Share with your CTO
Three AI infrastructure companies, NVIDIA, Broadcom, and Vertiv, posted earnings beats in Q2 2026 that confirm the capex cycle is still accelerating rather than plateauing. NVIDIA guided Q2 FY27 revenue to $91 billion at 75% gross margin. Broadcom expects AI semiconductor revenue to hit $16 billion next quarter, up 200% year over year. Vertiv, which supplies the power and thermal infrastructure those chips require, raised full-year guidance twice and now targets $14.2 billion in sales at 30-32% organic growth.
What this means for your business
The gap between compute demand and available supply isn’t closing, and that has a direct read-through for anyone currently negotiating data center contracts, custom silicon timelines, or power infrastructure buildouts. If your organization is mid-cycle on a GPU procurement or colocation deal, the pricing leverage you’re hoping for isn’t coming. Broadcom’s 200% year-over-year AI semiconductor growth projection means hyperscalers are locking in custom accelerator capacity aggressively, which pulls engineering resources and lead times away from everyone below them in the queue.
Vertiv’s numbers deserve particular attention from CTOs thinking about physical infrastructure. A 234% year-over-year jump in free cash flow, combined with 410 basis points of margin expansion, tells you this isn’t a company that merely benefits from AI spending. It’s a company where the AI buildout has fundamentally changed the cost structure of every data center project. Power density per rack is rising sharply as GPU clusters replace traditional server configurations, and thermal management (the systems that prevent chips from overheating at scale) is no longer a procurement afterthought. It’s a capacity constraint in its own right.
The article, published by 247 Wall St., which earns through retail investor traffic and is therefore writing for a portfolio audience rather than an enterprise one, frames all three names as August stock picks. That framing is worth discounting. The underlying data isn’t wrong, but the question for a CTO isn’t which stock to buy; it’s whether the supply signals these earnings reveal should pull forward a procurement decision or change a vendor dependency. Broadcom’s customer concentration risk, a handful of hyperscalers driving nearly all AI revenue, is the detail that matters most internally. If your architecture assumes continued access to commodity GPU supply while the hyperscalers absorb custom silicon, that assumption is getting harder to defend each quarter.
Concept deep-dive: Custom ASICs
A custom ASIC (application-specific integrated circuit) is a chip designed to do one class of task extremely well rather than a broad range of tasks adequately. Think of it as a specialized tool versus a Swiss Army knife. Hyperscalers like Google and Meta commission custom ASICs from Broadcom to run their specific AI workloads more efficiently than general-purpose GPUs allow. For enterprise buyers, the implication is that the chip supply chain is bifurcating: bespoke silicon for the largest players, and GPU allocation for everyone else.
Based on reporting from 3 AI Infrastructure Stocks That Could Make You Rich in August, originally published 2026-08-07 14:31:00.
