Source: Microsoft is in talks to design future custom chips with Broadcom, which would involve Microsoft switching its business from Marvell
Microsoft is in talks to design future custom chips with Broadcom, which would involve Microsoft switching its business from Marvell …
Context & Ripple Effects
Microsoft's reported supplier discussions extend a chip-design effort that dates to its work on in-house cloud-server and Surface chips. The company also previously signaled an intention to manufacture a Microsoft-designed chip on Intel's 18A process, making the Broadcom talks a potential change in design partner rather than a new move into custom silicon.
The reported switch matters because it would reallocate a major cloud customer's custom-chip work from Marvell to Broadcom. Subsequent coverage of Broadcom's proposed financing for OpenAI chip production underscores how custom-chip partners can become central to AI infrastructure supply arrangements.
First-order effects
- Broadcom could gain Microsoft's future custom-chip design business if the talks produce an agreement, while Marvell risks losing that work.
- Microsoft would gain a prospective new partner for designs it controls, but its existing custom-chip roadmap would need to transition away from Marvell.
Second-order effects
- A Microsoft win would strengthen Broadcom's position in bespoke data-center silicon and put pressure on Marvell to defend or replace the potential lost volume with other cloud customers.
- The shift could give Microsoft greater leverage over design-service terms and production planning by making partner choice an active competitive variable rather than a fixed supplier relationship.
Third-order effects
- If large cloud operators continue pairing in-house designs with specialized partners, value in AI infrastructure will increasingly sit in design enablement and supply coordination—not only in merchant chips.
- The pattern could further divide AI hardware strategies between companies buying standardized accelerators and those able to fund proprietary silicon programs, though this reported negotiation alone does not establish a completed transition.
The trend: Hyperscalers are turning custom silicon into a strategic procurement lever, selecting chip-design partners to better align AI infrastructure with their own roadmaps.