Broadcom Inc.
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
Dual-moat machine — irreplaceable custom silicon for AI hyperscalers and VMware's stranglehold on enterprise virtualization infrastructure. Apple renewed its supply agreement through 2031 (>$30B, spanning FBAR wireless filters and custom AI silicon, July 2026), joining Meta (through 2029) and Google (through 2031) as long-dated contracted customers. Moat statuses unchanged in the July 2026 review: the Q2 sell-off was a multiple event, not moat damage — though the CISPE EU escalation (five trade groups urging interim measures, July 15) now poses a live regulatory threat to VMware pricing power.
Broadcom's moat is built on Proprietary Silicon and Enterprise Infrastructure Lock-In:
- Custom AI Accelerator Dominance: Broadcom designs custom AI XPUs for six confirmed hyperscaler customers. Q2 FY2026 AI revenue hit $10.8B (+143% YoY), with Q3 guided to $16.0B (+200%+ YoY) and full-year FY2026 guidance raised to $56B (~+180% YoY); management reaffirmed $100B+ for FY2027. Meta is extended through 2029 (multi-GW MTIA rack-level supply); Google is formalized through 2031 (future TPU generations, AI networking, rack-scale components); Apple renewed through 2031 (>$30B covering FBAR filters and custom AI silicon, July 2026). Anthropic is tripling compute to 3GW+ before year-end 2026. The OpenAI 10GW inference program remains stalled on an $18B financing snag (Broadcom requiring a Microsoft 40% off-take that Microsoft has not agreed to), with first Jalapeno chips now expected mostly in 2027.
- Ethernet Networking Monopoly for AI: Broadcom's Tomahawk and Jericho switch chip families dominate the hyperscale data center networking market. As AI clusters require ever-faster intra-cluster networking at lower cost than InfiniBand, Broadcom's Ultra Ethernet ecosystem is positioned to capture the AI networking buildout. Broadcom is now selling fully assembled 'Ironwood Racks' (Google's 7th-gen TPU) directly to AI firms including Anthropic.
- VMware: The Enterprise Infrastructure Tax: VMware's vSphere and vCenter run the virtualization layer for ~70% of the Fortune 500. VMware Cloud Foundation (VCF) is described by management as the 'permanent abstraction layer' between AI software and physical silicon that 'cannot be disintermediated or replaced.' Q2 FY2026 infrastructure software revenue was $7.18B (+9% YoY) with ARR growth of 17% YoY (a modest deceleration from Q1's 19%) at 93% software gross margins. The regulatory front is heating up: after Broadcom terminated most of its European VMware Cloud Service Provider partnerships (January 2026), CISPE filed an EU antitrust complaint (March 2026), and on July 15 2026 five cloud trade groups jointly urged the EU Commission to impose interim measures suspending Broadcom's partner-program terminations.
- Hock Tan's Capital Allocation Engine: CEO Hock Tan has executed the most disciplined semiconductor M&A strategy of the past decade — acquiring CA Technologies, Symantec enterprise security, and VMware, then aggressively cutting costs, raising prices, and migrating customers to subscription contracts. In Q1 FY2026, Broadcom returned $10.9B to shareholders ($3.1B dividends + $7.8B buybacks). The board authorized a new $10B buyback through 2026.
Moat Verdict
Broadcom is a net AI beneficiary with two compounding engines: custom XPU co-development for six hyperscaler customers (AI revenue +143% YoY to $10.8B in Q2 FY2026; FY2026 guided to $56B, ~+180% YoY; FY2027 $100B+ reaffirmed) and VMware's stranglehold on enterprise virtualization (ARR +17% YoY at 93% gross margins). The strongest AI-resilient moats are systemOfRecord (vCenter as compute-inventory authority), businessLogic (custom silicon embedded in hyperscaler training pipelines), and talentScarcity (scarce custom ASIC design teams). Long-dated contracts — Meta through 2029, Google through 2031, and now Apple through 2031 (>$30B, July 2026) — meaningfully reduce medium-term customer concentration risk. July 2026 review: all ten moat statuses unchanged — the ~25% drawdown from the June 3 ATH ($495) was a multiple event (Q3 AI guide of $16B vs ~$17.2B whisper; $1.3T chip-sector sell-off June 5), not moat damage. Three risks require monitoring: (1) AI-capex digestion — the whisper-miss reaction proved expectations fragility, and the OpenAI $18B financing snag remains unresolved with Jalapeno chips slipping to 2027; (2) the CISPE EU complaint has escalated — five cloud trade groups jointly urged interim measures against Broadcom's VCSP terminations on July 15 2026; an adverse interim decision would directly weaken the bundling moat and warrant a status downgrade; (3) customer concentration among a handful of hyperscalers. Reassess after Q3 FY2026 results (early September 2026).
71.7 resilient · 74.5 vulnerable · 80/20 = 72.2 · = 72
Open a moat to read its note.
VMware vCenter and vSphere have been the interface for enterprise virtualization for 20 years. Infrastructure teams are deeply trained on VMware workflows — retraining is a genuine organizational cost.
VMware runs the compute layer that enterprise logic executes on, and custom ASICs embed Broadcom silicon in hyperscaler pipelines — but that is the embedding rated under transactionEmbedding and systemOfRecord, not vendor-owned business logic. Re-rated from strong to intact so one fact is not scored three times.
Not a data company: Broadcom's value is in silicon IP and infrastructure software, and it never held a public-data moat. A pillar that never applied is not applicable rather than weakened. Previously weakened.
Custom-ASIC design at hyperscale needs rare teams, but Marvell, MediaTek, Alchip and the hyperscalers' in-house groups compete for the same programmes. Deep bench, not an irreplaceable scarcity. Re-rated from strong to intact in the proof-point pass.
VMware Cloud Foundation bundles compute (vSphere), networking (NSX), storage (vSAN) and management into one licence, and Broadcom has moved the base onto it: VMware ARR grew 17% YoY at 93% gross margin, pricing power a point-product rival could not extract. Management calls VCF the permanent abstraction layer for AI workloads.
Broadcom does not build or monetise a proprietary dataset: VMware telemetry is customer-environment data it does not pool, and custom-silicon design data belongs to each hyperscaler customer. The pillar does not apply. Previously weakened, which scored a pillar that never applied.
VMware holds FIPS 140-2, FedRAMP, and Common Criteria certifications critical for government and regulated industry deployments. Re-certification on a new platform takes 18-36 months — a meaningful lock-in for public sector and financial services customers.
VMware's partner and ISV ecosystem was a real indirect network effect, but Broadcom cut the partner programme to a far smaller invite-only set and repriced VMware into VCF bundles, and a share of customers has been migrating workloads elsewhere — the ecosystem is shrinking by design. The semiconductor business has no network effect. Weakened on that erosion, not as a placeholder.
Custom accelerators are co-designed with each hyperscaler over multiple silicon generations, so Broadcom is embedded in their AI roadmaps: AI revenue reached $10.8B in Q2 FY2026 (+143% YoY) with FY2026 guided to $56B across 6 custom-XPU customers. Tomahawk and Jericho switching sits in the same data-centre fabric. Moving a programme mid-generation means re-taping silicon.
VMware vCenter is the system of record for enterprise compute inventory — every VM, workload and resource allocation lives there — across an installed base that entered the Broadcom deal at more than 300,000 customers. Migrating off it is a multi-year infrastructure programme, which is why repricing has held.
Fabless or equipment franchise whose edge is IP and process know-how, rated elsewhere; manufacturing volume does not give it a unit-cost lead rivals cannot buy.
Buyers are enterprises choosing on switching cost, integration and performance, which the other pillars rate. The name carries reputation, not a price premium it could hold on brand alone.
Combined average of Moat (AI Resilience), Growth, and Valuation scores.
Moat Score
Dual-moat machine — irreplaceable custom silicon for AI hyperscalers and VMware's stranglehold on enterprise virtualization infrastructure. Apple renewed its supply agreement through 2031 (>$30B, spanning FBAR wireless filters and custom AI silicon, July 2026), joining Meta (through 2029) and Google (through 2031) as long-dated contracted customers. Moat statuses unchanged in the July 2026 review: the Q2 sell-off was a multiple event, not moat damage — though the CISPE EU escalation (five trade groups urging interim measures, July 15) now poses a live regulatory threat to VMware pricing power.
Growth Score
Q3 FY2026 (reported September 2026) delivered $29.6B revenue (+86% YoY) with AI semiconductor revenue of $16.7B (+221% YoY, +54% QoQ), beating the $16.0B guide that had disappointed the street in June. Q4 is guided to $34.8B revenue (+93% YoY) with AI semiconductor revenue of $21.7B (+236% YoY), which puts FY2026 AI revenue near $58B; management now expects AI revenue to roughly double again to ~$115B in FY2027. Infrastructure software revenue was $8.8B (+29% YoY, from +9% in Q2), lifted by license revenue recognised upfront now that most new VCF contracts carry no termination-for-convenience right — a recognition change as much as a demand change. Non-AI semiconductors were $4.2B, flat on Q2. Apple renewed its supply agreement through 2031 (>$30B); Meta extended through 2029; Google through 2031; Anthropic tripling to 3GW+ by year-end; OpenAI's first-generation XPU is slated for volume deployment in 2027 at more than 1GW.
Valuation Score
At ~$370 (July 18, 2026), Broadcom trades ~24% below the $490 base case — the most attractive entry since April, courtesy of the post-earnings de-rating rather than any fundamental deterioration. The stock fell ~25% from its June 3 ATH ($495) after the Q3 AI guide of $16B missed the street's ~$17.2B whisper despite Q2 beating and FY2026 AI guidance being raised to $56B. The forward P/E has compressed to ~20× (non-GAAP NTM) — now below the semiconductor industry median (~34×) — and the PEG sits well under 0.5. The market is pricing meaningful AI-capex digestion risk that the order book ($56B FY2026 guided, $100B+ FY2027 reaffirmed, Apple contracted through 2031) does not yet show.
The Dual-Moat Architecture
Broadcom's moat is built on Proprietary Silicon and Enterprise Infrastructure Lock-In:
- Custom AI Accelerator Dominance: Broadcom designs custom AI XPUs for six confirmed hyperscaler customers. Q2 FY2026 AI revenue hit $10.8B (+143% YoY), with Q3 guided to $16.0B (+200%+ YoY) and full-year FY2026 guidance raised to $56B (~+180% YoY); management reaffirmed $100B+ for FY2027. Meta is extended through 2029 (multi-GW MTIA rack-level supply); Google is formalized through 2031 (future TPU generations, AI networking, rack-scale components); Apple renewed through 2031 (>$30B covering FBAR filters and custom AI silicon, July 2026). Anthropic is tripling compute to 3GW+ before year-end 2026. The OpenAI 10GW inference program remains stalled on an $18B financing snag (Broadcom requiring a Microsoft 40% off-take that Microsoft has not agreed to), with first Jalapeno chips now expected mostly in 2027.
- Ethernet Networking Monopoly for AI: Broadcom's Tomahawk and Jericho switch chip families dominate the hyperscale data center networking market. As AI clusters require ever-faster intra-cluster networking at lower cost than InfiniBand, Broadcom's Ultra Ethernet ecosystem is positioned to capture the AI networking buildout. Broadcom is now selling fully assembled 'Ironwood Racks' (Google's 7th-gen TPU) directly to AI firms including Anthropic.
- VMware: The Enterprise Infrastructure Tax: VMware's vSphere and vCenter run the virtualization layer for ~70% of the Fortune 500. VMware Cloud Foundation (VCF) is described by management as the 'permanent abstraction layer' between AI software and physical silicon that 'cannot be disintermediated or replaced.' Q2 FY2026 infrastructure software revenue was $7.18B (+9% YoY) with ARR growth of 17% YoY (a modest deceleration from Q1's 19%) at 93% software gross margins. The regulatory front is heating up: after Broadcom terminated most of its European VMware Cloud Service Provider partnerships (January 2026), CISPE filed an EU antitrust complaint (March 2026), and on July 15 2026 five cloud trade groups jointly urged the EU Commission to impose interim measures suspending Broadcom's partner-program terminations.
- Hock Tan's Capital Allocation Engine: CEO Hock Tan has executed the most disciplined semiconductor M&A strategy of the past decade — acquiring CA Technologies, Symantec enterprise security, and VMware, then aggressively cutting costs, raising prices, and migrating customers to subscription contracts. In Q1 FY2026, Broadcom returned $10.9B to shareholders ($3.1B dividends + $7.8B buybacks). The board authorized a new $10B buyback through 2026.
Moat Verdict
Broadcom is a net AI beneficiary with two compounding engines: custom XPU co-development for six hyperscaler customers (AI revenue +143% YoY to $10.8B in Q2 FY2026; FY2026 guided to $56B, ~+180% YoY; FY2027 $100B+ reaffirmed) and VMware's stranglehold on enterprise virtualization (ARR +17% YoY at 93% gross margins). The strongest AI-resilient moats are systemOfRecord (vCenter as compute-inventory authority), businessLogic (custom silicon embedded in hyperscaler training pipelines), and talentScarcity (scarce custom ASIC design teams). Long-dated contracts — Meta through 2029, Google through 2031, and now Apple through 2031 (>$30B, July 2026) — meaningfully reduce medium-term customer concentration risk. July 2026 review: all ten moat statuses unchanged — the ~25% drawdown from the June 3 ATH ($495) was a multiple event (Q3 AI guide of $16B vs ~$17.2B whisper; $1.3T chip-sector sell-off June 5), not moat damage. Three risks require monitoring: (1) AI-capex digestion — the whisper-miss reaction proved expectations fragility, and the OpenAI $18B financing snag remains unresolved with Jalapeno chips slipping to 2027; (2) the CISPE EU complaint has escalated — five cloud trade groups jointly urged interim measures against Broadcom's VCSP terminations on July 15 2026; an adverse interim decision would directly weaken the bundling moat and warrant a status downgrade; (3) customer concentration among a handful of hyperscalers. Reassess after Q3 FY2026 results (early September 2026).
71.7 resilient · 74.5 vulnerable · 80/20 = 72.2 · = 72
Open a moat to read its note.
VMware vCenter and vSphere have been the interface for enterprise virtualization for 20 years. Infrastructure teams are deeply trained on VMware workflows — retraining is a genuine organizational cost.
VMware runs the compute layer that enterprise logic executes on, and custom ASICs embed Broadcom silicon in hyperscaler pipelines — but that is the embedding rated under transactionEmbedding and systemOfRecord, not vendor-owned business logic. Re-rated from strong to intact so one fact is not scored three times.
Not a data company: Broadcom's value is in silicon IP and infrastructure software, and it never held a public-data moat. A pillar that never applied is not applicable rather than weakened. Previously weakened.
Custom-ASIC design at hyperscale needs rare teams, but Marvell, MediaTek, Alchip and the hyperscalers' in-house groups compete for the same programmes. Deep bench, not an irreplaceable scarcity. Re-rated from strong to intact in the proof-point pass.
VMware Cloud Foundation bundles compute (vSphere), networking (NSX), storage (vSAN) and management into one licence, and Broadcom has moved the base onto it: VMware ARR grew 17% YoY at 93% gross margin, pricing power a point-product rival could not extract. Management calls VCF the permanent abstraction layer for AI workloads.
Broadcom does not build or monetise a proprietary dataset: VMware telemetry is customer-environment data it does not pool, and custom-silicon design data belongs to each hyperscaler customer. The pillar does not apply. Previously weakened, which scored a pillar that never applied.
VMware holds FIPS 140-2, FedRAMP, and Common Criteria certifications critical for government and regulated industry deployments. Re-certification on a new platform takes 18-36 months — a meaningful lock-in for public sector and financial services customers.
VMware's partner and ISV ecosystem was a real indirect network effect, but Broadcom cut the partner programme to a far smaller invite-only set and repriced VMware into VCF bundles, and a share of customers has been migrating workloads elsewhere — the ecosystem is shrinking by design. The semiconductor business has no network effect. Weakened on that erosion, not as a placeholder.
Custom accelerators are co-designed with each hyperscaler over multiple silicon generations, so Broadcom is embedded in their AI roadmaps: AI revenue reached $10.8B in Q2 FY2026 (+143% YoY) with FY2026 guided to $56B across 6 custom-XPU customers. Tomahawk and Jericho switching sits in the same data-centre fabric. Moving a programme mid-generation means re-taping silicon.
VMware vCenter is the system of record for enterprise compute inventory — every VM, workload and resource allocation lives there — across an installed base that entered the Broadcom deal at more than 300,000 customers. Migrating off it is a multi-year infrastructure programme, which is why repricing has held.
Fabless or equipment franchise whose edge is IP and process know-how, rated elsewhere; manufacturing volume does not give it a unit-cost lead rivals cannot buy.
Buyers are enterprises choosing on switching cost, integration and performance, which the other pillars rate. The name carries reputation, not a price premium it could hold on brand alone.
Growth Analysis
Growth Drivers
Key Risk
Hyperscaler AI-capex digestion and customer concentration: the FY2027 ~$115B AI revenue path rests on a handful of XPU programs (Google, Meta, Anthropic, OpenAI and a few others), so if two or more large customers cut FY2027 capex or push XPU deliveries out by 2+ quarters, the doubling slips and the multiple de-rates with it. OpenAI's first-generation XPU is slated for 2027 volume at more than 1GW, but its financing structure has not been shown to be settled — if that program shrinks or slips again, a large slice of FY2027 growth goes with it. Separately, an EU interim measure on the CISPE complaint (five trade groups urged action July 15 2026) forcing VCSP partner reinstatement or pricing rollback would stall VMware ARR growth below 10%
Score Derivation
92.5 base + 1.3 trajectory − 5 risk = 89
Base 92.5 (30–50% CAGR, midpoint 40%; saturating above 30%) + 1.3 trajectory (AI semiconductors accelerating from +143% to +221% YoY; infrastructure software and non-AI semis stable: (1 − 0) / 3 × 4) + 0 margin (stable) − 5 moderate risk (AI-capex digestion and customer concentration, graded with NVDA, TSM and ANET). Software is held stable despite +29% against +9% in Q2 because the step-up is largely upfront license recognition on contracts that lost their termination right, not a demand acceleration. XPU and networking are one AI semiconductor line here because Broadcom reports them as one figure; the old separate networking driver had no revenue number of its own. The old +3 TAM expansion term is retired — primaryType does not score = 89
Key Growth Catalysts
FY2026 AI revenue guidance raised to $56B (~+180% YoY) at Q2 earnings (June 3, 2026); FY2027 $100B+ reaffirmed
Apple renewed its chip supply agreement through 2031 (>$30B), spanning FBAR wireless filters and custom AI silicon — moving Apple from rumored to contracted customer (July 2026)
Meta extended partnership through 2029 (April 14, 2026): multi-GW MTIA custom silicon, Ethernet switches, optical connectivity, and SerDes — most comprehensive rack-level supply agreement to date
Google agreement formalized through 2031 covering future TPU generations, AI networking, and rack-scale components
Anthropic tripling compute commitment to 3GW+ before year-end 2026; Anthropic's $21B in committed XPU orders ($10B placed, $11B additional for late 2026)
OpenAI custom inference engine (Jalapeno) targeting 10GW — but financing unresolved and volume chips now expected mostly in 2027
VMware Cloud Foundation (VCF) becoming the dominant private cloud AI platform as generative AI workloads require VCF-level orchestration
Ironwood Rack solutions (integrated Google TPU racks) representing a new rack-scale revenue stream
Price Scenarios (12–24 Months)
Valuation Analysis
Multi-year customer commitments — Apple through 2031 (>$30B, added July 2026), Meta through 2029, Google through 2031, Anthropic tripling to 3GW+ by year-end — provide forward visibility that few semiconductor names can match. The Q3 FY2026 print (early September 2026) is the next catalyst: guidance is ~$29.4B revenue (+84% YoY) and $16.0B AI revenue (+200%+ YoY), and after June's lesson the setup matters more than the print — a beat-and-raise against a now-reset multiple has asymmetric upside. Two open items to monitor: (1) the OpenAI $18B financing snag — still unresolved, Jalapeno chips slipping to 2027; formal resolution either way removes an overhang or confirms the first major program casualty; (2) the CISPE escalation — five cloud trade groups urged EU interim measures on July 15; an adverse interim decision would hit VMware pricing power directly. Sell-side consensus remains Buy with an average target ~$501. $490.
Valuation Multiples
| Trailing P/E (GAAP) | ~62× |
| Forward P/E (NTM) | ~20× |
| PEG Ratio | ~0.45× |
| Price / Sales (NTM) | ~15× |
| Price / FCF | ~55× |
The June–July de-rating flipped AVGO from premium-priced to relatively cheap: forward P/E compressed from ~36× (May) to ~20×, now ~41% below the semiconductor industry median (~34×) even as forward EPS estimates rose on the $56B FY2026 AI guide and the guided Q3 revenue step-up to ~$29.4B (+84% YoY). The PEG of ~0.45 is deep GARP territory. The multiple compression reflects sector-wide AI fatigue ($1.3T chip-sector wipeout June 5, insider selling, 'great isn't good' whisper-number dynamics) rather than an AVGO-specific demand problem — Q2 actually beat and full-year guidance went up. The risk is that the market is early on capex digestion; the opportunity is that the order book says it isn't.
Approximate figures as of July 2026.
Where We Are vs Targets
Loading live price…
EU antitrust ruling forces VMware pricing rollback; hyperscaler XPU customers accelerate in-house ASIC programs; AI semiconductor demand disappoints expectations.
- EU Commission issues adverse ruling on CISPE complaint, ordering VMware pricing rollback and partner program reinstatement, cutting software ARR growth below 5%
- Google, Meta, or Anthropic materially reduces custom XPU orders or accelerates fully in-house ASIC design, collapsing the path to $100B AI chip revenue in FY2027
- Semiconductor industry downcycle hits networking and broadband segments while AI chip ramp slows from supply or execution challenges
- GAAP amortization burden from VMware acquisition weighs on reported earnings longer than expected, compressing the stock's re-rating
AI chip revenue reaches the guided $56B in FY2026 and $100B+ in FY2027; VMware ARR sustains mid-to-high-teens growth; 27× forward non-GAAP P/E on FY2027 EPS of ~$18.
- VMware Cloud Foundation (VCF) adoption grows across enterprise base, sustaining 15%+ ARR growth at 93% software gross margins
- AI semiconductor revenue reaches the guided $56B in FY2026 and $100B+ in FY2027 as contracted customers (Meta through 2029, Google and Apple through 2031) scale to multi-GW deployments
- Broadcom sustains 68%+ adjusted EBITDA margins while returning $10B+ per quarter to shareholders
- Networking segment benefits from Ultra Ethernet Consortium standardization, cementing Tomahawk as the AI cluster switch of choice
AI revenue exceeds $100B in FY2027 ahead of schedule; sovereign AI programs and new enterprise verticals become XPU customers; VMware becomes the de facto private cloud AI platform.
- AI XPU market expands beyond the 6 known customers; sovereign AI programs (US DoD, EU agencies) and enterprise verticals become incremental XPU customers
- VMware Cloud Foundation becomes the dominant enterprise private cloud platform for AI workloads, with ARR scaling to $15B+ at 75%+ margins
- Broadcom's custom silicon expertise and rack-scale integration prove cost/performance superior to NVIDIA GPUs for inference at scale, capturing >30% of hyperscaler AI accelerator spend
- Dividend doubles within 3 years as free cash flow exceeds $40B annually, triggering multiple expansion