Infrastructure Market: IaaS for AI Workloads Becomes "Disruptive Growth Engine"

According to Gartner, AI-optimized infrastructure services are developing into “the next disruptive growth engine” in the infrastructure market. Analysts at the US market research firm estimate that global end-user spending on AI-optimized Infrastructure as a Services (IaaS) will increase by 146 percent to a total of 18.3 billion US dollars by the end of 2025. For 2026, a doubling of the market volume to an expected 37.5 billion dollars is forecast.

While the growth rate will lose momentum in the coming years, reaching only 34 percent in 2029 with a volume of just under 109 billion dollars, it will still significantly outperform the growth rates of ordinary IaaS spending. The share of investments in AI-optimized infrastructure services in total infrastructure spending is expected to double this year to just under nine percent. By 2029, market researchers expect an increase to almost 22 percent.

More than half for inference

“As companies increasingly expand their AI and GenAI applications, they need specialized infrastructures such as GPUs, Tensor Processing Units (TPUs), or other AI ASICs, complemented by high-speed networks and optimized storage solutions for fast parallel processing and efficient data transfer,” Gartner analyst Hardeep Singh interprets the market situation. It is becoming increasingly difficult to meet these requirements with classic, CPU-based IaaS offerings.

With the increasing prevalence of artificial intelligence across all industries, market researchers observe that inference workloads, in particular, are becoming a dominant force behind the growing demand. According to forecasts, investments in inference-focused applications will account for 55 percent of all AI-IaaS spending in 2026, amounting to 20.6 billion dollars. By 2029, this share is expected to rise to over 65 percent.

“Unlike training, which requires intensive, large-scale computing cycles during model development and ongoing updates, inference takes place continuously,” explains Singh, why the expenses for applying models, for example, for recommendation engines or fraud detection systems, exceed those for training.

(olb)