AI has pumped hyperscale capex, capacity – but how long can it last?

Hyperscale datacenter operators nearly tripled their spending on infrastructure over the past three years in response to the AI craze, while the amount of operational capacity added each quarter has increased by 170 percent, with little sign so far of any slowdown. Synergy Research has released its latest data tracking the growth of the hyperscale companies, which includes the big cloud providers Amazon, Microsoft, and Google, plus others such as Meta, Alibaba, Tencent, and Oracle. While the market has long enjoyed strong growth, Synergy's figures and charts show how it went into overdrive following the launch of ChatGPT towards the end of 2022. After this, it seemed like world+dog wanted to train their own generative AI models, and demand shot up for suitable IT infrastructure and the datacenters to house it in. As a result, quarterly capex reported by the hyperscale operators has inflated, growing by almost 180 percent and reaching $142 billion in the third quarter of this year, according to the firm's research. The volume of datacenter capacity added each quarter has also risen by 170 percent. The effect of this can be seen in the number of capacious bit barns operated by those hyperscale providers, which has now reached 1,297 worldwide, almost three times the number that existed back in 2018. During the same period, total operational capacity has increased more than fourfold, due to the datacenters themselves getting larger and larger. Where a typical facility might have previously supported tens of megawatts (MW) of IT infrastructure, this is now increasing to hundreds of MW and even gigawatts (GW) of capacity. Meta, for example, announced plans earlier this year for several multi-gigawatt datacenter clusters, including a 2 GW cluster called "Hyperion" that could scale up to 5 GW and will take up a space almost as large as Manhattan island. The US was already the largest datacenter market in the world, accounting for nearly half the global installed capacity. But it takes an even larger slice of the hyperscale operational capacity, now standing at 55 percent, according to Synergy, up from 52 percent three years ago, reinforcing its position as the dominant hub for massive bit barns. These figures have come from analyzing the datacenter footprint of 21 of the world's major cloud and internet service firms, including the largest operators in SaaS, IaaS, PaaS, search, social media, e-commerce and gaming, Synergy disclosed. As you might expect, the companies with the broadest footprint are the big three cloud providers, which together now account for 58 percent of all hyperscale datacenter capacity. Silicon photonics won't matter 'anytime soon' says Broadcom CEO Activist groups urge Congress to pause US datacenter buildouts Meta and Google turn to NextEra to feed insatiable datacenter power hunger HPC won't be an x86 monoculture forever – and it's starting to show Meanwhile, there are no apparent signs of slowing down. Synergy says that the pipeline of known future hyperscale facilities currently stands at 770 worldwide, which are at various stages of being planned, constructed, or fitted out. "Based on this analysis, Synergy has revised its five-year outlook upward for several core metrics. The firm now expects total hyperscale datacenter capacity to double in just over twelve quarters, underscoring the scale and speed at which AI-driven infrastructure investment is reshaping global cloud-related markets," said chief analyst John Dinsdale. However, there has been growing concern over what will happen to all this extra bit barn capacity if the AI investment bubble should burst. Financial giant Goldman Sachs warned recently that datacenter investments may fail to pay off if the industry is unable to monetize AI models, in which case capacity could actually decline between now and the end of the decade. ®