Marvell projects its custom AI chip business reaching $12 billion annually, with a potential Google deal valued up to $120 billion.

Marvell now expects $12 billion in custom AI chip revenue by fiscal 2029 as its Google partnership could generate up to $120 billion through 2033. The company has also lifted its fiscal 2028 revenue outlook to roughly $20 billion, above Wall Street expectations, as demand for custom silicon, networking, and data-center infrastructure accelerates. Marvell disclosed the expanded Google relationship in an SEC filing covering custom products tied to Google's TPU ecosystem, including inference accelerators, networking controllers, and near-memory compute.

The biggest long-term opportunity is the $120 billion cumulative potential. Marvell's expanded Google agreement could generate up to that amount in cumulative sales through fiscal 2033 if performance targets are met. Google also received warrants to buy nearly 59 million Marvell shares at $206.58 each, aligning a portion of the deal with Marvell's future execution. The forecast shows Marvell now sees custom silicon revenue reaching $12 billion by fiscal 2029, up from its previous $10 billion target.

Custom chips are increasingly becoming Nvidia's biggest alternative as hyperscalers shift investment. Google, Meta, Amazon, and other major cloud providers are developing proprietary accelerators optimized for specific workloads, particularly inference. This gives suppliers such as Marvell and Broadcom an opportunity to capture spending that might otherwise flow to general-purpose Nvidia GPUs.

Marvell has already generated about $2.17 billion of data-center revenue in its latest quarter, up 46 percent year over year, with the segment accounting for nearly 80 percent of total sales. The company is rapidly becoming an AI infrastructure business. Competition is intensifying as Broadcom expects its own AI semiconductor revenue to reach roughly $115 billion in 2027 and $230 billion in 2028.