Chip Duopoly Shapes Google AI Supply

Broadcom (AVGO) and Marvell (MRVL) currently control roughly 80% of the custom chip market used by hyperscalers. In practice, this concentrates most incremental demand for specialized ASIC and AI silicon into two vendors, particularly for large cloud and data‑center contracts.

Google parent Alphabet (GOOGL) is expanding its custom AI efforts while already relying on Broadcom (AVGO) for TPUs and exploring new AI inference chips with Marvell (MRVL). With the supplier base so concentrated, positive developments in outsourced custom silicon largely reinforce existing relationships rather than reshaping the competitive field.

Historical parallels in semiconductors show that when one or two firms dominate a critical niche, as with Nvidia (NVDA) in AI accelerators or TSMC (TSM) in advanced foundry, new capex or design‑win headlines typically channel disproportionate benefit to those incumbents. Markets often treat such news as confirmation of an entrenched trajectory rather than a thesis‑changing surprise.

Applied to AVGO and MRVL, the current custom chip structure implies that additional AI or data‑center build‑out announcements are more likely to refine expectations than to rerate the group dramatically. For Alphabet (GOOGL), reliable access to a narrow set of high‑end suppliers supports execution on AI and cloud roadmaps, but the more direct revenue sensitivity resides with Broadcom (AVGO) and Marvell (MRVL) as the primary custom silicon vendors.