Available in April, 2026
This report examines the transformation of AI infrastructure from the server-based data centers of today into integrated production systems designed to “manufacture” computational intelligence at scale. AI factories take a completely new direction. Rather than treating an AI factory simply as a very large GPU cluster, this report analyzes it as a new computing architecture in which semiconductors, high-bandwidth memory, advanced packaging, networking, storage, software, electrical infrastructure and cooling planned together. Although a growing number of large firms are involved in the birthing of AI Factories, NVIDIA is understandably and especially enthusiastic – AI factories may contain as many as millions of GPUs. NVIDIA’s current Rubin architecture illustrates the direction in which the AI factory is pointing.
CIR’s understanding is that the concept of the “AI factory” is increasingly to be taken literally and should be framed in industrial language. Thus, in the AI Factory, managers will talk about such measures as tokens/sec, tokens/watt, cost/token, utilization and uptime. Also, as if to add to the industrial theme the designer of the AI factory must take into consideration the emergence of novel power and cooling technologies. This means that, for example, transformers, switchgear, power electronics, energy storage, generation, grid infrastructure, liquid cooling, heat rejection and data-center control systems also become part of AI factory. The same can be said of software, which will increasingly coordinate these physical systems with computing workloads; emerging AI-factory architectures already contemplate dynamically managing compute, networking, cooling and electrical demand within a fixed power envelope.
For investors and senior executives, the report’s objective will therefore be to identify where economic value, and technological bottlenecks migrate as AI scales. NVIDIA is an obvious beneficiary, but perhaps the more interesting question is which markets become disproportionately valuable because they constrain the amount of intelligence that can be produced from a given dollar, rack or megawatt. Those constraints could create major opportunities in memory, advanced packaging, networking silicon, optical interconnects, power equipment, cooling, specialized inference architectures and AI infrastructure software—while simultaneously exposing portions of today’s server and accelerator markets to commoditization or architectural displacement.
The report will ultimately treat the AI factory not as a data-center trend but as the emergence of a new industrial system: one that converts semiconductors + data + electricity into intelligence, with potentially profound consequences for the structure of the semiconductor, computing, energy and capital-equipment industries. All of this is covered in this new report.
