Data center infrastructure and operating costs projected to exceed $76 billion by 2028 driven by AI demand.
Generative AI has captured global attention with its remarkable capabilities—holding conversations, passing tests, developing research papers, and writing software code—yet these achievements come at a substantial cost in processing performance and power consumption. According to Tirias Research, generative AI data center server infrastructure plus operating costs will exceed $76 billion by 2028 on the current trajectory. This figure is more than twice the estimated annual operating cost of Amazon's AWS, which holds one third of the cloud infrastructure services market. The forecast accounts for an aggressive 4X improvement in hardware compute performance, but this gain is overwhelmed by a 50X increase in processing workloads, even considering rapid innovation in inference algorithms and efficiency improvements.
The computational challenge is driven by the architecture of neural networks, which execute primarily on servers accelerated by Graphics or Tensor Processing Units (GPUs or TPUs) designed for parallel matrix calculations. Each accelerator applies thousands of coefficient "parameters" to each "node" across layers containing thousands of nodes with thousands of connections. In Large Language Models, these nodes map to tokens—text language objects and symbols—and the history of previously generated tokens is used to assign probabilities and select likely next tokens. Future models such as GPT-4 are being trained on massive data sets to create neural networks exceeding one trillion parameters, requiring multiple accelerators and servers to execute, further driving costs upward.
Current GenAI applications span text generation, software code, and images, with emerging capabilities in video, sound, and 3D animation. OpenAI's ChatGPT is rapidly approaching 2 billion monthly visitors, while Midjourney, the popular GenAI art community, has over 15 million users. Tirias Research forecasts demand across three foundational capabilities: for text GenAI, demand for tokens is forecast to exceed 10 trillion by the end of 2023 with over 400 million monthly active users concentrated in developed markets; by the end of 2028, the forecast estimates over 6 billion users—approximately 90% of smartphone market penetration—with over 1 quadrillion annual tokens, representing a 100X increase. Image GenAI demand is forecast to increase significantly higher at over 400X to over 10 trillion images, driven by sophisticated video generation requiring sequences of thematically and visually connected images and complex prompting loops.