Same Playbook, Different Silicon: Broadcom Holds the Throne, Marvell Plays the AMD Challenger.
Broadcom and Marvell both crushed AI estimates and raised guidance, but they are winning in very different ways, and only one of those strategies scales fast enough to matter when hyperscaler spending shifts.
Broadcom (NASDAQ:AVGO | AVGO Price Prediction) and Marvell Technology (NASDAQ:MRVL) both beat estimates on AI demand and raised their guidance. They run the same playbook: custom accelerators for hyperscalers, plus the networking that links them. Broadcom operates at far larger scale. In custom silicon, Broadcom leads much like NVIDIA (NASDAQ:NVDA) does in GPUs. Marvell plays the challenger role that AMD (NASDAQ:AMD) plays against NVIDIA.
XPUs Power Broadcom While Optics Power Marvell
Broadcom’s AI semiconductor revenue hit $16.70B, up 221% YoY, and now makes up 56% of total revenue. Custom XPUs supplied 73% of AI revenue. Broadcom is now mainly an accelerator company, with VMware software attached. CEO Hock Tan described his pitch simply: “When you co-develop a chip that is optimized for your particular large language model workloads, you will outperform any GPU.”
Marvell’s revenue came in at $2.74B, up 36.5% YoY. Data center sales rose 46% to reach 79% of total revenue, so the business keeps focusing toward AI. Networking drove much of that growth. The 1.6T optics business is ramping, and management expects scale-out switching sales to more than double in fiscal 2027 as 51.2T products roll out.
| Business Driver | Broadcom | Marvell |
|---|---|---|
| Main Growth Engine | Custom XPUs (TPUs, Jalapeno, MTIA) | Optical DSPs, 51.2T switches, XPU-attach |
| Networking Edge | Tomahawk 6 Ethernet switches | 800G and 1.6T optics, NPO |
| Customer Model | Six large XPU customers | Broad hyperscaler design wins |
Broadcom Owns the Chip. Marvell Wins the Parts Around It.
Broadcom has signed long-term TPU supply deals with Alphabet (NASDAQ:GOOGL) unit Google, and it expects Anthropic to become its largest XPU customer in 2027. Scale squeezed margins: gross margin slipped to 75% as memory-heavy XPUs grew as a share of sales. Tan asked investors to focus on operating margin, which reached 67.9%. Operating margin held up even as gross margin slipped.
Marvell is working inside the same Google ecosystem, but with chips that attach to the TPU: NICs, storage controllers and memory controllers. Its Google deal includes a warrant that lets Google acquire up to 7% of Marvell shares if revenue milestones are met. The arrangement is a strong endorsement, and it also means existing shareholders could be diluted. Management said that “every hop in the network at some point was going to get some level of customization.”
Supply Chains Will Decide Who Pulls Ahead
Broadcom guided to Q4 revenue of ~$34.8B and AI revenue of ~$21.7B. It also listed power, HBM and data center shells as possible bottlenecks. Tan was careful on deployment timing: “We judge conservatively that somewhat less will be deployed.” Marvell guided to $3.15B and plans about $1 billion in supplier prepayments to lock down capacity. I will be watching Broadcom’s report on December 9, 2026 to see whether its supply grows as fast as demand.
What Investors Should Watch as Broadcom and Marvell Diverge
In stock performance, Marvell is up 163.4% year to date, while Broadcom is down 6.09%. Over the past month, both fell sharply: 11.33% for Broadcom and 14.8% for Marvell.
Broadcom’s multi-generation roadmaps and Tomahawk switches anchor its AI business. I would want to see Marvell’s 1.6T and custom ramps show up in reported revenue first. Investors can keep an eye on Marvell’s optics ramp and Broadcom’s custom XPU roadmap in the next earnings reports. The traits that mark an early AI-silicon winner are the same ones we cataloged in a free playbook on spotting the next NVIDIA.
Contact [email protected] for any questions or corrections.








