Services · the new software · Research Note №1 · Memo 008 of 185AVGO · ← Overview
Semiconductors & Software
AVGO
Broadcom
Second-order NVDA — diversified enabler exposure.
Highly PositiveRank 8 · Nasdaq-100 constituent
Last price
$406.54
Market cap
$1.93T
As of
18 April 2026
Live quote sourced from Yahoo Finance. Prices cited in narrative below reflect the original memo date and may be stale.
Scores · adapted framework
Enabler
10 / 10
Autopilot adoption
2 / 10
Disruption risk
2 / 10
Efficiency upside
5 / 10
The Sequoia matrix
Intelligence / Judgment
Not applicableSemiconductor and networking vendor; doesn't sell services work.
Copilot posture
NoneNo direct copilot product line.
Autopilot posture
NoneNo direct autopilot product line.
Data moat
ModerateNot a data-moat story — IP plus customer relationships plus the VMware install base provide the moat.
Execution layer
Not applicableHardware vendor; execution happens at the customer level.
The memo
State of play · AVGO
Trading ~$407 in mid-April 2026, down ~22% from the $217 peak. Market cap ~$800B. FY26 Q1 (ended March) revenue $8.9B (+18% YoY), with Custom Silicon (hyperscaler ASIC) segment up 35%+ YoY. Broadcom supplies custom-silicon components for Google TPU, Amazon Trainium, Meta MTIA. VMware acquisition (Jan 2024) now contributes ~$25B run-rate revenue. Next earnings: FY26 Q2 end of May 2026.
Thesis angle
The silicon supplier for hyperscaler custom AI chips (including Google TPU) plus networking infrastructure that every AI data centre requires. Benefits whether autopilot providers use merchant or custom compute.
The framing
Broadcom is the picks-and-shovels vendor for the autopilot economy, with an inversion of NVIDIA's risk profile. NVIDIA fears custom silicon; Broadcom sells the components that custom silicon is built on. Meta-Broadcom co-development deal for MTIA through 2030 is a major tailwind that NVIDIA lists as a headwind. Broadcom also benefits from networking-infrastructure scaling as every data centre builds for AI inference.
Two forces, opposite directions
Tailwind · custom silicon boom is a Broadcom enabler
Meta-Broadcom MTIA v3+ co-development deal extends through 2030; pre-commits Meta infrastructure growth to Broadcom
Google TPU v7 uses Broadcom components; as TPU deployment scales with Agentspace, Broadcom captures silicon content
Amazon Trainium and custom networking (Tomahawk, Jericho) scale with AWS compute growth; Broadcom is the supplier to both
Custom silicon is the fastest-growing AI segment (35%+ growth); Broadcom is the default design partner across hyperscalers
The Meta-Broadcom deal is a permanent structural advantage that shifts the custom-silicon economics away from NVIDIA and toward pure enablers.
Headwind · customer concentration and cyclicality
Top 4 hyperscalers (Meta, Google, Amazon, Microsoft) drive ~60%+ of Broadcom custom-silicon revenue; customer concentration is high
If any hyperscaler cuts capex (recession, capex discipline cycle), Broadcom takes the first hit
Networking segment (Tomahawk, Jericho) is cyclical and tied to data-centre refresh rates
Valuation has expanded sharply (now ~45x P/E); any growth slowdown triggers multiple compression
Broadcom is a quality supplier in a cyclical position. Customer diversification is limited.
Broadcom exposure across the AI infrastructure stack
Customer
Product
Revenue contribution
Tailwind / headwind
Co-dev status
Meta
MTIA custom silicon components
~$2B+ (est.)
Tailwind
MTIA v3+, 2030 lock-in
Google
TPU v7 components
~$1.5B (est.)
Tailwind
Ongoing
Amazon
Trainium + Networking
~$1B+ (est.)
Tailwind
Ongoing
Other hyperscalers
Custom silicon + networking
~$2B+ (est.)
Headwind
Competing with NVIDIA
Broadcom is the vendor to every hyperscaler's custom-silicon program. The Meta lock-in is structural; others are project-by-project. Concentration risk is high but long-term visibility is strong.
Bull case
The Meta-Broadcom MTIA deal is a permanent structural advantage.
Meta is locked in through 2030; this is a multi-year pre-commitment worth billions. Unlike NVIDIA (which fears custom silicon), Broadcom profits from it.
Custom silicon is the fastest-growing AI segment.
Growth is 35%+ YoY; merchant GPU is 15–20%. Broadcom captures the silicon content of every custom ASIC deployed by every hyperscaler.
Networking infrastructure scales with every data-centre buildout.
Tomahawk and Jericho grow with AI cluster density. This segment is less concentrated than custom silicon and more predictable.
VMware acquisition provides high-margin software cashflow and diversification.
~25B run-rate revenue at 70%+ gross margins. Buffers the cyclicality of semiconductor segments.
Bear case
Customer concentration is extreme.
Top 4 hyperscalers are 60%+ of custom-silicon revenue. If any cuts capex, Broadcom takes the hit immediately.
Valuation has expanded sharply while growth is cyclical.
~45x forward P/E is richly valued for a semiconductor supplier. Any deceleration in hyperscaler capex triggers multiple compression.
Broadcom is a component supplier, not a system owner.
Hyperscalers can design Broadcom out if margins tighten. Moat is weaker than platform vendors (MSFT, GOOGL, AMZN).
Networking segment is tied to cyclical data-centre refresh rates.
Not as resilient as custom silicon. Margin pressure from competition is a structural headwind.
Sequoia-framework fit
Broadcom is NVIDIA's structural hedge in the custom-silicon economy. While NVIDIA loses inference share to hyperscaler ASICs, Broadcom supplies the silicon components that ASICs are built on. The Meta-Broadcom MTIA co-development deal through 2030 is a permanent structural advantage. Broadcom is a "highly positive" on the thesis because it profits whether the market uses merchant GPU or custom silicon — it is architecture-agnostic. The risk is customer concentration and cyclicality; the moat is weaker than platform vendors. Size accordingly as a pick-and-shovels hedge, not a core thesis holding.
Investor takeaway
Second-order NVDA play for diversified enabler exposure.