thenextweb.com
|
ksl
|
|
Google is in discussions with Marvell Technology to design custom AI inference chips, adding a third supplier alongside Broadcom and MediaTek for its TPU line. No contract is signed yet and production is years out, but the logic is clear – inference costs now dominate as billions of daily queries run continuously, and Google wants to avoid Broadcom’s 70% market share in custom accelerators becoming leverage. Marvell stock rallied 50% year-to-date on the news cycle alone. The custom ASIC market is projected to hit $118 billion by 2033, and Amazon, Microsoft, and Meta have all been building or expanding their own silicon programs in parallel. Google treating chip supply like an automotive diversification strategy is a sign of how seriously hyperscalers now view inference economics.
