Broadcom makes the networking semiconductors that route data inside every data center and enterprise network, plus the Wi-Fi chips in every iPhone. Its VMware acquisition ($69B) made it a major enterprise software company overnight. Broadcoms custom AI chip (XPU) business is booming - Google, Meta, and ByteDance all use Broadcom-designed AI chips to reduce NVIDIA dependence.
Key relationships in depth
Broadcom & Apple
Broadcom is one of Apple's most important chip suppliers, providing Wi-Fi, Bluetooth, and custom RF components embedded in every iPhone, iPad, Mac, and Apple Watch. Apple and Broadcom signed a multi-year supply agreement in 2023 worth $15B+ for wireless components, reflecting Apple's inability to in-source RF semiconductor manufacturing at competitive cost and performance. Broadcom also supplies touchscreen controllers and storage controllers used across Apple's device lineup.
Why it matters: Apple's ecosystem depends on best-in-class wireless performance for Wi-Fi 6E and upcoming Wi-Fi 7 connectivity, and Broadcom's Wi-Fi chip technology leads the industry in throughput and power efficiency for mobile devices. Despite Apple's extensive in-house chip program (A-series, M-series, Apple Neural Engine), wireless semiconductor design remains a specialized area where Broadcom's scale across billions of devices gives it performance and cost advantages Apple cannot replicate.
Microsoft and Broadcom have a growing relationship centered on AI network infrastructure and Azure's evolving silicon strategy. Broadcom supplies Ethernet switching ASICs (Tomahawk and Jericho series) for Azure's data-center networking backbone, interconnecting its tens of thousands of AI GPU servers. Additionally, Microsoft has reportedly engaged Broadcom as a potential partner for custom AI ASIC development, as part of Microsoft's Azure Maia program to design custom AI chips that reduce its NVIDIA GPU dependency.
Why it matters: Azure's AI infrastructure growth requires both networking silicon to connect GPU clusters at high bandwidth and custom AI inference chips to lower the per-token cost of serving Microsoft Copilot and Azure AI workloads. Broadcom's expertise in both networking ASICs and custom AI chip design for hyperscalers (demonstrated through Google's TPU program) makes it a natural partner for Microsoft as it scales its Azure Maia AI silicon program.
Broadcom and NVIDIA represent the two dominant forces in AI accelerator semiconductors, with NVIDIA controlling 86-92% of general-purpose AI GPU market share and Broadcom commanding 70%+ of the custom ASIC market for hyperscalers. While they operate in adjacent segments, their markets are converging as hyperscalers (Google, Meta, Amazon, Microsoft) shift portions of AI workloads from NVIDIA GPUs to Broadcom-designed custom chips. Broadcom's AI revenue target of $100B+ by 2027 is directly funded by hyperscaler dollars that would otherwise flow to NVIDIA, creating fundamental competitive dynamics.
Why it matters: NVIDIA's general-purpose GPU architecture provides the broadest compatibility and developer ecosystem (CUDA, 4M+ developers) but carries higher cost-per-FLOP for specific inference workloads than purpose-built ASICs. Broadcom's custom TPUs and XPUs deliver better performance-per-watt for specific model architectures, enabling hyperscalers to reduce AI infrastructure costs at scale. The competitive dynamic is additive for hyperscalers (they buy both) but structurally limits NVIDIA's TAM expansion as custom silicon captures an increasing share of new AI compute investment.
Broadcom and Google have a landmark 5-year custom AI chip supply agreement through 2031, with Broadcom manufacturing Google's custom Tensor Processing Units (TPUs) and AI networking ASICs. Google's TPU v5 and v6 series are designed in-house and manufactured by Broadcom, representing one of the largest custom silicon relationships in technology history. In April 2026, Broadcom received $10B+ orders from Google (through Alphabet's Anthropic investment relationship) for custom TPU compute, with demand projected to exceed $21B in 2026 alone. Broadcom commands 70%+ of the custom AI accelerator (ASIC) market, with Google as its anchor customer.
Why it matters: Google's decision to develop custom TPUs rather than exclusively buying NVIDIA GPUs reflects the hyperscaler's strategy to optimize AI compute for specific workloads (transformer inference, recommendation systems) at lower cost per FLOP than general-purpose GPUs. Broadcom's custom ASIC capabilities allow Google to achieve better performance-per-watt for its specific AI architectures. The 5-year supply agreement through 2031 provides Broadcom with revenue visibility that supports massive capital investment in advanced packaging and TSMC leading-node production allocation.
Relationships are compiled by VexMap from company filings, announcements and reporting, and last reviewed September 2026. Spotted something wrong? Tell us. Not investment advice.