The AI gold rush has sent Nvidia’s market cap soaring to $2.7 trillion, but Broadcom has quietly outperformed its flashier rival over the past year. Thanks to hyperscalers like Google, Broadcom’s AI-related revenue has exploded from nothing to a quarter of its business—and is set to hit 30 per cent by June, growing at a blistering 40 per cent per year.

While Nvidia dominates the GPU market with its off-the-shelf AI accelerators, Broadcom is carving out a niche by helping data centre giants design custom silicon—XPUs built for specific workloads. Google is already a major customer, and Broadcom’s CEO, Hock Tan, claims four more hyperscalers are lining up to join the party.

The trade-off between bespoke and general-purpose AI chips isn’t straightforward. While custom silicon can be more efficient, integration costs can affect those savings. Mizuho Securities estimates that by 2027, tailored chips could make up a fifth of the $350 billion AI accelerator market, but Broadcom faces stiff competition from Marvell, MediaTek, and Alchip.

If there’s one thing Broadcom has in abundance, it’s ambition. Since 2016, the company has spent nearly $100 billion on acquisitions and would have doubled that if Donald [hamburger eating surrender monkey] Trump hadn’t blocked its bid for Qualcomm in 2018.

In December, Broadcom chief Hock Tan [pictured] teased that his top three customers would pump up to $90 billion into AI investments by 2027, with another four possibly matching that figure. How much of that will flow to Broadcom is anyone’s guess, but the cocaine nose jobs of Wall Street are buying the hype.

At 15 times this year’s forecast sales—twice its 10-year average—Broadcom’s valuation is flying high. Its stock is up 16 per cent since November’s US election, while Nvidia has slid by the same margin. There’s room for both in the AI race, but for now, Broadcom is proving that sometimes, the quiet ones make the loudest gains.