AI Infrastructure's Hidden Costs: Power and Cooling Emerge as Dominant Investment Areas

The narrative around AI infrastructure has long been dominated by the quest for faster, more powerful chips. However, recent financial disclosures from key players in the data center ecosystem, Vertiv and Bloom Energy, reveal a significant reorientation of investment. These companies, specializing in power and thermal management, have reported substantial revenue growth, indicating that the bulk of AI infrastructure spending is now flowing into the often-overlooked areas of power and cooling. Analysts estimate that chips account for only about 25% of total AI infrastructure spending, with the remaining 75% dedicated to the essential, yet less glamorous, components like power management, cooling systems, and data center construction itself. This shift profoundly impacts anyone involved in deploying or managing AI workloads. For DevOps and cloud engineers, it means that infrastructure planning must prioritize energy efficiency and advanced cooling solutions with the same rigor previously applied to GPU selection. The sheer power demands of modern AI models, particularly large language models and generative AI, are pushing existing data center capabilities to their limits. Companies like Vertiv, which reported a 30% year-over-year revenue jump to $2.65 billion in Q1 2026, and Bloom Energy, with a dramatic 130% revenue increase to $751 million and a net profit of $71 million, are directly capitalizing on this demand. Their success underscores that the physical environment supporting AI is now as critical, if not more so, than the computational hardware itself. This trend is a natural evolution within the broader context of cloud and AI infrastructure. As compute density increases exponentially, the challenges of power delivery and heat dissipation scale commensurately. This isn't a new problem for data centers, but AI's unique workload characteristics—sustained high utilization of powerful accelerators—exacerbate it. The pivot of former crypto mining operations, such as IREN (formerly Iris Energy) and Hut 8, into AI data center hosting further illustrates this point. These entities already possess the critical assets: massive power access, land, cooling infrastructure, and operational expertise in running hardware at scale. Their strategic shift highlights that the fundamental requirements for high-performance computing, whether for blockchain or AI, converge on robust, scalable physical infrastructure. The "capex-light" approach favored by some investors, as noted in other market analyses, also implicitly acknowledges the immense capital expenditure required for the physical buildout, pushing some to seek value in less capital-intensive parts of the AI value chain. For practitioners, this means several concrete implications. Firstly, when evaluating cloud providers or designing on-premise solutions, scrutinize their power and cooling capabilities, not just their GPU offerings. Look for advancements in liquid cooling, efficient power distribution units (PDUs), and sustainable energy sourcing. Secondly, cost optimization strategies for AI should extend beyond software and model efficiency to include infrastructure-level power consumption. Monitoring power usage effectiveness (PUE) and developing strategies to reduce it will become increasingly important. Finally, the growing backlog at companies like Vertiv, exceeding $15 billion, suggests that lead times for critical power and cooling equipment could lengthen. Proactive planning and partnerships with specialized infrastructure providers will be crucial to avoid bottlenecks in scaling AI deployments. The focus is shifting from merely acquiring compute to ensuring that compute can be reliably and affordably powered and cooled.