Microsoft has embarked on an ambitious plan to significantly increase its data-center capacity, aiming to support the burgeoning demand for artificial intelligence (AI) technologies. An internal presentation leaked to Business Insider reveals that Microsoft has already added more than 500 megawatts of new data-center capacity since July 2023. The company now boasts over 5 gigawatts of total installed data-center capacity, a milestone that underscores the rapid growth in its infrastructure to accommodate the needs of generative AI and large foundation models.
Microsoft’s expansion of its datacenter capacity aims to triple the rate at which it adds new space in the early part of its fiscal year 2025. This move is in anticipation of the burgeoning demand for artificial intelligence (AI) services, particularly in the realm of foundation models and generative AI, which require substantial computational power.
Generative AI, exemplified by technologies like ChatGPT and GPT-4 developed by OpenAI, requires extensive data processing capabilities. Training these models involves analyzing vast datasets and fine-tuning the algorithms, tasks that demand significant computational power, primarily provided by GPUs (Graphics Processing Units) housed in large data centers. These facilities, due to their energy consumption, are measured in megawatts and gigawatts, akin to the power usage of small countries.
Doubling Down on Future Growth
Looking ahead, Microsoft has outlined plans for even more aggressive expansion. The company intends to double its new data-center capacity in the latter half of its fiscal year, running into mid-2024. Furthermore, in the first half of the 2025 fiscal year, Microsoft aims to achieve a threefold increase in new data-center capacity, equating to the delivery of over 200 megawatts of capacity every month. Such rapid expansion is indicative of either a response to surging demand or a strategic move to secure future power capacity in a competitive market.
Shaolei Ren, a professor at the University of California, Riverside, commented on the pace of Microsoft’s expansion as “astonishing.” He highlighted the implications of such growth, including the potential environmental impact due to increased carbon emissions and water consumption associated with large data-center operations.
Sustaining AI Workloads with Record GPU Capacity
To support the intensive workloads demanded by AI applications, Microsoft has significantly increased its GPU capacity. In the second half of the previous year, the company achieved a “record-level GPU capacity,” more than doubling its total installed GPU base. This expansion has extended Microsoft’s AI clusters to 98 locations globally, indicating a robust infrastructure capable of handling complex AI tasks and computations.
In preparation for this expansion, Microsoft is “securing a record number of GPUs” to manage the AI workloads effectively, having more than doubled its total installed base of these accelerators in the last six months of the previous year. This strategy aligns with Microsoft’s previously announced commitment to invest “many billions of dollars” in building extra capacity to cater to a fast-growing clientele interested in running generative AI projects.
The push to expand datacenter capacity is not limited to Microsoft alone; other hyperscalers are also enhancing their infrastructure to meet the increasing demand for AI-powered applications. According to a report from Synergy Research Group, the total capacity of facilities operated by hyperscalers has doubled in just four years, with projections indicating another doubling in the next four years. As of the end of 2023, the United States hosts 51 percent of the global capacity, followed by Europe and China with 17 percent and 16 percent, respectively. The major players in this space include Amazon, Microsoft, and Google, which collectively account for 60 percent of all hyperscale datacenter capacity.
Last Updated on November 7, 2024 8:55 pm CET