Thesis Labv0.2.0

NASDAQ · Information Technology · Semiconductors

AVGO Broadcom Inc.

As of 2026-05-20 · Palo Alto, CA

HOLD 6.0% target Conviction: high 12-24 months

Custom-silicon hedge on NVDA. Google TPU franchise alone justifies the position; VMware is the unknown.

Entry: Add 1-2% on any pullback to $390. VMware integration noise is the entry-creating force.

Reverse on: AVGO-F1, AVGO-F2

Business summary

Broadcom is two businesses in one. The semiconductor side designs custom ASICs (most importantly for Google's TPU program and emerging Meta MTIA work), Ethernet switching silicon (Tomahawk, Jericho — the backbone of AI training clusters), and a portfolio of legacy networking, storage, and wireless chips. The software side, expanded enormously by the 2023 VMware acquisition, sells virtualization, security, and mainframe software at industrial-grade pricing.

The AI investment case is the ASIC franchise. Google's TPU v5/v6/v7 are designed by Broadcom; the relationship is multi-year and produces materially better margins than merchant GPUs. Meta MTIA work is reportedly migrating to Broadcom as well. As hyperscalers diversify away from sole-sourcing NVIDIA, Broadcom captures the dispersion. VMware integration is largely complete; investors are watching for re-acceleration of free cash flow and how aggressively Broadcom can reprice the VMware book.

Connected theses

Key metrics

AI revenue run-rate (FY26 est)$25-30B
Driven by Google TPU + AI networking silicon. Run-rate growth >50% YoY.
Consensus
VMware integrationCost synergies largely realized; revenue stabilizing
Aggressive subscription conversion; some customer churn but ARR expanding.
FY25-26

Valuation snapshot

Price$420.71
Market cap$1,980B
Forward P/E33.0×
EV / EBITDA22.0×
FCF yield2.8%

Higher multiple than the legacy semiconductor business deserves, lower than pure-play AI. The blended position reflects market's uncertainty about how much VMware is structural vs accretion. AI franchise alone justifies a premium.

Evidence

Catalysts

  • FY Q3 earnings high
    What to watch: AI revenue trajectory, VMware ARR, FY27 capex guidance from hyperscaler customers

Falsifiers

  • Google or Meta cancels/delays custom silicon program
    armed · Hyperscaler earnings + AI event announcements
  • VMware integration drag worse than expected (subscription conversion stalls, churn >15%)
    armed · Software segment disclosures
  • AI networking share lost to Marvell or in-house hyperscaler designs
    armed · Earnings commentary + AI cluster vendor disclosures

Agent notes

Hold. The Google TPU franchise alone justifies the position. VMware is the unknowable. Sized to express the custom-silicon angle without taking VMware-specific risk in size.

Educational notes

📚 ASIC design win

An ASIC (Application-Specific Integrated Circuit) is custom silicon. When a hyperscaler picks a design partner like Broadcom for their custom chip, that's a 'design win' worth typically $1-3B over the program lifecycle, with margins that can exceed merchant GPU economics. Design wins are sticky — the same hyperscaler typically keeps the same designer for 2-3 generations because the IP and verification work compound. That's the moat: not just being best today, but being the designer of record across generations.

Open questions

  • What's the specific Meta MTIA timing?
  • VMware ARR vs license churn ratio?
  • Marvell competitive pressure on AI networking?