Topic: Additional custom chip prospects beyond the seven identified
Key points:
AVGO segments the AI market into LLM developers (frontier model players) and enterprise (not addressed).
Currently has seven identified: four are customers, three are prospects; AVGO is "very picky and careful" about qualifying.
Sees potential for "one more perhaps as a prospect," but no more beyond that for now.
Mgmt stance: Neutral — Hock Tan states the universe is narrow and AVGO is selective, with no expansion expected soon.
Q7 — Tom O'Malley
Topic: Jericho 4 switch and scale-across networking for AI clusters
Key points:
Jericho 4 is a 51.2 terabit per second switch for networking across multiple data center sites (within ~100 km) to form a single cluster.
Technology uses deep buffering and congestion control, proven over 10–20 years in telco routing; runs on Ethernet.
Jericho 3 (256 connections) has been shipping to a couple of hyperscalers for the last two years.
Mgmt stance: Bullish — Hock Tan emphasizes proven, stable technology already in use, with Jericho 4 addressing growing bandwidth needs for AI training.
Q8 — Carl Ackerman
Topic: VMware VCF adoption and cross-selling benefits
Key points:
Over 90% of top 10,000 accounts have bought VCF licenses, but full deployment is ongoing (hard work over next two years).
Second phase: enabling customers to deploy and operate VCF on-prem, then selling advanced services (security, disaster recovery, AI).
No cross-selling benefit to hardware (servers, storage, networking) — VCF commoditizes underlying hardware.
For next 20,000–30,000 midsized companies, AVGO is still learning; TCO and skill sets may limit adoption.
Mgmt stance: Neutral — Hock Tan is confident in top 10,000 but cautious on broader enterprise tail, calling it "an area we're still learning."
Q9 — CJ Muse
Topic: Gross margin guidance (down 70 bps)
Key points:
Q3 guide implies gross margin of 77.7%, down 70 bps sequentially.
Drivers: XPUs (TPUs) and wireless revenue (both lower-margin) are increasing; software revenue is also coming up slightly.
Kirsten Spears confirms wireless is typically the heaviest quarter of the year.
Mgmt stance: Neutral — Kirsten Spears explains the mix shift (higher XPU/wireless, slightly higher software) without providing specific margin breakdowns.
Q10 — Joe Moore
Topic: Categorization of the fourth customer and timing of $10 billion orders
Key points:
All seven customers do LLMs; some have large platforms, others may create them — hard to differentiate hyperscale vs. LLM makers.
The $10 billion in orders will likely be delivered in the second half of fiscal 2026, more precisely Q3 of fiscal 2026.
Hock Tan does not specify the start/end of deployment within Q3.
Mgmt stance: Neutral — Hock Tan provides a specific timeframe for delivery but avoids further categorization of customers.
Q&A Batch (11-12 of 12)
Q11 — Joshua Buchhalter
Topic: Scale-up Ethernet momentum vs. UALink/PCIe, and AI networking opportunity
Key points:
Ethernet solutions are disaggregated from AI accelerators; Broadcom treats them as separate, open-source, and customer-choice driven.
For XPU customers, Broadcom optimizes networking switches hand-in-hand with XPUs, all using Ethernet interfaces; hyperscalers use Broadcom Ethernet switches for scale-out GPU clusters.
Scale-up Ethernet (e.g., Tomahawk with sub-250 nanosecond latency) is expected to see increased use in scale-up scenarios, especially with XPUs.
Mgmt stance: Bullish — Ethernet is openly enabled and proven; Broadcom leads in Ethernet switches for hyperscaler scale-out, and scale-up Ethernet momentum is tied to XPU adoption.
Q12 — Christopher Rolland
Topic: Competition in ASIC and networking (Ethernet vs. UALink/PCIe)
Key points:
Ethernet is well-proven, known to hyperscaler architects; Broadcom can tweak switches to achieve latency better than NVLink/InfiniBand (sub-250 nanoseconds).
Broadcom faces competition in Ethernet (open-source system) and ASIC; strategy is to out-invest and out-innovate, leveraging IP like SerDes and low-power design.
For XPUs, all customer agreements make them compatible with Ethernet, not proprietary protocols; Broadcom is first to create the XPU model on silicon.
Mgmt stance: Bullish — Ethernet is the logical choice; Broadcom’s investment in IP and innovation keeps it ahead in both networking and ASIC competition.