Every dinner-party conversation about AI investing starts and ends with Nvidia. That’s understandable – it’s the biggest number, the most recognizable name, the easiest story to tell. It’s also, at this point, the most crowded trade in the group, with 96.8 percent of covering analysts already bullish and essentially no room left for a positive surprise nobody’s priced in.
The more interesting positioning, at least to me, sits one and two layers down the supply chain, where the businesses have real pricing power but far less retail attention.
Start with Taiwan Semiconductor. TSM trades at $437.68, a $2.27 trillion market cap, at a trailing P/E of 36.35 times – essentially in line with Nvidia’s own multiple, despite TSM sitting at the actual physical chokepoint of the entire AI hardware buildout. Every advanced-node chip that Nvidia, AMD, and the custom-silicon teams at the hyperscalers need gets fabricated through a supply chain TSM effectively controls at the leading edge. That’s about as close to a toll-booth position as exists anywhere in this industry, and yet the market is pricing TSM at a multiple comparable to – not materially higher than – the fabless design houses that depend entirely on TSM’s capacity allocation decisions.
Broadcom is the other name worth serious attention, and it’s more expensive on paper: $399.74 a share, a market cap around $1.90 trillion, trading at 49.17 times trailing earnings. On forward-looking annual figures the multiple actually runs hotter – 77.5 times earnings and an EV/EBITDA of 52.6 times – alongside a price-to-sales ratio that’s climbed to 27.4 times. Analyst conviction here is nearly as strong as it is on Nvidia: 92 percent of 25 tracked ratings are bullish, with an average price target of $504.16. Broadcom’s gross margin sitting at 67.8 percent reflects a business mix that’s shifted heavily toward custom AI accelerator design work for hyperscalers building their own silicon – a segment that, if anything, benefits from customers trying to diversify away from Nvidia dependency, not from customers buying more Nvidia GPUs.
That’s the structural insight that gets lost in the Nvidia-centric framing of this entire trade: TSM and Broadcom don’t need Nvidia to keep winning in order to keep winning themselves. If hyperscalers succeed in reducing Nvidia dependency by designing more custom silicon in-house – the exact scenario that shows up as a bear case in every Nvidia research note – TSM still fabricates most of that custom silicon, and Broadcom is one of the primary design partners helping build it. The picks-and-shovels layer of this cycle has a hedge built into its own business model that the marquee name at the top of the stack simply doesn’t have.
None of this makes TSM or Broadcom “cheap” in any absolute sense – a 49x trailing multiple isn’t a value stock by any definition, and Broadcom’s forward multiple pushing past 77x deserves the same scrutiny anyone applies to Nvidia. But the risk-reward math is different when a business benefits from multiple competing paths the AI buildout could take, rather than betting everything on one company’s continued market dominance. That diversification of outcome, more than any specific multiple, is what makes the supply chain layer worth more attention than it currently gets.
