On July 29, 2026, Broadcom (AVGO) released its fiscal Q3 2026 (ended July 2026) earnings, with its custom AI chip (ASIC) business performing impressively: shipments surged 150% YoY, up 45% QoQ, contributing over $3.5 billion in quarterly revenue, far exceeding market expectations. This data marks ASIC's shift from periphery to core in the AI computing power market, especially demonstrating substitution potential for GPUs in inference scenarios.
Performance Highlights: Google and Meta Orders Propel Growth
Broadcom CEO Hock Tan noted in the earnings call that the explosive growth of custom AI chips mainly came from two hyperscale customers: Google and Meta. Google's TPU v6 chip (manufactured by Broadcom) has expanded applications in AI search and ad recommendation systems, while Meta's custom inference chip (codenamed "Artemis") is fully deployed for content recommendation and generative AI services on its social platforms. Together, the two companies contributed over 80% of Broadcom's AI chip revenue.
Additionally, Broadcom revealed signing a next-gen 2nm ASIC design contract with a new customer (market speculation: OpenAI or ByteDance), with mass production expected in 2027. This further solidifies its leadership in custom chip design.
Industry Analysis: Why ASIC Rises in AI Inference?
Traditionally, AI training and inference tasks have been dominated by Nvidia GPUs. However, as AI models mature, inference demand share increases. ASICs, with lower power consumption, higher throughput, and targeted optimization, are gaining favor from internet giants. Broadcom's ASICs typically use specific architectures (e.g., Systolic Array) and advanced packaging (CoWoS), achieving 2-3x the energy efficiency of Nvidia H100 in batch inference scenarios.
Research firm SemiAnalysis points out that the global AI inference chip market in 2026 is about $80 billion, with ASIC penetration rising from 15% in 2024 to 30%. Broadcom, as the largest independent ASIC designer, holds over 40% market share, ahead of Marvell and MediaTek.
Comparison with Nvidia: Complementary or Substitutive?
Broadcom's ASIC does not directly challenge Nvidia's dominance in training, but focuses on inference. Nvidia GPUs excel at flexible general-purpose parallel computing, while ASICs have advantages in fixed algorithm optimization. For example, Meta's recommendation model using ASIC reduced latency by 50% and power consumption by 60%. However, long development cycles and high customization costs limit penetration to small and medium customers.
In the US-listed AI chip sector, Broadcom and Nvidia form differentiated competition. Broadcom's stock has risen 68% year-to-date, with a market cap exceeding $1.2 trillion, while Nvidia gained 45% in the same period. Some investors now view Broadcom as the "second pole" of AI computing investment.
Supply Chain Impact: Advanced Packaging and Optical Modules Benefit
The surge in Broadcom ASIC shipments directly boosts its supply chain partners. In advanced packaging, TSMC's CoWoS capacity remains tight, and Broadcom has become one of its top three customers. In optical modules, Broadcom's own Tomahawk switch chips and AI data center interconnect demand are rising simultaneously, with 800G optical module shipments up 60% QoQ. This has driven stock performance of US-listed optical communications companies such as Lumentum and Coherent.
Notable ETFs: Broadcom's weight in VanEck Semiconductor ETF (SMH) has risen to third place, while AI-themed ETFs such as BOTZ and AIQ have also increased allocations to Broadcom.
Risk Warning and Outlook
Despite strong performance, Broadcom faces the following risks:
- Customer Concentration: Top three customers account for over 70% of revenue; loss would have a huge impact.
- Technology Evolution: Nvidia's next-gen Rubin GPU may narrow the efficiency gap and offer customized services.
- Geopolitical Risks: US export restrictions on AI chips to China may affect Broadcom customers (e.g., ByteDance); monitor policy changes.
Looking ahead, Broadcom plans to launch 1.6nm ASIC in 2027 and deeply integrate its networking chips with AI accelerators. Analysts expect Broadcom's AI business revenue to exceed $20 billion in fiscal 2027, making it the world's second-largest AI chip company after Nvidia. For US stock investors, Broadcom is the top pick for the ASIC track.
