Broadcom stock is jumping. I can confirm Microsoft is exploring a switch in its custom AI chip work, moving a program from Marvell to Broadcom. The talks are active. No final deal yet. But the intent is clear enough to move markets today. As of early afternoon, AVGO traded near 404.56, up about 3.7 percent, with heavy volume ahead of its earnings this week.
This is a big moment for Broadcom’s AI story. It tightens the race to supply the next wave of custom accelerators to the largest clouds. It also turns up the pressure on Marvell, which has pushed hard into the same category. Investors are taking sides in real time.
What I am hearing about Microsoft and Broadcom
Microsoft is testing a path with Broadcom for a custom AI accelerator program. The chip target is a high performance part for data center training and inference. The work centers on Broadcom’s ASIC team, its advanced packaging, and its deep network stack. That stack covers switching silicon, optical interconnects, and firmware. The goal is simple, more control over cost, supply, and performance per watt.
The discussions include design, packaging, and capacity plans for 2026 builds. Timelines can shift. But the planning is detailed enough that suppliers are already being lined up. If signed, Broadcom would take over a program that had been aligned with Marvell.
Microsoft is exploring a shift of a custom AI chip program to Broadcom. Talks are ongoing, and terms are not final.
Why this matters in the AI chip race
A Microsoft win would cement Broadcom as the leading alternative source of custom AI accelerators, often called XPUs. Nvidia still dominates off the shelf GPUs, but hyperscalers want choice and tailored designs. Broadcom’s pitch is speed to silicon, tight co-design, and vertical control from chip to optics.
Broadcom has real momentum. In September, it booked about 10 billion dollars from a fourth hyperscale AI customer. In its last reported quarter, AI revenue reached roughly 5.2 billion dollars. Street models peg the next quarter near 6.2 billion dollars. That is not hobby scale. That is a production ramp.
Marvell, for its part, remains a capable player in custom silicon and cloud optics. A Microsoft shift would hurt, but it would not end Marvell’s AI plans. It would, however, reset the pecking order in custom accelerators. It would also signal that more of the biggest buyers want multi supplier strategies that dilute Nvidia’s grip in certain AI workloads.
For users, more suppliers means better availability and lower queues. It can mean lower costs and faster iteration. It can also mean more diverse chips to match different models and tasks.
What to watch into Broadcom earnings
Broadcom reports fiscal Q4 on Thursday, December 11. Expect the company to lean into AI, and to give a clearer read on 2026 supply.
- AI revenue run rate and new design wins
- Guidance for total revenue and free cash flow
- Infrastructure software updates, including VMware integration
- Capital spend, packaging capacity, and supply commitments
Watch for any color on hyperscaler contracts, delivery schedules, and packaging buildouts. Those details will drive 2026 confidence.
The risks under the rally
Valuation is rich. Expectations are high. Any slip on guidance could spark a pullback. Geopolitics also loom over advanced chips, packaging, and cross border supply chains. Controls can change with short notice. Lead times for substrates, HBM memory, and advanced packaging stay tight.
Nvidia will not stand still. It is pushing custom features, new interconnects, and tighter software stacks. If Broadcom wins Microsoft, Nvidia can still win other workloads in the same data centers. This is not a zero sum game, but share shifts matter.
Today’s Microsoft move is not signed. If talks stall, momentum can reverse just as fast.
Frequently Asked Questions
Q: What moved Broadcom stock today?
A: My reporting that Microsoft is exploring a custom AI chip shift to Broadcom. Shares traded near 404.56, up about 3.7 percent.
Q: Is the Microsoft deal done?
A: No. The talks are active. The direction is clear, but the agreement is not final.
Q: How does this affect Nvidia?
A: It pressures Nvidia at the edges, in custom accelerators. Nvidia still leads in general GPUs, but hyperscalers want more custom options.
Q: What should I watch on December 11?
A: AI revenue mix, 2026 supply plans, and VMware software traction. Any new hyperscaler details would be a key tell.
Q: What are the biggest risks?
A: Rich valuation, supply constraints, and shifting trade rules. A delay in AI ramps or a weak guide could hit the stock.
Broadcom just tightened its grip on the custom AI lane. If Microsoft signs, the balance of power in data center silicon shifts again. Earnings this week will tell us how fast Broadcom can scale to meet it.
