Global spending on artificial intelligence (AI) infrastructure will exceed $100 billion by 2028, according to a report from IDC. In the first half of 2024, hardware infrastructure for AI was $31.8 billion.

The growth of AI infrastructure is primarily driven by investments in servers. In the first half of 2024, 89 percent of spending was on servers, an increase of 37 percent compared to the same period last year.

Servers with integrated accelerators are the most popular, accounting for 58 percent of total server spending, a category that increased 63 percent. IDC expects this share to continue to rise to over 60 percent by 2028, with average annual growth of 19 percent.

The cloud remains dominant

Cloud-based infrastructure remains dominant and will account for 65 percent of server spending by 2024. Hyperscalers and cloud providers continue to expand their capacity, while traditional companies lag behind in adopting on-premises AI infrastructure.

Storage also plays an important role, especially due to the need to manage large data sets for AI training and inference. In the first half of 2024, storage spending increased by 36 percent, with 56 percent of investments in cloud environments.

Regional distribution and future growth

The United States remains the largest market for AI infrastructure, with nearly half of spending occurring in the first half of 2024. This is followed by China (23%), Asia-Pacific (16%) and Europe (EMEA, 10%). IDC predicts that the Asia-Pacific region will experience the fastest growth over the next five years at 20 percent, followed by the US (16%), Europe (EMEA, 13%) and China (11%).

By 2028, global spending is expected to reach $107 billion. Cloud-based servers will account for 75 percent of the market, while servers with accelerators will account for 56 percent of total spending.

Challenges surrounding energy consumption

IDC notes growing concerns about the energy consumption of AI infrastructure. The energy consumption of European data centers could triple in five years. The energy demand of AI workloads worldwide is already approaching that of a small country.

In addition, the issue of water consumption must also be kept in mind. A one-hundred-word AI-generated email requires about three 0.5-liter bottles of water.

Many technology giants are being forced to adjust their climate targets. For example, Mirosoft representative Craig Cincotta told The Washington Post that the company “will work on data center cooling methods that completely eliminate water use.” But AI’s profit motive often seems to be in the foreground and environmental goals lag behind. For example, Google saw its greenhouse gas emissions increase by 48 percent in just five years due to its uncontrolled AI push.