Broadcom AI Forecast Disappoints Investors

🔑 Enhanced Key Takeaways

• •Broadcom's AI networking backlog exceeds $73 billion over the next six quarters, with FY26 and FY27 AI networking revenues projected at $17 billion and $30 billion respectively—43% and 64% above market expectations[1], indicating substantial pipeline strength despite investor tepidness[3].
• •The company has secured a landmark $350 billion deal with OpenAI to supply 10 gigawatts of custom AI chips, with deliveries beginning in H2 2026 and completion by end of 2029, equating to approximately $100 billion in annual revenue[2].
• •Broadcom's custom ASIC strategy is enabling hyperscalers to bypass the 'Nvidia Tax' by co-designing application-specific chips at lower costs; Alphabet's Sunfish TPU and Meta's MTIA v4 chips exemplify this trend, reducing reliance on general-purpose GPUs[4].
• •The semiconductor industry is projected to reach $1 trillion in annual revenue by 2026, with global sales at $791.7 billion in 2025 and expected 26% growth in 2026, positioning Broadcom as a critical infrastructure provider across the AI stack[3].
• •Big Tech capital expenditure is forecast to reach $550–$600 billion in 2026 (up from $380 billion in 2025), with Google committing $175–$185 billion and Amazon $200 billion, creating sustained demand tailwinds for Broadcom's AI hardware and networking solutions[3].

🛠️ Technical Deep Dive

• •Tomahawk 6 Switching ASIC: World's first 102.4 Tbps chip capable of supporting 1.6T Ethernet ports, enabling networking of up to 1 million AI accelerators in a single cluster—a scale previously considered impossible[4].
• •Custom ASIC Architecture: Application-specific integrated circuits are hardwired before manufacturing to perform particular tasks more efficiently than general-purpose GPUs, reducing power consumption critical for AI workloads[2].
• •Jericho3-AI Chips: Specialized hardware designed to manage complex traffic patterns in multimodal AI models (video, audio, text simultaneously), addressing emerging infrastructure demands[4].
• •Agentic AI Infrastructure Pivot: Broadcom is developing specialized hardware for autonomous AI agents requiring near-zero latency, representing a strategic shift beyond traditional data center networking[4].
• •OpenAI Custom Chip Specifications: 10 gigawatts of custom AI chips valued at $350 billion based on current Nvidia GPU pricing, with H2 2026 delivery start and 2029 completion timeline[2].

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline