More

        

          

    HomeCloud/NFVPower issues and recession talk limit EMEA data centre secondary market growth 

    Power issues and recession talk limit EMEA data centre secondary market growth 

    -

    Some EMEA emerging DC markets have hit pause but Cairo, Istanbul and Manchester power on in Q2 

    Higher rates of borrowing, coupled with limitations on local power networks and looming concerns of a recession, have temporarily dampened growth in several EMEA emerging data centre markets. 

    Global property company Knight Frank Data Centre Report for Q2 2023 examined Bucharest, Cairo, Casablanca, Istanbul, Johannesburg, Manchester, Munich, Nairobi, Vienna, Warsaw and Zurich and found total capacity across these markets has reached just over 1,577MW, following a Q2 uptick of 137.2MW. 

    The report, published in partnership with DC Byte, found growth between markets is varied, with strong growth registered in the Cairo, Istanbul, and Manchester markets. Istanbul saw the largest new supply at 59.2MW, primarily bolstered by Equinix announcing a new facility expected to deliver 28MW of IT capacity. Turkey has also commissioned the development of a new nuclear power plant, which will help meet the increasing power demands in the market. 

    Cairo registered a 37MW increase with UAE-based Khazna Data Centers announcing the development of a state-of-the-art facility costing $250 million. Also, with the development of the Medusa subsea cable expected to go live in 2025 east (Portugal, Spain France, Italy, Morocco), and 2026 west (Tunisia, Greece, Cyprus, Egypt), the authors expect to see more growth in the market, brought on by the promise of a more cost-efficient and high-speed data delivery network. 

    Not included in the original Knight Frank report was GPX Global Systems’ two data centres in Cairo. The first opened in 2007. A second 3,000sqm fully redundant Tier4 DC in Cairo was opened and fully operational in 2015. In Q2 [May] GPX announced a EGP2.7bn investment to expand its Cairo 2 data centre. The expansion will add an additional 12MW and 9,000sqm bringing the DC operator’s total equipped space in the Cairo market to 12,600sqm.

    Kao Data recently announced a new 40MW facility in Manchester, with Microsoft expanding into Leeds. These reflect a trend of data centre demand in the UK spilling North, with Manchester tipped to experience significant growth in upcoming years. 

    The rest are slower but new cables may change that  

    In Nairobi, the authors witnessed no new supply or major announcements this quarter. Casablanca experienced a similar absence. However, as with Cairo, Casablanca will also benefit from the Medusa Subsea cable, which will see an 8,700km 24-pair fibre optic cable offering 20Tbps of capacity per fibre pair. 

    Warsaw and Vienna recorded limited market activity in Q2, with power limitations placing constraints on new data centre development. However, both feature in Microsoft’s plans for developing new “Data Centre Regions”. In Warsaw, Microsoft have developed a 38MW Poland Central Campus, whilst in Vienna, Microsoft’s deployment in Achua is expected to stimulate growth in the market. 

    “The markets covered in this report are set to enter an exciting phase of growth as data localisation requirements become more pronounced and power constraints continue to stifle growth in the more mature markets across EMEA,” said DC Byte managing director (APAC) James Murphy.

    Generative AI set to reshape data centre market 

    The authors also examined the impact generative AI will have on data centre development going forward. DigitalBridge CEO Marc Ganzi last week claimed generative AI workloads will consume 80% of data centre power over the next 15 years. He suggested while the global public cloud is 13GW, AI will dwarf this heading to 38GW. 

    As a result of this huge processing power, current data centres may not be about to cope. AI applications typically use more than 30kW of power per rack, which is quite a jump from the average rack that uses 8-10kW. AI-based servers also require greater power density, which means more chips crammed into each rack and will unquestionably require greater cooling to handle the added heat. High cooling requirements raise power demands. 

    Telco racks certainly won’t cope with that level of power. However, a nuance is developing as Generative AI machine learning workloads typically come in two flavours – training and inference – with different data centre requirements for each, much like the centralised versus edge models that are developing as services are pushed closed to users.  

    The training aspect of generative AI uses the most amount of computing power, as it requires many advanced GPUs to run at capacity for long periods. Inference, on the other hand, involves generating predictions or decisions based on input data and does not require nearly the same amount of processing power, instead requiring low latency with proximity to the end user. 

    A tale of two DCs

    The authors believe specialised facilities for both the training side of generative AI and the inference side could develop. Each function has its own unique data centre needs and requirements, which could reshape the data centre market in upcoming years.  

    While AI is already driving existing hyperscale demand by around 20%, the authors are already seeing operators and hyperscalers expanding outside of the established clusters where there is a greater offering of land and power, highlighted by Microsoft’s recent purchase of a site in Leeds.  

    “In North America, Microsoft has also had to rapidly speed up their development of a new 750,000 sq. ft hyperscale data centre campus in Quincy, Washington, designed to support and accommodate artificial intelligence with three 250,000 sq. ft data centres,” they write.  

    They contend two data centre models may appear, even sitting side-by-side, where one serves the needs of AI training, and one serves the needs of AI inference. 

    The training model DC would be a larger data, more resource-intensive data centre requiring more advanced cooling techniques with greater capacity and high computing/storage power. The inference model DC would be a smaller data centre with less capacity and rack power density and different cooling infrastructure. The latter may bring more impetus to edge data centre business models, and a place telcos can play.

    “The rapid emergence of AI from a relatively unknown base six months ago to an everyday tool for work and home life has been the most significant data centre development since the creation of AWS public cloud in the early years of 21st century,” said Knight Frank data centre lead (APAC) Fred Fitzalan Howard. “Design, power and locational requirements vary immensely from traditional cloud or colocation facilities which is causing operators to rethink previous strategies in order to gain exposure to this next phase of data centre growth.”