Broadcom vs Samsung

Custom AI Accelerators: Broadcom's Core Moat

Broadcom has established a commanding 70-80% market share in custom AI accelerators, making it the go-to partner for hyperscalers who want alternatives to NVIDIA GPUs. The company co-develops XPUs with customers like Google (whose TPUs are Broadcom-designed), Meta (MTIA v2), and most recently OpenAI, which announced a collaboration to deploy 10 gigawatts of OpenAI-designed accelerators built in partnership with Broadcom. CEO Hock Tan has stated the company has line of sight to $100 billion in AI chip revenue by 2027.

Samsung has no comparable custom silicon design business for external customers. While Samsung manufactures chips in its foundry, it does not offer the kind of deep co-design partnership that Broadcom provides. For organizations building bespoke AI training and inference silicon, Broadcom is the clear leader — and this gap is unlikely to close given the years of co-development relationships Broadcom has built with hyperscalers.

Memory and HBM: Samsung's Strategic Stronghold

Samsung is one of only three companies globally — alongside SK Hynix and Micron — that can manufacture High Bandwidth Memory, the critical component that enables AI accelerators to process massive datasets. In early 2026, Samsung shipped the industry's first commercial HBM4, delivering 11.7 Gbps per pin (46% above the industry standard) with 40% better power efficiency than HBM3E.

Samsung has secured an exclusive deal to supply HBM4 to OpenAI for its first-generation Titan chip, and has allocated over half of its Pyeongtaek foundry capacity to HBM4 base die production. With HBM4E sampling expected in H2 2026, Samsung is positioning itself at the bleeding edge of AI memory technology. Broadcom, as a fabless designer, has no memory manufacturing capability — it relies on companies like Samsung to supply the HBM that goes into the very accelerators Broadcom helps design.

Foundry Services: Samsung's Uphill Battle Against TSMC

Samsung Foundry began mass production of its 2nm GAA (Gate-All-Around) process in November 2025, beating TSMC to market on this transistor architecture. However, Samsung's foundry business holds only about 7.3% global revenue share compared to TSMC's dominant 70%. Yield rates remain a challenge: Samsung's 2nm process reportedly achieves 40-55% yields versus TSMC's 65% at comparable nodes.

Broadcom is a key customer of TSMC rather than Samsung Foundry, and the two have collaborated on advanced 3.5D packaging for AI chips. Samsung's foundry ambitions are strategically important to the broader semiconductor ecosystem because the industry needs manufacturing diversity beyond TSMC, but Broadcom's chip designs currently flow primarily through TSMC's fabs. Samsung is projecting 130% growth in 2nm orders for 2026, signaling aggressive efforts to win more design wins.

Datacenter Networking: Broadcom's Other AI Advantage

Beyond custom accelerators, Broadcom dominates the datacenter networking silicon market with its Tomahawk switching and Jericho routing chip families. The latest Tomahawk 6 and Jericho 4 were purpose-built for AI cluster fabrics and have seen record demand as hyperscalers build out massive GPU and XPU clusters. This dual position in both compute and networking silicon gives Broadcom unique end-to-end influence over AI datacenter architecture.

Samsung has minimal presence in datacenter networking silicon. While Samsung manufactures networking equipment through its enterprise division, it does not design the high-performance switching ASICs that form the backbone of AI datacenter fabrics. For infrastructure architects building AI infrastructure, Broadcom's networking chips are as critical as the accelerators themselves.

Enterprise Software vs. Consumer Ecosystem

Broadcom's $69 billion acquisition of VMware in 2023 gave it a massive enterprise software business, with infrastructure software revenue reaching $27 billion in FY2025. VMware Cloud Foundation is the virtualization layer running in most enterprise datacenters, and Broadcom sees AI workloads as a growth driver for VMware adoption. This software moat gives Broadcom recurring revenue and deep enterprise relationships that Samsung lacks.

Samsung's consumer ecosystem — spanning Galaxy smartphones, tablets, TVs, and home appliances — represents a different kind of platform advantage. Samsung is integrating AI capabilities across its device portfolio through Galaxy AI, and its massive installed base of consumer devices could become a significant edge computing platform. These are fundamentally different business models: Broadcom sells infrastructure to enterprises, while Samsung sells hardware and experiences to both consumers and enterprises.

Supply Chain Interdependence

Perhaps the most important insight in this comparison is that Broadcom and Samsung are not just competitors — they are supply chain partners. Broadcom's custom AI accelerators require HBM manufactured by Samsung (among others). Samsung's foundry competes for Broadcom's chip fabrication contracts. And both companies supply components that end up in the same AI servers and datacenter racks built by their shared customers.

This interdependence means evaluating Broadcom and Samsung is less about picking a winner and more about understanding the different layers of the AI hardware stack. Broadcom dominates at the design and architecture layer, while Samsung dominates at the physical manufacturing layer. Both are essential, and both capture substantial value from the AI infrastructure buildout.