Broadcom Q2 FY26 earnings call AVGO

AI semiconductor revenue hit a record $10.8B in Q2, up 143% YoY and above outlook, and is guided to accelerate to $16B in Q3 (>200% YoY). Management expects H2 FY26 AI revenue to double H1 (~$19B), supporting a full-year FY26 figure of ~$56B (+~180% vs FY25). Beyond that, FY27 AI revenue is reiterated at 'in excess of $100 billion' (~2x FY26) with FY28 expected to grow substantially further, underpinned by initiatives across six core customers. Hock characterized demand for XPUs and networking as 'simply insatiable.'

Broadcom detailed a widening AI customer roster: a long-term Google agreement for multiple TPU and AI-networking generations (April); Anthropic access to >1 GW of TPU compute in 2026 plus another 5 GW of next-gen TPU compute from 2027; OpenAI's contractual 1.3 GW in 2027 within a 10 GW-by-2029 agreement (silicon delivered, production late 2026); and a Meta MTIA partnership to deploy 3 GW through end-2028, with an initial 1 GW order (XPUs + networking) delivering H2 2027. Two additional customers are set to ship late 2026 and have placed $6B of purchase orders to date. Management plans ~10 GW of shipments in 2027, back-half loaded^⚖️.

To fund chips for frontier labs that may struggle to finance capacity themselves, Broadcom is creating an 'AI XPV platform' with Apollo, Blackstone and other large-balance-sheet investors to deploy more than 20 GW of compute through 2028. The first tranche, valued at $35B, is currently being launched by Apollo. The vehicle is designed to deliver compute at the lowest cost and power to labs including Anthropic and OpenAI while keeping Broadcom in its pure chip-supplier model.

Networking represented almost 40% of Q2 AI revenue — a level Hock expects to normalize closer to 30% as XPU compute scales. Broadcom claims at least a one-generation lead: the 100-terabit Tomahawk 6 Ethernet switch has shipped for over a year, a next-gen 200-terabit switch tapes out this quarter, and the company is the de facto CPO standard (1.6T DSPs, CW and EML lasers). Jericho3 and Jericho4 fabric solutions enable the largest multi-hyperscaler cross-data-center deployments. Networking carries richer margins than XPUs/ASICs, partly offsetting mix dilution.

Consolidated gross margin fell 230 bps YoY to 77.1% and is guided to ~74% in Q3 as fast-growing, lower-margin AI/TPU revenue and semiconductors become a larger share of mix. Management stressed this is not a structural semiconductor-margin change: semiconductor gross margin held around 70% and operating leverage kept operating margin flat at ~67%. Investors were explicitly urged to model semiconductor and software margins separately given shifting revenue mix.

Non-AI semiconductor revenue was $4.2B, up 6% YoY, with bookings exceeding $6B — which management cited as a clear indication of a path toward full cyclical recovery. Broadband, server storage and enterprise networking rose, partially offset by a seasonal decline in wireless. Q3 non-AI revenue is guided to ~$4.5B, up 12% YoY.

Software revenue was $7.2B, up 9% YoY, with ARR sustaining 17% YoY growth and a 93% gross margin. The newly released VMware Cloud Foundation 9.1 adds heterogeneous compute support across AMD, Intel and NVIDIA platforms plus enterprise AI-inferencing support, and strong global server demand is driving robust on-prem private-cloud adoption. Q3 software revenue is guided to ~$8.9B, up 31% YoY; management sees no negative agentic-AI impact given the hypervisor-level positioning.

AI bookings exceeded $30B in the quarter against $10.8B shipped, as customers order far ahead to secure not just wafers, HBM and DRAM but also power. Management says visibility now runs to 2028 (versus roughly 2027 three months ago). On supply, Hock said Broadcom has secured needs for 2026-2027 and is working on 2028-2029, and can generally meet incremental customer demand for wafers and HBM. Inventory rose to 86 days (from 68 in Q1) as a deliberate build ahead of H2 AI acceleration.