What Powers Broadcom (NASDAQ:AVGO) Data Centre Strategy?
Broadcom (NASDAQ:AVGO) said revenue from AI-focused chips and high-speed networking systems rose 21% in the fiscal third quarter to $4.6 billion, driven by orders from cloud providers building AI clusters.
Source: Kalkine Media · July 21, 2026 at 8:29 AM · AI-assisted report
Single-source
KUALA LUMPUR, 21 JULY 2026 —
Broadcom (NASDAQ:AVGO) said revenue from AI-focused chips and high-speed networking systems rose 21% in the fiscal third quarter to $4.6 billion, driven by orders from cloud providers building AI clusters.
The increase beat the $4.4 billion brokerage consensus tracked by FactSet and came as hyperscalers accelerated spending on custom silicon and Ethernet switches that link thousands of graphics processing units inside new data centres, the company told investors on its August 6 earnings call.
Broadcom’s custom silicon—application-specific integrated circuits designed for individual cloud customers—now accounts for about 40% of semiconductor revenue, up from 32% a year earlier. These chips cut latency and power use in AI training workloads, helping hyperscalers reduce the cost of running large language models and inference services.
Networking revenue climbed 18% to $1.8 billion on stronger demand for 800-gigabit and 1.6-terabit Ethernet switches, optical interconnects and custom accelerators. Broadcom chief executive officer Hock Tan said the Jericho3-AI family of Ethernet chips, introduced in 2023, is now shipping in volumes to three of the top five cloud vendors.
Enterprise software, which supplies virtualisation, security and IT operations tools to banks, manufacturers and telecoms, grew 12% to $1.3 billion. The division’s recurring license model provides steady cash flow that offsets cyclical swings in hardware orders.
Analysts at JPMorgan estimate AI-related sales could reach $15 billion in fiscal 2025, or roughly 55% of total semiconductor revenue, assuming hyperscaler capex grows at a 45% compound rate through next year. “Custom silicon is becoming the primary lever for hyperscalers to differentiate performance and cost,” the bank wrote in a note dated August 6.
Foundry partners TSMC and GlobalFoundries are ramping 5-nanometre and 4-nanometre processes to meet the surge in AI chip orders, while assembly subcontractors in Malaysia and China are expanding pin-count and thermal test capacity to handle larger, power-dense packages, according to a 2024 supply-chain report from TrendForce.
Broadcom expects AI infrastructure revenue to grow at a mid-30% annual rate through 2027, underpinned by demand for next-generation networking silicon that can push 3.2-terabit throughput per switch. The company plans to allocate 70% of its fiscal 2025 R&D budget—about $1.9 billion—to AI-focused products, up from 55% two years ago.
Malaysia Impact
Global development — watch for knock-on effects on oil prices, the ringgit, and KLCI risk sentiment.