Huawei sets out its vision for intelligent data infrastructure

Speaking at Huawei Connect 2026 in Florence, Huawei's Data Storage leadership made the case that the AI race has entered a new phase, one where data quality and architecture - not raw compute - will determine which enterprises succeed. Florence, Italy, played host to Huawei Connect 2026 on 16 October, with the company's Data Storage Summit drawing together enterprise leaders, technology partners, and industry experts to examine how data infrastructure must evolve to serve the demands of the AI era. The central argument, articulated most clearly by Yuan Yuan, President of Huawei's Data Storage Product Line, was deceptively simple: computing power and model development have had their moment, and AI has now entered a stage where data is the decisive factor. "The next chapter of AI is data," Yuan says. His case rests on a structural argument: deploying AI successfully at enterprise scale is no longer primarily a question of accessing sufficient compute or choosing the right foundation model. The bottleneck has shifted to data - its quality, its accessibility, and the infrastructure built to manage it. Enterprises that fail to address this will find their AI investments underperforming, regardless of how much they spend on GPUs. A platform built for AI inference The headline product announcement at Florence was Huawei's 3+1 AI Data Platform Solution - the company's claim to be the first in the industry to integrate a knowledge base, key-value (KV) cache store, memory bank, and Unified Cache Manager into a single platform. The solution has already been deployed across more than 20 service scenarios, and the customer cases presented at the summit illustrate the practical impact. A leading hospital used the platform to address a persistent problem with large-model AI in clinical settings: limited understanding of specialist medical terminology and a tendency towards hallucinations. By integrating multimodal medical data and applying knowledge parsing, vectorisation, and knowledge graph construction, the hospital developed a clinical-grade agent for diagnosis and treatment, lifting the adoption rate of AI-assisted diagnosis reports from 50% to 90%. In a different context, a major internet company was struggling with the storage demands of multi-turn, long-sequence inference tasks for AI coding workflows. Huawei's platform improved inference throughput by a factor of 2.2 - measured in tokens per second - by replacing repeated computation with associative search, using multi-level cache scheduling and prefix cache techniques. Meanwhile, a large bank applied the platform to its customer service operations, automatically extracting historical conversation data and enabling memory recall in seconds, resulting in a 16% increase in customer satisfaction. Infrastructure that keeps pace with AI Lucio Ceruti, CTO of Huawei Data Storage Italy, outlined the product-level upgrades underpinning the broader platform story. Performance improvements centre on FlashLink intelligent algorithms, built on a data and control plane separation architecture, which deliver a 50% performance uplift. On resilience, an integrated ransomware protection system with network-storage collaboration achieves 99.99% ransomware identification accuracy, supported by intelligent health inspection and automatic diagnosis capabilities. Energy efficiency gains come from hardware-based offloading of deduplication and compression, combined with dynamic CPU frequency reduction, reducing overall power consumption without compromising throughput. Yuan also highlighted an upgraded AI Data Lake Solution, built on higher-density green hardware, with global data visibility, multi-dimensional vector retrieval, and the ability to deliver data in seconds - designed to act as a one-stop supply of high-quality data for AI training and inference pipelines. European deployments in practice Two European customer cases grounded the summit's broader narrative in concrete infrastructure terms. Daniele Cesarini, Head of AI/HPC Architecture at Cineca - Italy's national supercomputing centre - described how the organisation adopted Huawei OceanStor Pacific scale-out storage to build an exabyte-scale scientific research data lake with tiered storage for hot, warm, and cold data. The deployment supports the Square Kilometre Array project, among other large-scale scientific programmes, and has significantly reduced both system complexity and total cost of ownership. Marco Frontini, Chief Information and Digital Officer of Cerba HealthCare Italia, described a different challenge: consolidating a legacy IT architecture spread across 50 labs and five central nodes into a single full-stack platform. The company, which operates 400 medical centres, serves more than 1.2 million patients, and conducts over 50 million examinations each year, achieved a 23% reduction in five-year TCO while adding ransomware protection and zero data loss capabilities. Building for what comes next Adley Zhang, Director of Huawei's European Data Centre Marketing and Solution Sales Department, contextualised the product and customer announcements within Huawei's broader European trajectory. The company has been expanding its presence across finance, healthcare, education and research, and retail, and Zhang emphasised Huawei's commitment to an open and innovative ecosystem as the foundation for continued growth in the region. For data centre and infrastructure professionals, Huawei believes its Florence summit's message is worth taking seriously. The shift Yuan describes - from compute-centric to data-centric AI infrastructure - has real architectural implications. It demands storage systems capable of supporting high-throughput inference, low-latency retrieval, long-term memory, and seamless data governance at scale. Huawei's argument is that these requirements call for a purpose-built platform, not a collection of upgraded legacy components. The customer deployments on show in Florence suggest that argument is beginning to find traction across European enterprises. For more from Huawei, click here. This article is sponsored by Huawei

JSM completes 110kV substation at Berlin data centre

JSM Services, part of utility infrastructure provider JSM Group, has completed the structural works on a 110kV substation at maincubes’ mainHub data centre campus in Nauen, near Berlin, six months after construction began. The topping-out ceremony took place on 13 October and marks progress on the power infrastructure for the campus. As general contractor, JSM is delivering the 200MVA substation and an approximately 6km high-voltage cable route connecting the campus to the Wustermark substation. The project is JSM’s first in Germany. Construction began at the end of February, but ground investigations found topsoil up to 3m deep, compared with the expected 1m. Around 4,200m³ of soil was replaced to provide a secure foundation. During summer temperatures of up to 41°C, concrete pours were scheduled for the early hours to maintain progress. The substation incorporates around 250 tonnes of structural steel embedded in concrete and approximately 3,500m² of precast walls and ceilings. An 11m-span ceiling in the central block was lifted into place in a single day. Peter Östlund (pictured above), Managing Director of JSM Services, says, “Today, there is no longer a wooden roof truss here, but concrete and steel. Yet, the essence of a topping-out ceremony remains the same: it is a celebration of gratitude towards the people who built it. "10 design firms, 14 companies, and even more suppliers have worked together on this building. Despite its comparatively small size, it is highly complex, and every one of them has played their part in getting us here.” Cable route works progress in parallel The substation will house four transformers, each weighing just under 70 tonnes. The installation of building services is scheduled to begin at the end of October, followed by the delivery of the transformers. Work on the cable route is progressing across four construction phases. More than 27km of protective conduit has been laid, around 14,000m³ of cable trench excavated, and more than 3,000 tonnes of asphalt removed. 10 horizontal boreholes are planned to cross the L161 and B5 roads. The project has also involved managing archaeological finds, crossing existing water and gas pipes and other utilities, and protecting trees and wildlife. JSM reports that around 18,000 working hours have been completed at the substation and a further 6,200 on the cable route, totalling approximately 24,000 hours without an accident resulting in lost time or serious injury. Thomas Wacker, Chief Development Officer at maincubes, comments, “The topping-out of the substation marks an important milestone in the development of mainHub Berlin and brings us one step closer to delivering the infrastructure needed to support the campus. "A reliable power supply is fundamental to the long-term success of any data centre, and we are pleased to see this critical element taking shape on schedule. The progress achieved in just six months is a testament to the commitment and expertise of everyone involved. "We look forward to the next phase as we continue to bring our vision for mainHub Berlin to life.” The Nauen development expands maincubes’ presence in the Berlin region, building on the operation of its existing BER01 data centre. For more from JSM, click here.

center3, Epsilon to expand Saudi internet exchange access

Middle Eastern digital infrastructure provider center3 has appointed Epsilon Telecommunications, a global connectivity provider offering network, cloud, and colocation services, as a reseller for its one-IX (internet exchange) platform, enabling international networks and digital service providers to access exchange points in Jeddah, Riyadh, and Dammam. The partnership will extend the reach of center3’s internet exchange services through Epsilon’s international connectivity network, allowing global networks, cloud providers, and content aggregators to peer with networks in Saudi Arabia. Powered by the London Internet Exchange (LINX) since 2019, center3’s internet exchange platform supports interconnection across Saudi Arabia. The company says the partnership will help improve traffic localisation, reduce latency, and simplify access for international networks seeking to connect with the country’s digital ecosystem. Expanding international connectivity Hussein Alnajrani, Chief Commercial Officer at center3, says, “The selection of Epsilon as our reseller partner represents an important step in expanding the global reach of center3 internet exchange nodes. "Through our carrier-neutral data centres, international networks can connect to Saudi Arabia’s digital ecosystem and benefit from low-latency, secure, and scalable interconnection services. "As demand for streaming content, cloud, and digital services continues to accelerate across the region, center3 remains committed to building the infrastructure and partnerships that enable digital growth and contribute to Saudi Arabia’s ambition to become a leading global connectivity hub.” Warren Aw, Chief Commercial Officer at Epsilon Telecommunications, adds, “It's great to work with center3 to enable efficient, high-performance peering in and out of Saudi Arabia. Together, we're making it easier for content providers to connect with users across the Middle East, APAC, and beyond. "This partnership is an important milestone in our ongoing collaboration with center3 and supports growing demand for low-latency, high-capacity connectivity in the region.” The companies say the partnership will benefit sectors including streaming, gaming, digital media, cloud services, and AI applications, where low-latency connectivity and proximity to users are important. The expanded access is intended to support more efficient traffic exchange and localised content delivery as demand for digital services grows across the region. For more from Epsilon, click here.

ORCA launches PT-3 photonic quantum computer

ORCA Computing, a London-based developer of photonic quantum computers for data centres, has launched PT-3, a photonic quantum computer designed to deliver commercial quantum advantage. The first system is operational at Digital Realty’s Innovation Lab in London, where it runs alongside NVIDIA accelerated computing. ORCA says PT-3 works alongside graphics processing units (GPUs) to increase the expressivity of AI models, helping them learn from limited training data and identify candidates for new drugs and materials. The system has generated results using a hybrid quantum-classical extension of NVIDIA’s Proteína generative protein-design model. Trained using 32 NVIDIA Hopper GPUs and a PT-3 system operating with 64 photons in 128 qumodes, the model improved the ratio of useful proteins by 20% compared with a classical baseline. It also generated novel proteins with a different mix of secondary structures. These results build on earlier work using ORCA’s PT-2 system with the Technical University of Denmark, which produced around twice as many strong peptide binders as classical AI, validated in laboratory testing, and generated 49% more valid, unique molecules on the QM9 benchmark. Faster optimisation workflows PT-3 is also being used for industrial optimisation, including energy scheduling. In work involving JIJ and the UK’s National Quantum Computing Centre, the system ran a hybrid quantum-classical workflow for a 25,755-variable unit commitment problem in about an hour. ORCA says this is more than 12 times faster than its predecessor at the same solution quality. The associated whitepaper found that the hybrid solver produced results competitive with a commercial optimisation solver and, with additional runtime, found the best solutions among the methods tested. Richard Murray, co-founder and CEO of ORCA Computing, comments, “By integrating our photonic quantum computer with NVIDIA accelerated computing, we have taken a critical first step in the future of hybrid quantum supercomputing. "The PT-3 is already delivering results in AI for science and industrial optimisation, from protein design to energy scheduling. Through Digital Realty, enterprise users can now put it to work on their most challenging workloads, and customers are trialling use cases in life sciences, defence, energy, and logistics.” Sam Stanwyck, Director of Quantum Product at NVIDIA, adds, “With the PT-3, ORCA is bringing photonic quantum computing and NVIDIA accelerated computing together in a system researchers can use to develop and test new applications. "The value of a computing platform is in what researchers build on it. ORCA’s extension of NVIDIA’s Proteína model to generate a different range of protein structures with improved designability is a great example of what can be accomplished when these tools come together.” Séamus Dunne, Managing Director, UK and Ireland at Digital Realty, says, “Our partnership with ORCA reflects a shared ambition to make quantum computing practical for business. "Hosting the PT-3 in the DRIL gives us a real-world environment to explore how quantum can deliver value today, and it’s a significant step towards commercial quantum advantage at scale.”

Colt DCS, RMZ begin construction of 64MW Mumbai data centre

Colt Data Centre Services (Colt DCS), a hyperscale and colocation data centre operator, and RMZ, an Indian real estate and digital infrastructure group, have begun construction of Mumbai 3, a 64MW data centre at the Navi Mumbai campus in India. The AI-ready facility follows the signing of the campus's first hyperscale customer contract and is fully pre-let. The project is part of a joint venture announced in 2024, under which the companies committed to invest $1.7 billion (£1.2 billion) in India’s data centre market. Ground preparation began in December 2025. Mumbai 3 will combine air cooling with direct liquid-to-chip cooling to support AI and high-performance computing workloads. The facility will also have an on-campus 220kV gas-insulated switchgear (GIS) substation and carrier-neutral connectivity with access to high-capacity fibre routes. Pratap Mane, Country Head DCS, India, says, “The nature of data centre demand is evolving rapidly, with customers requiring larger deployments, higher rack densities, and infrastructure capable of supporting increasingly intensive AI workloads. Mumbai 3 is designed to address these needs, further strengthening our Navi Mumbai Campus as a scalable and future-ready digital infrastructure platform. "Built with AI readiness at its core, the facility incorporates advanced cooling capabilities, including liquid cooling, and enhanced power infrastructure to support high-density deployments. This enables customers to efficiently scale next-generation AI and high-performance computing workloads while maintaining operational resilience and efficiency.” Renewable energy and local jobs The facility is planned to operate on 100% renewable energy and use a closed-loop cooling system designed to produce near-zero wastewater from the cooling process. The project is expected to create more than 700 jobs across construction and operations whilst supporting local suppliers and contractors. Mumbai 3 will join the existing Mumbai 1 and Mumbai 2 facilities at the campus. Once complete, it will increase the site’s operational capacity to 86MW, with potential for the campus to reach 148MW at full development. Deepak Chhabria, President, RMZ Digital Infrastructure, comments, “The next wave of digital growth will be driven by infrastructure that brings together scale, resilience, and speed to market. "At Mumbai 3, we are developing a future-ready campus designed to enable hyperscalers and neo-cloud providers to scale seamlessly as their capacity and computing needs evolve. "This investment reinforces our commitment to building advanced data centre and AI infrastructure that meets global standards and supports the rapidly evolving demands of the digital economy.” The joint venture has committed to developing 250MW of data centre capacity across Mumbai and Chennai. For more from Colt DCS, click here.

1.3GW solar farm and 1GW data centre proposed in Norfolk

EPL Norfolk, a subsidiary of UK-based independent energy developer Evolution Power, has announced plans for Longmarsh Green Energy, a proposed 1.3GW solar farm and battery project alongside a 1GW data centre campus near King’s Lynn in west Norfolk. A community consultation on the proposal is set to run from 15 October to 26 November 2026. The project comprises two sites: one northeast of Tilney St Lawrence and another near Fincham and Barton Bendish. The development would connect to the existing transmission network substation at Walpole, with a connection date expected in 2031. The proposed data centre campus is expected to attract more than £25 billion in regional investment. According to EPL Norfolk, it would create thousands of construction jobs, more than 500 direct operational roles, and support a further 1,500 jobs. Renewable power and regional investment The company says the combination of local offshore wind generation and 1.3GW of colocated solar capacity would allow the campus to be powered predominantly by locally sourced renewable energy. It also proposes a closed-loop cooling system that recirculates liquid rather than continually drawing water from mains supplies. A spokesperson for the project says, “Longmarsh Green Energy would bring significant local and national economic benefits, providing secure renewable energy and data centre capacity, both of which are critical for future UK economic growth. "Our design includes measures and commitments to reduce local effects, and we welcome feedback on these, together with suggestions regarding how the project could support local community projects and initiatives.”

Pure DC completes final transformer installation at LON01

Pure Data Centres Group (Pure DC), a designer, developer, and operator of hyperscale data centres, has installed the third and final power transformer at its 90MW LON01 data centre campus in North London, bringing the substation’s total capacity to 150MVA. The 75MVA transformer, manufactured by Hitachi, will supply power to the campus’s B1 and B2 buildings. Energisation is planned for later this month, completing the dedicated power system serving the campus. LON01 is being developed to support cloud computing and AI inference workloads, using a closed-loop liquid cooling system designed for high-density computing. Specialist lifting system overcomes site constraints Installing the transformer presented a logistical challenge because restricted space around the live substation prevented the use of a conventional 500-tonne crane. The project team instead designed and deployed a specialist gantry system with a lifting capacity of 1,000 tonnes to move the transformer into position. The lift took four hours and involved Pure DC’s project team, MGE Group, logistics partners, and specialist suppliers. The operation reportedly followed extensive preparation, including route planning, road closures, and local authority approvals. The installation completed an 18-month programme that began with manufacturing in Colombia in March 2025. Following factory testing in June 2026, the transformer was transported by sea to the UK before being delivered to the campus. Once B2 is complete, the building will feature a living wall comprising more than 750,000 plants. Pure DC says the installation is intended to reduce noise and air pollution and encourage biodiversity, including insects and birds, near Brent Reservoir’s Site of Special Scientific Interest (SSSI). For more from Pure DC, click here.

Echion, Proper Voltage to co-develop 1.6MW energy system

Echion Technologies, a UK developer of niobium-based fast-charging battery anode materials, and Proper Voltage, a US developer of battery management software and electronics, are collaborating to develop a 1.6MW uninterruptible power supply (UPS) system designed to buffer fluctuating AI workloads and bridge power outages in data centres. The proposed VaultX system will integrate Echion’s niobium-based battery technology into Proper Voltage’s Vault UPS platform, which supports 480V and 800V configurations. The companies have signed a memorandum of understanding (MoU) covering the development. AI training workloads can cause power demand to fluctuate by tens of megawatts within milliseconds, placing additional demands on utility connections, switchgear, and generation equipment. The companies say buffering these fluctuations within data centres could help operators make better use of existing grid capacity. VaultX is planned to deliver 1.6MW of power and 84kWh of storage capacity. Its battery system will use Echion’s XNO anode technology, which the company says is designed to support high-rate charging and discharging over repeated cycles. The technology is intended to meet the demands of AI data centres, where frequent power fluctuations can require rapid charging and discharging. Echion says its battery chemistry can support charging or discharging in three to six minutes and more than 50,000 cycles under specified test conditions. UPS system combines backup power and load buffering Jean de La Verpilliere, CEO of Echion, comments, “AI data centres need exactly what XNO was built for, which is extremely high charge and discharge rates sustained across tens of thousands of cycles. "Our materials and cells are built for duty cycles that aren’t survivable for conventional chemistries, and collaborating with Proper Voltage enables us to implement these capabilities in a market-leading UPS platform and ensure that operators can benefit from VaultX in the shortest possible amount of time. “Conventional chemistries can deliver serviceable discharge but cannot safely handle the continuous and instantaneous high-C-rate charging that AI data centres require. XNO addresses this. "Operators will soon be able to incorporate a single system that serves both the backup mission and continuous load buffering. I look forward to working with Proper Voltage to launch what has the potential to be a step-change product to the market.” Charlie Welch, CEO of Proper Voltage, adds, “Our architecture was purposely designed so that operators can adopt new and emerging battery technology without redesigning their power systems, which is why integrating XNO into our Vault architecture is straightforward. "What this system changes for an operator is the business case. A UPS that also buffers load stops being an asset that sits idle between outages and starts earning its footprint every day. "We recognise the impact that XNO can have on data centre operations and its symmetric power capabilities fundamentally change what a data centre UPS can do to support emerging GPUs and ease the stress placed on upstream grid infrastructure. It’s exciting to be moving into the next phase of a project with this much potential to shape AI infrastructure.”

Aggreko invests in liquid cooling test equipment

Aggreko, a British multinational temporary power generation company, has invested in a fleet of liquid-cooled load banks to support the testing and commissioning of AI data centre cooling systems across Europe. The units are designed to simulate the thermal and electrical loads of IT equipment, allowing operators to test cooling and electrical infrastructure before a facility goes live or during refurbishment. Liquid cooling is increasingly used in high-density AI and high-performance computing environments. During commissioning, the load banks can help operators validate equipment including chiller lines, pipework, pumps, heat exchangers, and coolant distribution units (CDUs), reducing the risk of equipment failures during operation. Load banks support testing at different power levels The units can be configured to accept 200kW, 300kW, or 600kW across single or multiple connections, accommodating different electrical supply arrangements on site. Constructed from 316L stainless steel, the equipment is designed to reduce material degradation and the risk of contamination in liquid cooling systems. Each unit and its accessories undergo specialist cleaning, flushing, sealing, and testing before deployment. The load banks also offer adjustable operating modes, allowing engineers to control power loads, temperature differentials, and leaving water temperatures within specified parameters. Multiple units can be networked to manage larger loads and support testing under different operating conditions. Torben Hiller, Lead Development Engineer at Aggreko, comments, “The rapid growth of AI has ushered in a new era of data centre cooling. However, with even a single outage presenting six- or seven-figure ramifications, ensuring that this cooling infrastructure is fit to handle the demands of day-to-day operation is paramount. “Our new LCLBs are the first of their kind on the market, and feature bespoke control systems that allow them to be set up for precise, scenario-specific testing. "Combined with the knowledge and sector experience of our data centre team, this new investment will allow us to offer unrivalled testing and commissioning support for European AI data centres and their cooling equipment, providing operators with the best possible assurance that their infrastructure will be able to handle real load conditions.” Torben also says testing under different scenarios is important as rack densities increase, warning that inadequate testing can lead to overheating, poor heat rejection, coolant contamination, and reduced equipment lifespan. For more from Aggreko, click here.

Could smarter monitoring reduce data centre wastewater?

Liquid waste removal experts at BusinessWaste.co.uk say operators will need to think more carefully about how water is used across their sites, where wastewater is being created, and what can be done to reduce it: Water use depends on how sites are designed Water demand can vary considerably between data centres, depending on the cooling systems and infrastructure in place. Some cooling systems rely much more heavily on water, whilst others are designed to recirculate more of it. That means decisions made during site planning can have an impact on how much water it eventually uses. For new developments, that could mean considering cooling systems, local water availability, and whether any water reuse opportunities exist before construction begins. Existing sites can also get a clearer picture of where water is going by using meters and sensors to track usage over time. A sudden rise in consumption could prompt checks for a leak, an equipment problem, or another change in the site. Where cooling or maintenance processes produce wastewater, cutting unnecessary water use and reusing more of it could also help reduce the amount that eventually needs to be treated or discharged. Could AI help operators spot problems earlier? As large data centres already produce a huge amount of operational data, AI could help operators make more sense of it. Used alongside existing monitoring systems, AI could pull together information from water meters, cooling equipment, and other parts of a site, which could make unusual changes or any emerging patterns easier to spot. This could help operators investigate problems sooner, especially on large sites where it may be difficult to track everything manually. Where a site generates wastewater through cooling or maintenance, the same monitoring could also help operators track changes in the amount being produced and understand where those changes originate. There is still a lack of clear evidence around how much water data centres use and how AI could affect that demand. The Environment Agency has faced difficulties gathering reliable water-use data from the sector, which has made it difficult to assess future demand. Research into AI-led cooling is also developing, although much of the focus so far has been on energy efficiency. A recent review of 64 studies found that only three reported water usage effectiveness (WUE), suggesting there is still more to understand about how AI optimisation could affect water consumption. That means water use will need to be measured more consistently if operators are going to understand whether these systems are actually making a difference. Could data centres rely on less drinking water? As demand for AI continues to grow, there will likely come a point when operators will need to look beyond using drinking water as a means of cooling. Depending on the site and the type of cooling system being used, some options include rainwater harvesting, recycled wastewater, and other non-drinking water sources. Government research has already identified these as possible ways to ease pressure on public water supplies. Closed-loop cooling could also help, as it allows water to circulate through a system several times rather than constantly bringing in fresh supplies. The Environment Agency has pointed to this as a more water-efficient option and says reuse should be considered when new data centres are planned. There may also be opportunities to reuse some of the wastewater created on site. Water removed from cooling systems, for example, can sometimes be treated and fed back into the system instead of being discharged straight away. For future sites, the wider aim could be to get more use out of the water already available, rather than relying so heavily on fresh supplies each time. Calls for more transparency Mark Hall, Liquid Waste Removal Expert at BusinessWaste.co.uk, comments, “Data centres are becoming a bigger part of everyday infrastructure, and the relationship between them and water is becoming a bigger part of discussions. “As more sites are proposed, operators are likely to face more questions from local communities, councils, and regulators about the pressure they could place on water supplies. Being able to explain that is going to become increasingly important. “There needs to be more openness across the sector, because if we don’t know how much water data centres are using, it becomes much harder to judge what further growth could mean for local water supplies. “For businesses planning new sites, water should be treated as a key part of the decision-making process from the beginning. The availability of land and power will always matter, but access to water could also play a bigger role in where these facilities can realistically operate.”