If the US is 'substantially No. 1' in AI, the single name that most literally embodies that lead is Nvidia — its GPUs are the compute substrate every hyperscaler, lab and sovereign is fighting to secure. This is the purest equity expression of 'America wins the AI race' because Nvidia's data-center franchise scales directly with global AI capex.

The non-obvious step past Nvidia: the hyperscalers racing to be No. 1 don't want to depend on a single supplier, so they're designing custom AI silicon (TPUs, Trainium-class ASICs) — and Broadcom is the dominant merchant partner for those chips plus the high-speed networking that stitches AI clusters together. It captures the exact same buildout as Nvidia but from the custom-silicon and interconnect angle that scales as clusters get larger.

An AI boom 'bigger than the internet' is physically a data-center construction boom, and the binding constraint inside the building is power distribution and cooling of dense GPU racks. Vertiv is the leading pure-play in thermal management and power infrastructure for AI data centers — a picks-and-shovels name whose order book reprices harder than the broad tech index as capacity is added.

The deepest second-order consequence of AI 'bigger than the internet' is electricity: US data-center power demand is exploding and the grid is the true bottleneck. Constellation is the largest US nuclear operator — the source of the firm, carbon-free, 24/7 baseload that hyperscalers are signing multi-year deals for (including restarting Three Mile Island for AI compute). This is the AI trade routed through the power that feeds it.