AI Profit Pools Shift to Structural Supply Chain Bottlenecks
As artificial intelligence demand strains physical and technical limits, capital is flowing toward the five supply chain layers that money cannot quickly replicate, reshaping global investment strategies.
The profit pool in artificial intelligence is migrating away from general-purpose graphics processors toward structural supply chain bottlenecks. Investors are increasingly targeting the specific layers of the AI infrastructure chain where capital cannot quickly relieve capacity constraints.
The defining test for these assets is whether a constraint can be rapidly solved by funding. While software capacity can be added in an afternoon, physical realities like 128-week lead times for large power transformers or five-year median waits for United States grid connections create durable moats.
Custom silicon and high-bandwidth memory have emerged as primary gating inputs. Broadcom reported $10.8 billion in AI semiconductor revenue for the quarter ended 3 May 2026, a 143 per cent year-on-year increase, guiding the current quarter to $16 billion. Similarly, Micron saw quarterly revenue jump to $41.46 billion from $9.30 billion a year earlier, backed by $22 billion in strategic customer commitments, including $18 billion in upfront cash deposits.
Networking represents a constraint many market participants have yet to fully price. Nvidia’s networking revenue reached $14.8 billion in the quarter ended April 2026, up 199 per cent year on year and expanding faster than its graphics processor sales. Because copper fails beyond roughly one metre at modern speeds, optical networking is mandatory, driving Arista’s quarterly revenue up 35 per cent to $2.71 billion and Astera Labs’ revenue up 93 per cent to $308 million at a 76 per cent gross margin.
The physical power layer remains the most intractable bottleneck. In the PJM grid, capacity for the 2026/27 delivery year cleared at the regulated price cap of $329.17 per megawatt-day, with the subsequent auction hitting $333.44. GE Vernova booked $2.4 billion in data centre electrification orders in the first quarter of 2026 alone, while a gas turbine ordered today faces a lead time extending to 2031, and Vertiv grew quarterly revenue 30 per cent to $2.65 billion providing critical cooling.
The Bottleneck Premium
Valuation discipline remains critical, as a bottleneck is only valuable if not already reflected in the share price. As of 14 July, Broadcom trades at 24.4 times forward earnings with a PEG ratio of 0.53, whereas Vertiv trades at 44 times with a PEG of 1.36. Investors must assess how long it would take to relieve these constraints and how much of that eventual relief is already priced into the market.
For international portfolios, accessing these layers requires looking beyond domestic markets. SK hynix recently listed American depositary shares on the Nasdaq, raising $26.5 billion in the largest US listing by a foreign company on record. This move, alongside a 17 per cent year-on-year rise in equity and debt remittances under India’s Liberalised Remittance Scheme in April 2026, highlights how global capital is adapting to reach these critical industrial chokepoints.
