What are VCs saying about AI infrastructure spending

Analysts describe the current AI infrastructure build-out as a historic capital expenditure supercycle, dwarfing previous technology waves like cloud computing [3, 4]. Projections for the total investment are in the trillions, with one estimate suggesting a **$10 trillion total capital expenditure**, an order of magnitude larger than the cloud cycle [4, 16]. Current spending by large technology companies is estimated to be on a $400 billion annual run-rate, representing approximately 30% of their revenue [3, 7, 10, 21]. Other forecasts place the total supply chain capex at $1 trillion for the current year alone, with cloud-specific capex projected to exceed $1 trillion annually by 2027 [22, 26]. This massive, front-loaded investment by profitable mega-cap companies is seen as de-risking the ecosystem for application-layer startups by laying a robust foundation [7, 10], driven by a competitive race dynamic and a "winner-takes-all" mentality . The scale of this build-out is compared to a combination of the internet's development, the space race, and the Manhattan Project .

The primary bottleneck for scaling AI is rapidly shifting from semiconductor availability to fundamental physical infrastructure [1, 6, 14]. The most significant constraints are now the availability of electricity and the components required for data centers and power distribution [1, 12]. Specific shortages in the next 6 to 12 months are expected in structural steel, electricians, substations, transformers, and air chillers . This shift makes **power generation and distribution** the critical path for AI growth, moving the strategic focus from pure R&D to investment in energy and construction [12, 14]. This physical constraint represents a tangible vulnerability for Western AI development, as competitors like China are aggressively building out the necessary infrastructure to overcome these hurdles [6, 14].

This infrastructure-led boom is causing a significant shift in value accrual compared to previous technology cycles. Currently, value is concentrating in the infrastructure and semiconductor layers, a **reversal from the cloud era** where software and application companies captured the most value . This dynamic creates challenging unit economics for AI application startups, as a significant portion of venture capital flows through them directly to foundation model and hardware providers, creating a subsidized market with poor margins [6, 15]. While this massive infrastructure spending by incumbents de-risks the platform for startups [7, 10], it also makes it difficult for them to capture value, leading investors to prioritize "picks and shovels" infrastructure software companies that are less likely to be commoditized [11, 19]. The scarcity of human capital, with only about 30 teams globally experienced in training large models, also drives high-value "mega acqui-hire" acquisitions .

Despite the bullish long-term outlook, which pegs the AI market opportunity as an order of magnitude larger than the mobile and cloud wave , analysts identify near-term risks. The primary concern is a potential **slowdown in enterprise AI adoption** if companies fail to realize a clear and timely return on their massive investments . Some also foresee a potential overbuild of data center capacity . However, the consensus view is that high capital expenditures will continue for at least the next 12 to 15 months, fueled by intense competition and the sheer scale of the opportunity [27, 28]. The long-term demand is believed to be underestimated, as evidenced by the full utilization of even older-generation hardware, suggesting the infrastructure build-out is still in its early stages .