AI needs more than advanced chips

An AI rack combines computation with memory, communication, power delivery and cooling control. The June 2026 SIA/Deloitte report describes those functions separately, rather than reducing the system to its fastest processor. Its process-node breakdown makes the distinction visible: different trays use different mixes of chip technologies. This is a representative architecture study, not evidence of the installed base across all AI data centers. The market implication is a research question about dependencies, not an unsourced claim about demand growth.

Mature nodes have a distinct role

A process node describes a generation of manufacturing technology; a smaller node label is not a universal ranking of usefulness. Figure 17 defines foundational or mature nodes as 28 nanometers and above. It assigns those dies a 38% share within the compute tray, 94% within the accelerator-interconnect tray, 19% within the management tray, 83% within the power tray and 100% within the coolant-distribution tray. Each share has its own tray’s die count as the denominator. They must not be added together or interpreted as shares of the entire rack. (SIA/Deloitte, June 2026, p. 35, figure 17.)

A system view

The diagram below groups functions described in the report. An accelerator performs computation; memory supplies working data; networking moves information between devices; power and cooling systems keep the hardware operating. These functions interact. The sequence is an editorial reading aid, not a wiring diagram, a manufacturing sequence or a claim that one function always comes before another. (SIA/Deloitte, pp. 11–29.)

Why node mix and value mix differ

The mature-node chart counts individual dies. A die is an individual integrated circuit; a package can contain more than one. Content value is a different measure, discussed elsewhere in the report. A tray with many mature-node dies need not command a large share of rack expenditure. Conversely, a lower-value function may still be essential to operation. Those distinctions prevent a common analytical shortcut: equating the smallest process node with all of the system’s economic importance. (SIA/Deloitte, pp. 33–36.)

A framework for examining exposure

For a supplier, the relevant questions are the function it serves, product qualification, manufacturing flexibility and the customers disclosed in its filings. The source alone cannot establish which company has won a socket or how an individual product is priced. It also cannot prove that a given node category is currently short of capacity. The chart should therefore be read as a map of technological breadth in the report’s modeled rack, not as a ranking of stocks or an estimate of future sales.