When investors think about AI infrastructure, the first name that comes to mind is almost always Nvidia. The GPU giant has become synonymous with the AI boom — and deservedly so, with Blackwell on track for $100B in data center revenue. But every one of those GPUs needs to talk to every other GPU. The company that makes that possible — providing the high-speed networking fabric that stitches tens of thousands of accelerators into a single compute cluster — is Broadcom.
Broadcom has quietly become one of the most consequential companies in AI. Through its Tomahawk switching silicon, custom ASIC design capabilities, and a transformed VMware business unlocking high-margin subscription revenue, AVGO is generating the kind of compounding growth that rewards patient investors. The dividend just grew 18% to $0.65/quarter, marking Broadcom's 14th consecutive annual increase.
Tomahawk 5: The Networking Engine Powering AI Clusters
Broadcom's Tomahawk 5 is the industry's highest-bandwidth Ethernet switch, delivering 51.2 Tbps of switching capacity in a single device. It is deployed in every major hyperscaler data center — Google, Meta, Microsoft, and Amazon all rely on Tomahawk-based switches to connect their GPU clusters. In an era where cluster sizes have grown from 4,000 to 100,000+ GPUs, the networking layer has become just as critical as compute.
The transition from InfiniBand to Ethernet is accelerating. While Nvidia's NVLink and InfiniBand have traditionally dominated AI training, Ultra Ethernet — a consortium backed by Broadcom, Meta, Microsoft, and others — is gaining traction as the standard for scale-out AI workloads. Broadcom sits at the center of this shift, with Tomahawk 5 already powering the largest GPU clusters in the world.
The Signal's Take: Every $1 spent on AI GPUs generates roughly $0.10–0.15 in networking spend. With AI GPU spending exceeding $200B in 2026, Broadcom's networking TAM is approaching $30B — and Tomahawk captures the lion's share.
Custom ASICs: The Hyperscaler Pivot
Beyond networking, Broadcom's custom ASIC (application-specific integrated circuit) business has become a multi-billion-dollar growth engine. The company designs custom AI accelerators for Google (TPU), Meta, and Apple, among others. These are not small programs — Google's TPU v6 alone is an annual $5B+ program that Broadcom co-develops and manufactures. Unlike merchant silicon, these are multi-year, high-margin design wins with deep customer lock-in.
Analysts estimate Broadcom's custom ASIC revenue reached $10–12B in fiscal 2026, growing 40%+ year over year. The pipeline is even stronger: Apple is designing its own AI server chips with Broadcom, and multiple tier-2 hyperscalers are lining up for custom designs rather than buying merchant GPUs from Nvidia or AMD.
- Google TPU: Broadcom-designed — now on v6, scaling to $8B+ annual spend
- Meta: Custom AI accelerators for inference workloads
- Apple: Server-grade AI chips in development for cloud inference
- OpenAI/Microsoft: Reportedly exploring custom silicon with Broadcom
VMware Transformation and the Dividend Machine
Broadcom's $69B acquisition of VMware in 2023 was initially met with skepticism. The integration has been aggressive — Broadcom cut thousands of VMware employees, ended perpetual licensing, and pushed customers to subscription models. The result: VMware revenue has stabilized after a transitional dip, subscription backlog is growing, and margins are expanding.
VMware now generates over $15B in annualized revenue with 85%+ gross margins. The transition to subscription has created a powerful recurring revenue base that Broadcom is using to fund its dividend growth. With a $5.67 quarterly dividend (yield ~1.6%), 14 years of consecutive increases, and a payout ratio still below 50%, AVGO is one of the few AI plays that also qualifies as a reliable dividend growth stock.
The sum of the parts is compelling. Tomahawk connects the GPUs, custom ASICs design the accelerators, and VMware monetizes the enterprise AI stack. With revenue tracking toward $60B in fiscal 2026 and operating margins above 45%, Broadcom is executing on multiple fronts simultaneously. For investors looking for AI exposure that doesn't rely solely on GPU spot pricing, AVGO offers diversification, recurring revenue, and compounding dividends — a rare combination in semiconductor investing.





Discussion
Comments powered by Giscus — sign in with GitHub to join the discussion.