AI investing has become a herd sport, and AMD sails at the center of that stampede. This piece skips the rehashed GPU duel and points to four companies that collect revenue no matter who wins the next benchmark: Micron, Broadcom, Marvell and ASML. Each plays a distinct, often overlooked role in the AI supply chain, from memory to custom accelerators to the machines that etch silicon. If you want exposure to the AI buildout without debating GPU market share, these are the names to follow.
Everyone calls AMD the “NVIDIA alternative” and treats that label like a thesis. That consensus has pushed AMD into the spotlight and made the trade crowded, with retail accounts and analysts piling into the same narrative. Popularity can be a risk when the moat you need to beat runs deeper than a decade of software advantage.
CUDA is not just code; it’s an ecosystem that keeps edge cases within NVIDIA’s orbit, and hyperscalers are quietly funding custom silicon projects to sidestep merchant GPUs entirely. Those trends make trying to out-forecast a duopoly a high-stakes wager. The smarter route often lies with firms that sell into every potential winner, not just the loudest name on the scoreboard.
Micron is a textbook example of a supplier that benefits regardless of GPU branding. Fiscal Q1 revenue hit $13.64 billion, a 56.6% year-over-year jump, with non-GAAP EPS beating consensus. Cloud Memory revenue nearly doubled to $5.28 billion at robust margins, management guided sharply higher and order books reportedly stretch into 2027. As the only major U.S.-based memory manufacturer focused on HBM, Micron’s CEO calls the company an “essential AI enabler,” and that position translates into sticky demand.
Custom silicon is where hyperscalers are spending the big dollars, and Broadcom sits squarely in that lane. The company reported $8.40 billion in AI semiconductor revenue last quarter, up 106% year over year, with management forecasting even bigger numbers ahead. Broadcom builds the kinds of accelerators and Ethernet switches that get deployed at hyperscale, the plumbing that keeps data centers humming under heavy AI loads.
Marvell is the other design-partner play to watch, with data center revenue that now dominates the top line. Recent results showed $1.519 billion in data center sales, representing about 73% of revenue, and management says custom AI design activity is at an all-time high across dozens of opportunities. The Celestial AI acquisition was called “transformational” for optical interconnect, and that kind of capability matters when customers demand bandwidth at scale.
Then there’s ASML, the industrial backbone of modern chipmaking nobody easily replaces. Quarterly revenue landed north of $10 billion and the company carries a multi-decade backlog measured in tens of billions, forcing customers to accelerate capacity plans. ASML holds an effective monopoly on EUV lithography, and that position means virtually every advanced AI chip, regardless of whose logo ends up on it, requires ASML machines to be manufactured.
Put simply, the picks-and-shovels names collect the toll no matter which accelerator wins the next benchmark. Memory makers, custom-design partners, and equipment suppliers get paid whether the data center installs GPUs from AMD, NVIDIA, or bespoke silicon from hyperscalers. Those businesses can offer clearer revenue visibility and less theater than the headline-grabbing GPU race.
If the current market feels like repeating the same contrarian call after it’s become mainstream, consider where contracts, backlog and engineering partnerships are actually locking in cash flow. These supplier and design plays don’t need you to pick a GPU winner to participate in the AI megatrend, and that reality is worth a much closer look.
