 1.6T networking platforms for AI fabrics

Arista Networks, a provider of cloud and AI networking systems, has introduced the 7060XE7 Series, a new portfolio of 1.6T networking platforms designed for rack-scale AI infrastructure. The launch reflects growing demand for networking architectures capable of supporting increasingly large AI deployments, as training and inference environments scale from thousands to hundreds of thousands of accelerators. According to Arista, the new platforms are designed to support both scale-up and scale-out AI fabrics across air-cooled, liquid-cooled, and hybrid environments. The company says the 7060XE7 Series is intended to address the density, power, and thermal requirements associated with large AI clusters, whilst also enabling greater compute density within a given power envelope. Tyson Lamoreaux, Senior Vice President, Cloud and AI Networking at Arista Networks, comments, "The AI era requires a shift in how we think about the network. It is no longer a standalone layer of infrastructure, but a tightly integrated component of the AI supersystem. "With the 7060XE7 Series, we are delivering massive-scale 1.6T systems that combine world-class reliability and the differentiation of EOS with liquid cooling and low-power optics to help our customers build AI fabrics designed for maximum performance and power efficiency." Designed for large-scale AI deployments The 7060XE7 Series includes fixed-switch platforms and configurable rack-scale systems designed to support a range of AI workloads and infrastructure requirements. According to Arista, the systems provide low-latency connectivity and intelligent packet buffering to manage the traffic patterns associated with AI training and inference workloads. The platforms also support a range of EOS features aimed at improving resilience, congestion management, and operational visibility within AI environments. The portfolio comprises three main configurations: • 7060XE7-64PS and 7060XE7-64PRS rack switches, offering 64 1.6T ports in an air-cooled 4RU design• 7060XE7-64PRS-RV3-L, a liquid-cooled 2OU platform designed for high-density AI clusters• 7060XE7-128PE, providing 128 800G ports in an air-cooled 4RU form factor The systems use 224G and 100G SerDes technologies, depending on configuration, and support Linear Pluggable Optics (LPO), which Arista says can reduce interconnect power consumption by around 60%. Industry support for 1.6T ethernet Several major cloud providers have provided statements supporting the development of higher-capacity ethernet infrastructure for AI environments. Gaya Nagarajan, Vice President of Infrastructure at Meta, says, "Arista’s 1.6T platforms and liquid-cooled designs align with our focus on open, scalable AI fabrics that meet the requirements of next-generation training and inference." Rani Borkar, President, Azure Hardware Systems and Infrastructure at Microsoft, notes, "Our collaboration with Arista on the 1.6T ethernet interface helps enable the next generation of AI clusters with greater interconnect capacity for Azure Maia, Microsoft's AI accelerator, and Fairwater, Microsoft's extreme-scale AI data centres, while preserving operational simplicity across our infrastructure." Mahesh Thiagarajan, Executive Vice President, Oracle Cloud Infrastructure, adds, "Arista Networks’ 1.6T platforms provide the throughput, determinism, and stability needed for our RDMA-based AI fabrics, while Arista EOS delivers operational consistency and performance at scale across our global AI infrastructure." The 7060XE7 Series is also supported through collaborations with AMD and Broadcom, with the platforms utilising Broadcom's Tomahawk 6 ethernet switching silicon. Arista expects the first systems in the portfolio to become available during Q4 2026, with additional models scheduled for release during Q1 2027. For more from Arista, click here.

Raltron launches compact OCXOs for DC timing

Raltron, a manufacturer of frequency control components, has announced the OX7000 Series of oven-controlled crystal oscillators (OCXOs), developed to provide a timing option for network interface cards (NICs) and other space-constrained data centre hardware. The OX7000 Series is designed to support networking applications requiring stable, low-phase-noise reference clocks while maintaining a compact footprint. The devices are housed in a 9mm x 7mm surface-mount package, making them suitable for use in NIC cards, servers, storage systems, and network switching platforms. The company says the oscillators are intended to help designers address challenges around board space, power consumption, and system cost as networking and data centre infrastructure continues to scale. Designed for networking and data centre hardware The OX7000 Series operates from a single 3.3V supply and consumes between 0.35W and 0.5W during steady-state operation. During warm-up at 25°C, power consumption ranges from 0.8W to 0.9W. According to Raltron, the devices combine OCXO stability with low phase noise in a package aimed at high-volume networking applications. Sasha Wolloch, President of Raltron, says, "Network and data centre designers need timing products that balance performance, size, and cost. With the OX7000 Series, we're providing a compact, lower-cost OCXO that is especially well suited for NIC cards and other high-volume networking platforms." The OX7000 Series has also been qualified for environmental testing, including vibration, mechanical shock, and thermal cycling, which are common requirements for continuously operating infrastructure equipment.

Pure DC completes cross-border biomethane deal

Pure Data Centres Group (Pure DC), a designer, developer, and operator of hyperscale data centres, has completed what it describes as Europe's first large-scale cross-border biomethane purchase for a data centre, transferring 9GWh of certified biomethane from Germany to the Irish gas network. The transaction follows a proof-of-concept project announced in March 2026 and represents a significant increase in scale. According to Pure DC, the deal demonstrates that biomethane can be used to support the decarbonisation of gas-connected data centres through existing energy infrastructure and certification frameworks. The biomethane was produced in Germany during 2025 using waste and residue feedstocks. The fuel is certified under the International Sustainability and Carbon Certification (ISCC) scheme and complies with the requirements of the Renewable Energy Directive (RED) II and RED III. Pure DC says the biomethane has a carbon intensity of less than 12gCO₂/MJ, meeting thresholds required for zero-rated treatment under the EU Emissions Trading System, subject to approval of monitoring plans by the relevant authorities. The gas was mass balanced from the German network to Ireland through existing interconnectors, with renewable attributes tracked through the Gas Networks Ireland (GNI) Renewable Gas Registry. Biomethane forms part of wider net zero strategy Pure DC says renewable gas is intended to act as a transitional measure within its broader strategy to achieve net zero emissions by 2040. The company is continuing to pursue biomethane procurement from Irish, European, and UK sources while also developing longer-term initiatives including renewable energy integration, energy storage, and efficiency improvements. Ireland's National Biomethane Strategy targets production of up to 5.7TWh of domestic biomethane annually by 2030. While local production capacity continues to develop, imports can help meet demand through existing gas infrastructure and certification systems. Maria Jose Rivas Duarte, Director of Sustainability at Pure DC, says, "This milestone supports Ireland’s Climate Action Plan and aligns with the LEU policy, under which data centres must meet at least 80% of their annual energy demand with additional renewable electricity. "By demonstrating that cross-border biomethane can be procured, mass balanced, and registered at volume through existing infrastructure, we are helping to pave the way for broader data centre sector adoption as well as other industries seeking a credible route to decarbonise natural gas." Pure DC says the transaction forms part of a strategy to develop a diversified biomethane portfolio spanning multiple geographies, feedstocks, and supply arrangements. Agnes Warner, Property Director at Pure DC, says, "Our biomethane procurement strategy is designed to build a balanced portfolio that provides long-term security and credibility. "For our customers, this means a demonstrable, auditable pathway to lower embedded emissions, making our platform more attractive to hyperscalers and enterprise customers while supporting their sustainability commitments." For more from Pure DC, click here.

Green Horizon secures approval for Norway data centre

Green Horizon, a Norwegian developer of hydropower-backed, AI-ready data centres, has received planning approval for Norway 1, a 36MW data centre development near Stavanger that is scheduled to enter service in the second half of 2027. The approval follows the earlier granting of zoning permission for the site and allows the Norwegian developer to progress to final design and construction. The company is currently working with consultants and contractors ahead of a planned construction start later this year. Located on Norway's southwest coast, Norway 1 is being developed as a carrier-neutral and cloud-neutral facility designed to provide connectivity to the UK, mainland Europe, and onward routes to North America. The facility will be built to Tier III standards and is designed to support high-density AI, GPU, and high-performance computing (HPC) workloads. Green Horizon says the site will include two 'Meet-Me Rooms', diverse connectivity options, and access to multiple network providers. Norway 1 forms the first phase of the company's wider data centre platform in the Stavanger region, where 96MW of power capacity has been secured across three planned developments. The company says the facility will be powered by renewable hydropower and is targeting a power usage effectiveness (PUE) rating of 1.1 at full load. Heat reuse strategy integrated into design A key element of the project is its heat reuse strategy. Green Horizon plans to supply excess heat generated by the data centre to both a new greenhouse that will be integrated into the facility's design and an existing commercial greenhouse located adjacent to the site. According to the company, the new greenhouse will be constructed directly above the data centre, enabling waste heat to be reused as part of a wider symbiosis partnership with Norway's largest greenhouse operator. The concept has been technically validated and approved by the local municipality. Operations at the site will be supported by CBRE, which will provide operational services and monitoring. Richard Rettedal, CEO of Green Horizon, comments, "Securing planning approval for Norway 1 marks a major milestone for Green Horizon and for our ambition to build Norway’s AI data centre platform. "Customers deploying AI and high-performance compute need dependable capacity, resilience, and a clear route to scale. Norway 1 is designed to deliver high-density infrastructure powered by renewable hydropower, with heat reuse enabled by design, supporting both lower-cost operation and a lower operational footprint. "We’re proud that this project will contribute to the local community and bring new, renewable powered capacity to the market." The €300 million (£259 million) development is expected to create around 400 construction jobs during the build phase and contribute additional renewable-powered data centre capacity to Norway's digital infrastructure. Construction is expected to begin later in 2026, with the facility targeted to become operational in the second half of 2027.