Big Techs invest billions in datacenters

Two tech giants recently announced billion-dollar investments in datacenters. Amazon plans to spend around $150 billion over the next 15 years to guarantee the capacity to process the growing demand for Artificial Intelligence applications and other digital services. Not to be outdone, Microsoft and OpenAI, its strategic partner in the field of Artificial Intelligence, are working on a datacenter project that could cost up to $100 billion, including a supercomputer called Stargate, scheduled for launch in 2028.

According to Bloomberg, the investments are strategic to maintain Amazon’s leadership in the cloud services market, which is more closely followed by Microsoft and then Google. According to estimates by Synergy Research Group, Amazon’s share of the global cloud infrastructure market stood at 31 per cent in the fourth quarter of 2023, down from 33 per cent a year earlier. In second place is Microsoft, slowly increasing its share of the market by reaching an all-time high of 24 per cent in the same period. Google is in third place with 11 per cent.

Amazon has already earmarked $148 billion over the last two years to build and operate datacenters in various regions of the world, including Mississippi (USA), Saudi Arabia, Malaysia and Thailand. It also plans to improve other datacenters already built, for example in the US states of Virginia and Oregon, recognised for their attractive energy costs and tax incentives.

As reported by The Information, Microsoft and OpenAI are going head to head in this billion-dollar dispute. Microsoft will probably finance the project, which is expected to be 100 times more expensive than some of the largest datacenters ever built.

The Stargate supercomputer, which will have millions of specialised chips and will be installed in the United States, is intended to be the largest in a series that the two companies will build over the next six years. According to  Reuters, Microsoft and OpenAI have devised supercomputers in five phases, with Stargate being part of the fifth. Microsoft is working on a smaller, fourth-phase supercomputer for OpenAI, which should be launched in 2026.

Microsoft and OpenAI are in the middle of the third phase of this five-phase plan, with a significant part of the cost for the next two phases involving the acquisition of the necessary AI chips, the article said.

Microsoft made billion-dollar investments in OpenAI, the developer of ChatGPT, last January. The partnership between the two companies is under scrutiny by the European Union for violating the region’s merger laws, as well as by regulators in the United States and the United Kingdom. In addition, Elon Musk, co-founder and former executive of OpenAI, has already sued the company for seeking profit and diverging from its original mission of being a non-profit organisation. OpenAI hit back by publishing emails from Musk that seemed to show his interest in making money to fund his AI ambitions.

Is so much investment needed?

Duel of the titans aside, there are those who wonder whether all organisations need to invest in datacenter infrastructures to accommodate the growing demand for Artificial Intelligence (AI) applications. It might seem at first glance that the more you invest in AI, the more datacenter capacity you’ll need, but that might not be true, according to a column on the DataCenter Knowledge website.

There are four reasons why the impact of AI on the datacenter industry will end up being limited, according to columnist Christopher Tozzi:

1) AI infrastructure may be needed temporarily: Many AI use cases will not require specific infrastructure on a permanent basis, needing, for example, computing power only during training. After this phase, you won’t need as much datacenter capacity until the next model is trained.

2) AI infrastructures are already available: AI infrastructure and datacenter space can be purchased from major IaaS providers such as Amazon, Microsoft and Google. For example, instances of virtual machines can be used to carry out AI training.

3) Few companies will develop their own AI models: Training and deploying AI models from scratch is an arduous task, so very few suppliers are likely to do it and provide the models for other companies to use without the need to acquire AI infrastructures of their own.

4) The AI boom phase will eventually give way. At the moment, Generative AI is the hot topic and is putting pressure on companies to invest in AI solutions. But in five or 10 years’ time, most companies’ AI strategies will probably have matured. By then, there will be time to prudently expand datacenter capacity if necessary.