Technical Intelligence Analyst Report Date: 2026-03-05

Executive Summary

  • Broadcom’s Rise as a Merchant Silicon Alternative: Broadcom’s Q1 F2026 results highlight a massive shift toward custom AI XPUs (ASICs) among hyperscalers, directly threatening the merchant GPU market share of NVIDIA and AMD.
  • Major Custom Silicon Roadmaps Revealed: New details emerged regarding Google’s TPU v7 (“Ironwood”), Meta’s MTIA v3, and OpenAI’s “Titan” chip, with deployment scales measured in gigawatts for 2026-2027.
  • Supply Chain Lock-in: Broadcom has secured guaranteed capacity for wafers and HBM through 2028 for its six major AI partners, potentially squeezing merchant GPU vendors for manufacturing resources.

🤼‍♂️ Market & Competitors

[2026-03-05] Broadcom May Become The Biggest Counterbalance To Nvidia

Source: The Next Platform

Key takeaway relevant to AMD:

  • Direct Threat to Instinct TAM: Broadcom’s success in enabling “homegrown” XPUs for Google, Meta, and OpenAI removes these massive volume buyers from the merchant GPU market (AMD Instinct/Nvidia H/B-series) for internal workloads.
  • Supply Chain Competition: Broadcom has secured specific wafer and HBM capacity for 2026–2028 for its partners. This indicates that AMD will face continued stiff competition for TSMC CoWoS and HBM supply, not just from Nvidia, but from the hyperscalers themselves via Broadcom.
  • System Integration Rivalry: The article notes Broadcom’s expansion into rackscale systems may make AMD “rue the day” it spun off ZT Systems, as Broadcom creates a vertical “one throat to choke” model for AI infrastructure that bypasses OEMs.

Summary:

  • Broadcom reported Q1 F2026 financials showing a 54.2% jump in semiconductor solutions, driven almost entirely by custom AI accelerators and networking.
  • CEO Hock Tan provided specific roadmap updates for six major custom silicon customers (including Google, Meta, and OpenAI).
  • The company forecasts AI revenues to exceed $100 billion by Fiscal 2027.

Details:

  • Financial Metrics (Q1 F2026):
    • Total Revenue: $19.31 billion (+29.5% YoY).
    • Semiconductor Solutions: $12.52 billion (+54.2% YoY).
    • AI Chip/System Revenue (Estimated): ~$8.44 billion (>2X YoY).
    • Operating Margin (Semiconductors): 60%.
  • Customer Specific Technical Roadmaps:
    • Google: High demand projected for TPU v7 (codenamed “Ironwood”) in 2026/2027.
      • Anthropic: purchasing $10B in TPU racks; installing 1 gigawatt of TPU v7 capacity in 2026, scaling to >3 gigawatts in 2027.
    • Meta:
      • MTIA v2: Currently shipping (“alive and well”).
      • MTIA v3: Next-gen chip will scale production to “multiple gigawatts” of compute capacity.
    • OpenAI (Customer 6): Confirmed use of Broadcom for the “Titan” AI XPU. OpenAI plans to install >1 gigawatt of Titan capacity in Fiscal 2027.
    • ByteDance & Apple (Rumored): Identified as Customers 4 and 5; shipments expected to double in Fiscal 2027.
  • Networking vs. Compute Dynamics:
    • AI networking revenue was $2.78 billion (+60% YoY) but down 51.5% sequentially.
    • Analysis suggests hyperscalers pre-bought networking in Q4 F2025 to prepare for the rollout of new compute engines (TPU v7, Titan, etc.) throughout 2026.
  • Supply Chain Security:
    • Broadcom has fully secured capacity for leading-edge wafers, HBM (High Bandwidth Memory), and substrates for 2026 through 2028 to support these custom ASIC roadmaps.