2025年数据中心趋势报告：电力、AI与行业进展

Expect theunexpected Foreword A wave of change caused by the surge in deployment of regenerative AI is upon us, with arace to build massive new facilities for training AI in the face of labour shortages, limitedskilled contractor resource and, in some locations, power constraints. Now we are bracingourselves for an even bigger wave, and what that will bring, as algorithms are trained anddeployed, and more businesses embrace AI. Adaptability and flexibility are the watchwords for 2025. The speed of technologicaldevelopment is already outpacing that of construction, which means that we will continueto see changes to scope – and some plans could be ripped up altogether. There are still a lot of question marks over exactly how the world is going to take hold of AIand use it day to day. And that will have a big impact on what we ask of our data centresand exactly where we need them to be located. The race is on to find new locations, new sources of power and sympathetic regulatoryregimes for those regions and governments looking to fulfil their digital growth ambitions. January, 2025By Scott SmythFounder and Group CEO Read on for ourtop ten predictions for 2025, in Soben’s third annual trends report. Data CentreTrends 2025 Click on the rightto explore each trend. 1|AI: pause for thought Big is getting biggerImpact of AI inference still unpredictableRobotics ups the ante Scaling up:hyperscale and beyond The increased processing power required by AI data centres means campuseswhich were considered large 10 years ago would today be thought of as aminnow. A hyperscale data centre was initially 20MW or more; five years ago,that had risen to perhaps 60 MW. Today, we are seeing plans for 1 GWcampuses, such as that planned by ECL, east of Houston, Texas – mooted tobe powered by hydrogen[2].“These AI facilities are huge, exponentially biggerthan cloud computing campuses,”saysJoe Cusick, Soben’s Group COO. The burgeoning use of generative AI has seen a wave of planneddata centre projects which is stronger than anyone predicted.Looking forward, demand for AI data centres could rise at a rate of30% per year between 2023 and 2030, according to McKinsey[1].This would mean that it accounted for 70% of all data centrecapacity, or 154 GW out of 219 GW in a midrange scenario. So, will big keep getting bigger? Not everywhere. For now, these projects are largely focused on providing capacity for machinelearning, training up the algorithms that will be used to deliver AI tools andapplications.“These can be located virtually anywhere, where there are largeplots of land with associated power, preferably renewable,”saysPieterSchaap, Group Development Director at Soben. Expanding horizons:inference and emerging technologies Perhaps three years from now, we can expect the next wave of AI-drivendemand to hit. There will be an increased demand for data centre space closeto urban and industrial centres, smaller scale inference facilities that need lowlatency so that data moves fast between the data centre and the machinery orequipment being operated; think autonomous vehicles such as Waymo taxis orEinride freight vehicles. Almost every organisation could benefit from the deployment of AI, and manymore will over the next five years. The EU wants to see 75% of firms using AIby 2030. And its deployment could offer huge efficiencies in the public sector,an ambition underlined by the UK government’s recently announced plans toturn the UK into an ‘AI Superpower’. With generative AI set to become business as usual for private and publicsector organisations, and embedded in almost every app we use via ourphones, there will be a need for inference data centres with low latencyconnections to where the apps are located. These could be located close toexisting cloud regions. The use of AI in healthcare, finance, and defence - already advanced comparedto other sectors - will ramp up. Applications range from medical diagnoses, drugdevelopment and AI-assisted robotic surgery in healthcare to loan appraisals,agent modelling which simulates individual behaviour, and fraud detection. Theenergy sector is catching up fast, with uses that include more precise demandforecasting and optimised storage and distribution of energy. And then there is the vision of robots as part of the everyday. The news that Nvidia has set its sights on robotics, promising to showcasetechnology for humanoid robots in the first half of 2025, accelerates the arrivalof that vision. Tesla too is pushing that narrative – the realisation of itshumanoid robot Tesla Optimus will follow on from the launch of its Cybercabby 2027 and later its 20-person capacity Robovan, says Elon Musk. Also on the horizon is quantum computing which, while promising huge leapsin processing speed and capacity, will require significant changes to datacentre infrastructure. Currently quantum computers require super-lowtemperatures to operate which would demand more energy for cooling. The