AVGO · US
AVGO.US: Q2'26 update — custom XPU + AI networking dual engine
Broadcom FY Q2 (ended 2026-05-03) revenue $22.2B (+48%), AI semiconductors $10.8B (+143%), adj EBITDA margin 69%. Q3 guide ~$29.4B total and ~$16B AI semi; 2027 AI semi target >$100B. Custom ASIC (XPU) + AI networking thesis is clear with Google / Meta / OpenAI as customers, but concentration and rich valuation matter. Complements NVDA rather than replacing it. See also Oracle and TSMC.
Cite a section with a deep link, e.g. /en/r/avgo-us-research#thesis
Q2 FY2026 snapshot
- Q2 revenue (+48%)
- USD 22.2B
- AI semiconductors (+143%)
- USD 10.8B
- Non-GAAP EPS
- USD 2.44
- Adj EBITDA margin
- 1
Q3 guidance
- Q3 total revenue
- USD 29.4B
- Guide YoY
- 1
- Q3 AI semiconductors
- USD 16.0B
- 2027 AI target
- >$100B
Thesis
Broadcom’s AI infrastructure role has converged from diversified semiconductors + software to a custom ASIC (XPU) + AI networking dual engine. FY Q2 2026 (ended 2026-05-03) validated that mix with $22.2B revenue (+48% YoY) and AI semiconductors $10.8B (+143%). Q3 guide of ~$29.4B total (+84%) and ~$16B AI semi (+200%+), plus a reiterated 2027 AI semi >$100B, shows hyperscaler custom-silicon demand is still compounding.
The premium holds only if:
- XPU custom ASICs stay indispensable — Google TPU generations, Anthropic, Meta, OpenAI co-design programs keep ramping; Broadcom is one of few partners that can deliver at scale;
- AI networking scales with cluster size — Tomahawk / Jericho switching is the vascular system of rack-scale AI; competes and coexists with NVIDIA Spectrum;
- VMware anchors cash flow — infrastructure software ~$7.18B (+9%), 69% adj EBITDA margin, ~$10.3B FCF fund semi capex and returns.
Hock Tan’s playbook is M&A integration plus deep customer lock-in; the current mandate is taking AI semiconductors from a handful of mega-programs to a >$100B platform. This complements NVDA’s GPU training stack rather than replacing it.
Business
Model: Custom AI XPUs with hyperscalers + AI networking ASICs + VMware software subscriptions.
Flagships: Custom XPU/AI semi; Tomahawk/Jericho; VMware.
Competitiveness: Co-design lock-in; semi + software margin stack.
Strategy: 2027 AI semi >$100B narrative; watch captive ASIC substitution.
Value chain
Mid-stack custom silicon & networking between TSMC and cloud CapEx.
Upstream: TSMC. Downstream: hyperscalers. Risks: leading-edge capacity; rising captive silicon share.
Valuation
Custom ASIC themes drove a strong 2026 move. As of this note (2026-07-30), valuation already prices part of the 2027 >$100B AI semi path. A 69% adj EBITDA margin is best-in-class, but customer concentration + AI capex sensitivity make risk/reward asymmetric at elevated levels.
Valuation frame
| Metric | Status | Read |
|---|---|---|
| 盈利增长 | 强劲加速 | Q2 +48% YoY；AI 半导体 +143% |
| 盈利能力 | 行业顶尖 | Adj EBITDA 69% of rev；FCF ~$10.3B |
| 估值 | 偏高 | AI 定制 ASIC 叙事已大幅定价 |
| 客户集中度 | 高风险 | Google / Meta / OpenAI 等少数 hyperscaler 主导 |
| 相对 NVDA | 互补非替代 | XPU 定制 + 网络 vs GPU 训练平台 |
| AI CapEx 周期 | 敏感 | hyperscaler 支出放缓将直接冲击订单 |
Versus NVDA: AVGO does not compete head-on in training GPUs; it captures custom XPU + networking increment. The valuation frame is closer to “AI infrastructure necessity” than “AI platform monopoly.” Versus Oracle’s cloud + AI infra narrative, AVGO sits deeper in silicon.
Financial trend (~24 months)
Eight-quarter revenue and gross margin with YoY and QoQ. Semi/AI hardware seasonality is secondary to supply, ASP, and order timing.
Revenue · last 8 quarters
Interactive chart available in the reader.
Gross margin · last 8 quarters
Interactive chart available in the reader.
Operations
Q2 FY2026 operations
Q2 P&L and segments
| Item | Q2 FY2026 | Note |
|---|---|---|
| 营收 | $22.187B（+48%） | FY Q2 截至 2026-05-03 |
| 半导体解决方案 | ~$15.1B | 含 AI 半导体 $10.8B |
| AI 半导体 | $10.8B（+143%） | 定制 XPU + AI 网络 |
| 基础设施软件 | ~$7.18B（+9%） | VMware 订阅模式 |
| Adj EBITDA | $15.2B（69%） | 利润率行业领先 |
| FCF | ~$10.26B | 强劲现金转化 |
| Non-GAAP EPS | $2.44 | — |
Segment mix
| Segment | Revenue mix | Watchpoint |
|---|---|---|
| 半导体解决方案 | ~68% | Q2 ~$15.1B；AI 占半导体 ~71% |
| AI 半导体 | ~49% | Q2 $10.8B；+143% YoY |
| 基础设施软件 | ~32% | VMware ~$7.18B；+9% YoY |
| 非 AI 半导体 | ~19% | 宽带 / 存储 / 工业等 |
Takeaways:
- AI semiconductors ~49% of revenue — upgraded from a semi sub-segment to the company’s primary growth engine; non-AI semi (broadband, storage) still funds the base but carries less weight.
- Adj EBITDA $15.2B (69%) and FCF ~$10.26B validate earnings quality — why hyperscalers stay locked in long term.
- VMware software +9% YoY — subscription transition on track; cash-flow ballast and valuation anchor.
AI semiconductors
Broadcom’s AI semi splits into two lines:
- Custom XPU (ASIC) — co-designed AI accelerators with hyperscalers for training and inference. Google TPU is the most mature case; Anthropic, Meta, OpenAI programs are ramping.
- AI networking — Ethernet switching (Tomahawk 5/6, Jericho) is critical for AI cluster scale-out; every new GPU rack adds networking ASIC demand.
Watch AI semi quarterly revenue and QoQ slope, new customer / next-gen XPU announcements, networking ASP and share, and TSMC leading-edge capacity. If AI capex plateaus, AVGO’s elasticity may be lower than pure-GPU names (NVDA), but customer concentration risk is higher.
VMware software
Infrastructure software (VMware) FY Q2 ~$7.18B (+9% YoY). Post-acquisition subscription transition delivers:
- Recurring revenue anchor — dampens semi-cycle volatility;
- Enterprise IT penetration — private / hybrid cloud demand runs parallel to AI infra spend;
- M&A synergy — 2023 VMware integration still releasing efficiency.
Software growth trails semiconductors, but 69% adj EBITDA margin quality makes Broadcom a rare “AI theme + cash flow” compounder. Track VMware subscription ARR — if software slows while AI semi misses, double-hit risk rises.
Competition
Strategy / product / risk map
| Category | Detail | Implication |
|---|---|---|
| 定制 ASIC (XPU) | Google TPU 代际、Anthropic、Meta 等 co-design | AI 半导体核心增长引擎 |
| AI 网络 | Tomahawk / Jericho 交换芯片；机柜级互联 | 与 NVDA Spectrum 竞争 |
| VMware 软件 | 订阅转型；私有云 / 混合云 | 现金流锚；增速低于半导体 |
| 大客户 | Google、Meta、OpenAI、Anthropic | 集中度是双刃剑 |
| 制造 | TSMC 先进制程依赖 | 产能分配与 geopolitics 风险 |
| 估值 | AI 主题 + 高 margin 溢价 | CapEx 周期回调风险 |
Three axes:
| Axis | Broadcom | NVDA | Cloud in-house |
|---|---|---|---|
| AI accel | Custom XPU co-design | GPU training platform | TPU / Trainium / MTIA |
| Networking | Tomahawk / Jericho Ethernet | Spectrum / InfiniBand | Partial in-house |
| Customer lock | Deep co-design + long contracts | CUDA ecosystem | Internal priority |
AVGO’s moat is mega-customer co-design + delivery capability, not software ecosystem — a different dimension from NVDA’s CUDA wall. Complement > substitute. But rising in-house silicon (Google TPU internalization, Amazon Trainium, etc.) is a medium-term structural risk.
Peer comparison
| Company | Share / role | GM | Strength | Weakness |
|---|---|---|---|---|
| Broadcom (self) | Custom XPU + AI net | EBITDA ~69% | Hyperscaler co-design | Customer concentration |
| NVIDIA | Merchant GPU | ~75% | CUDA platform | Less captive ASIC |
| Marvell | Custom / optics | ~60% | Custom silicon | Scale vs AVGO |
| Arista | AI eth switching | ~64% | EOS software | No XPU franchise |
Management
CEO Hock Tan, CFO Kirsten Spears, Chair Henry Samueli stable. No material 24m change. Stability: steady.
Key management (24m)
| Role | Name | Since | 24m change |
|---|---|---|---|
| CEO / President | Hock Tan | 2006 | No change |
| CFO | Kirsten Spears | 2020-12 | No change |
| Chair | Henry Samueli | 2018 | No change |
Outlook
Company guide · Q3 FY2026
| Item | Guide | Note |
|---|---|---|
| Q3 总营收 | ~$29.4B | 约 +84% YoY |
| Q3 AI 半导体 | ~$16.0B | 约 +200%+ YoY |
| 2027 AI 半导体 | >$100B | 管理层重申长期目标 |
| 财报发布 | 2026-06-03 | FY Q2 业绩（截至 5/3） |
Framework:
- Delivering ~$29.4B Q3 total (+84% YoY) would re-rate the full-year path.
- ~$16B Q3 AI semi (+200%+) is the key validation metric — any miss triggers “capex slowdown” narratives.
- 2027 >$100B AI semi is the long-term valuation anchor; separate “guide reiteration” from “path dependency” (sustained customer capex, limited in-house substitution).
- Q2 Non-GAAP EPS $2.44 run-rate still has room to expand on Q3 growth, but margins are near ceiling.
Scenarios
| Scenario | Conditions | Implication |
|---|---|---|
| Bull | Q3 AI ≥$16B; 2027 >$100B path gets new mega-wins; VMware ARR accelerates | AI infrastructure necessity narrative strengthens |
| Base | Q3 roughly delivers; AI semi stays 100%+ YoY; VMware high-single-digit | Hold; watch concentration |
| Bear | Q3 AI miss; hyperscaler capex slows; in-house ASIC substitution accelerates | 15–25% multiple compression possible |
Positioning frame (research only): AI infrastructure complement; avoid aggressive chase into an overheated AI tape. Track AI semi QoQ slope, next-gen XPU, VMware ARR, customer capex commentary.