Financial analysts are urging investors to pivot from AI model-building companies toward the providers of AI infrastructure.
This shift matters because the current demand for compute power is outstripping the supply available from any single company, creating a tighter-supply environment with potentially higher margins for those providing the underlying hardware and services.
During a Trader Talk episode, Willy Lee, a principal at NeoSeller Capital, said, "Demand for compute is far outstripping the supply that any company can provide." Lee and other guests, including Kevin Mahn, chief investment officer of Hennion & Walsh, said that the infrastructure layer offers better upside than the hype surrounding AI software models.
Data suggests significant growth in the hardware sector. Vertiv has reported a backlog of $15 billion, which represents a 109% increase year-over-year [1]. According to industry data, power and cooling requirements now account for roughly 80% of AI datacenter spending [1]. This indicates that the physical requirements of running AI are becoming a primary cost driver.
However, the infrastructure narrative is split between physical hardware and cloud services. While hardware is surging, Google maintains a $460 billion cloud backlog and an operating margin of 36% [2]. This highlights a tension in the market between those investing in the physical cooling and power systems, and those investing in the cloud platforms that manage the compute.
Market volatility has already hit some hardware providers. Super Micro Computer saw its stock fall 41% in a single month [2]. Despite this volatility, analysts said the rotation into infrastructure remains a strategic move as the industry struggles to keep up with the physical demands of artificial intelligence.
“Demand for compute is far outstripping the supply that any company can provide.”
The transition from 'AI hype' to 'AI infrastructure' reflects a maturing market. Investors are moving away from speculative software applications and toward the 'picks and shovels' of the digital age—power, cooling, and cloud capacity. This suggests that the primary bottleneck for AI growth is no longer just algorithmic innovation, but the physical limits of electricity and thermal management in data centers.
