Broadcom’s Custom AI Chip Unit Doubles Year-Over-Year with 52% Growth Forecast

AVGOAVGO

Broadcom's custom AI chip business is expected to double year-over-year and it aims for ~52% annual revenue growth, driven by enterprise and hyperscaler demand for ASICs. Trading at a 32.4x forward P/E (vs Nvidia's 24.6x), the company remains a premium AI infrastructure play.

1. AVGO’s Custom AI Chip Momentum

Broadcom has rapidly expanded its custom AI ASIC business, reporting a year-over-year shipment increase of 100% in 2025 as hyperscale cloud providers shift from general-purpose GPUs to tailored accelerators. Management disclosed that AI-related networking and compute revenue grew to $6.8 billion in its fiscal Q2, representing 20% of total sales, and anticipates doubling this contribution by fiscal 2026-end.

2. Robust Revenue Growth Trajectory

Analysts project a compound annual revenue growth rate of 52% for Broadcom across fiscal 2024–2027, driven by strength in AI infrastructure and enterprise networking. In the trailing twelve months through Q2, AVGO generated $48 billion in revenue, up 45% year-over-year, while bookings for its Silicon One networking chips rose by 60%, reflecting accelerating demand for high-speed data-center interconnects.

3. Premium Valuation Reflects Growth Profile

Broadcom trades at a forward price-to-earnings multiple of 32.4x, compared with 24.6x for its nearest GPU-focused peer, underscoring investor willingness to pay for its diversified AI infrastructure exposure. With a projected fiscal 2026 earnings growth of 35%, AVGO’s PEG ratio stands near 0.9, suggesting valuation alignment with its high-visibility AI and networking growth.

4. Strategic Hyperscaler Partnerships

Broadcom has secured multi-year supply agreements with leading cloud providers that commit over $25 billion in aggregate spending on custom AI ASICs and high-bandwidth networking gear through 2028. These partnerships underpin AVGO’s expected annual R&D investment of $3.5 billion, ensuring leadership in next-generation silicon for both compute and data-transport layers.

Sources

FFFFF
+6 more