Alphabet and Amazon are committing roughly $420 billion to artificial intelligence infrastructure this year, a combined outlay that funnels directly into the supply chains of four semiconductor companies. The spending plans, outlined in recent guidance, underscore how dependent the largest cloud operators have become on a narrow set of hardware providers.

The spending breakdown

Alphabet has signaled capital expenditure between $195 billion and $205 billion for 2026, while Amazon raised its own target to about $220 billion from an earlier $200 billion figure. The increase at Amazon was driven explicitly by surging memory chip costs, according to the source material. Together the two budgets represent a single-year commitment that exceeds the gross domestic product of many mid-sized economies.

Nvidia and Broadcom split the compute layer

Nvidia remains the default choice for AI compute, with cloud customers treating its processors as the baseline for portability across providers. Broadcom, meanwhile, has secured a parallel revenue stream through its partnership with Alphabet on the Tensor Processing Unit, a custom accelerator designed for specific workloads. Broadcom’s management has told investors that its custom AI semiconductor division could generate more than $100 billion in annual sales, up from $10.8 billion in the most recent quarter. Nvidia, for its part, has projected that total hyperscaler AI spending will surpass $1 trillion next year.

Memory shortage lifts Micron and Sandisk

The memory market tells a different story. Prices have risen sharply because fabrication capacity has not kept pace with demand from the same hyperscalers buying compute. Sandisk disclosed that two-thirds of its recent revenue growth came from higher selling prices rather than volume gains. Micron’s leadership has stated the supply deficit will persist into 2028, implying several more years of pricing power for both memory vendors.

The timeline stretches to 2028

The source material frames the four stocks as a basket play on a multi-year capital cycle. Whether the spend sustains at these levels depends on whether cloud customers continue to absorb the resulting capacity, a question the current order books do not yet answer.