Artificial intelligence has been on the rise for some time, but suddenly, seemingly out of nowhere, hit a meaningful inflection point. Uptake in AI-based applications has surged and the demand for infrastructure capacity to support these workloads has risen sharply. AI demand has pushed public cloud growth and spawned a new wave of infrastructure service providers specializing in hosting GPU-based infrastructure. AI-oriented startups, enterprises and even hyperscalers are looking for AI-ready infrastructure wherever they can get it and this demand has not surprisingly crossed over into the Internet infrastructure ecosystem. This new generation of GPU-oriented service providers is taking down colocation capacity to support their operations, hyperscalers are in need of more data centre capacity and support for higher densities, while some end users are moving their own infrastructure into colocation environments. Some are even running on-premise, but connecting back to any and all of these infrastructure locations.
Like the cloud, AI does not live in the sky, but in a data centre, and the capacity requirements and inevitable growth are coming at a pace orders of magnitude higher than previously seen. But while hyperscale cloud may precede it, the AI story promises to be vastly different. The impact of AI will be far-reaching. It is going to prime demand for data centre colocation and force operators to fundamentally rethink how they approach design, ramping timelines, capacity planning, operations, supply chains, procurement and even business development. Needless to say, the rise of AI is a net positive for the sector and set to create levels of demand that will kick off another wave of data centre development, eventually on a global basis. It is important to keep in mind that AI is almost purely value-add for the sector with few hints of cannibalization effects. AI is enhancing raw public cloud infrastructure with tools and services and creating a new generation of workloads that enhance, augment and expand, rather than replace, what has already been built.