Marvell Technology shares rose 6% on Wednesday after the chipmaker disclosed a deal giving Google the right to buy up to $12.2 billion of its stock, tied to custom AI chip purchasing targets stretching through fiscal 2033. The securities filing sets the option at 58,970,907 shares priced at $206.58 apiece. Broadcom, Google's longtime custom-silicon partner, fell about 5% on the news.
The equity arrangement is unusually large for a supplier-customer relationship and effectively lets Google take a substantial stake in Marvell if it hits volume commitments over the next several fiscal years. Marvell said the expanded agreement covers products that "attach to the [tensor processing unit] ecosystem," naming AI inference accelerators along with storage and network interface controllers. That positions Marvell as a broader silicon partner to Google's TPU stack rather than a swap-in replacement for the TPU itself.
The structure — warrants or share-purchase rights tied to purchasing targets — is the same playbook OpenAI and other AI buyers have used to lock in supply while giving suppliers upside. For Marvell, $12.2 billion at a fixed $206.58 strike is a floor of demand visibility running to 2033, an eight-year commitment window that is rare in the merchant silicon business.
Key facts
- 01Google can buy up to 58,970,907 Marvell shares at $206.58 apiece, worth up to $12.2 billion.
- 02The stake vests against Google purchasing targets running through Marvell's fiscal 2033.
- 03Marvell shares rose 6% on the disclosure; Broadcom, Google's incumbent custom-chip partner, fell about 5%.
- 04Expanded agreement covers products that attach to Google's TPU ecosystem, including AI inference accelerators and storage and network interface controllers.
Google has spent the last decade building its custom AI accelerator effort primarily with Broadcom, which co-designs the TPU line. That partnership was itself expanded in April. Bringing Marvell in at this scale signals Google is diversifying its custom-silicon supply base rather than consolidating further with a single partner, and it explains Broadcom's 5% drop even though the April expansion left Broadcom's core TPU role intact.
This is not the first sign of the Marvell-Google tie-up. In April, Marvell shares also popped after reporting surfaced of a chip collaboration with Google for AI workloads, with the deal said to include a TPU and a memory processing unit. Wednesday's filing puts a dollar figure and a fiscal-year runway on what had been described in outline.
The broader context is that Google, Amazon, Meta and Microsoft have all been building custom silicon for AI workloads to reduce dependence on Nvidia, whose data-center GPUs still command the majority of AI training and inference spending. Custom accelerators from the hyperscalers are the most credible medium-term challenge to that position, and Marvell — which supplies the connectivity, storage controllers, and now inference silicon that surround those accelerators — sits directly in the flow of that capex.
For Marvell, the deal is a validation of its AI-adjacent product roadmap at a moment when investors have been trying to sort which merchant chip vendors are structural winners of the hyperscaler custom-silicon wave versus which are cyclical beneficiaries. A named, multi-year, $12.2 billion commitment from Google is about as clean a signal as the market gets. The 6% pop reflects that, though it is smaller than the reaction to April's initial report — some of the upside was already priced in.
The risk is that share-purchase options tied to purchasing targets are not the same as booked revenue. Marvell has to actually deliver against Google's volume targets through fiscal 2033 for the full $12.2 billion in equity to be exercised, and the specific products in the TPU-attached category — inference accelerators, storage controllers, NICs — are competitive markets. Broadcom, still Google's primary TPU partner, is not being displaced, and the dilution implied if Google exercises the full option is material for existing Marvell holders.
The read-through for the AI chip market is that hyperscalers are increasingly willing to trade equity for supply certainty, and that Nvidia's alternatives now include not just Broadcom-Google TPUs and AWS Trainium but a widening bench of merchant partners taking equity-linked positions in the hyperscaler supply chain. Marvell just moved from adjacent supplier to strategic Google partner on paper, and Broadcom's 5% drop is the market pricing in that the custom-silicon pie, while growing, is now being sliced more than one way.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
