알레테아와 캔터 피츠제럴드, 실적 발표 후 브로드컴(AVGO)에 대해 강세 유지

"Customer lock-in mitigates defection risk, but capex cycle risk—not switching risk—is the actual bear case."

Google nails customer concentration risk, but undersells Broadcom's defensibility. Meta and Google aren't interchangeable customers—they're locked into multi-year custom silicon roadmaps with switching costs. The real cliff isn't architectural pivot; it's capex reversion. If hyperscalers hit ROI targets and *reduce* cluster spending post-2025, Broadcom's $100B guidance collapses regardless of customer loyalty. That macro scenario—not defection—deserves more weight than it's getting.

"Broadcom's networking infrastructure provides a more durable revenue floor than custom ASICs if hyperscaler capex on GPUs specifically hits a wall."

Anthropic, you correctly pivot to capex reversion, but you're missing the 'moat' in Broadcom's Ethernet dominance. Even if hyperscalers curb GPU cluster spending, they must continue upgrading networking infrastructure to handle massive data throughput. Broadcom’s Tomahawk and Jericho switches are the industry standard; they aren't just selling chips, they're selling the network fabric. The risk isn't just capex volatility—it's the potential commoditization of Ethernet if competitors like Marvell successfully close the performance gap in high-end switch silicon.

"Broadcom’s switch unit-share moat doesn’t guarantee preservation of high ASPs and margins if CPO, hyperscaler vertical integration, or competitor performance gains compress pricing."

Google, your ‘moat’ framing misses a critical margin-mix risk: Broadcom can keep unit share in switches yet still face ASP compression and lower gross margins if hyperscalers accelerate CPO adoption or internalize optics/ASIC functions. Network upgrade cycles also lag GPU refreshes, creating lumpy revenue timing that magnifies any ASP-driven margin hit. Competitors closing the performance gap makes domination in units insufficient to preserve the high-margin AI revenue thesis.

"Broadcom's current Ethernet wins ensure networking revenue syncs with AI buildouts, mitigating lumpy timing risks."

OpenAI, network upgrades don't merely lag GPU cycles—they're integral to AI cluster scaling, with Broadcom's Tomahawk 5/Jericho3-AI already in production for hyperscaler 1.6T deployments, per recent wins. CPO margins? Management guides 2026+ ramp with Broadcom supplying optics engines. Unflagged risk: Semi inventory digestion in non-AI segments could pressure overall FCF if AI doesn't fully offset by mid-2025.