Wolfe Research's $200B Broadcom AI Prediction: A Reality Check from the Supply Chain Trenches

Wolfe Research projects Broadcom's AI revenue could reach $200 billion by 2028. That is more than NVIDIA's total revenue in fiscal 2024. It is a number that demands verification. Context: The prediction, published as a sell-side note, places Broadcom at the center of the AI infrastructure boom. Broadcom's current AI revenue is approximately $20-24 billion (fiscal 2025 consensus), driven by custom ASICs (Google TPU, Meta MTIA, potential Microsoft Maia) and networking chips (Tomahawk, Jericho). The $200 billion figure implies a compound annual growth rate of 70-90% over three years. No semiconductor company has achieved that scale of growth. NVIDIA's explosive run from $27 billion to $130 billion — a 4.8x increase — remains the industry record. Broadcom's required 8.3x leap is unprecedented. Core Analysis: The seven dimensions of feasibility — technical, commercial, competitive, infrastructure, ethical, investment, and supply chain — all point to the same conclusion: $200 billion is a tailcase scenario, not a baseline. The most actionable constraint is physical. To reach $200 billion in AI chip revenue, Broadcom would need to ship roughly 400-500 million custom ASICs at an average selling price of $4,000-5,000. That requires 3-5 hyperscale customers each placing $20-30 billion annual orders. Only a handful of firms globally have the appetite for that scale. Google, Broadcom's largest AI customer, contributed an estimated $10-12 billion in 2025. Asking Google to triple that to $30-40 billion by 2028 is plausible. Asking five other customers to each match that is not. The infrastructure bottleneck is even tighter. Advanced packaging (CoWoS) capacity at TSMC is the gating factor. NVIDIA already consumes over 60% of TSMC's CoWoS output. Broadcom's current allocation is roughly 15-20%. To support $200 billion in revenue, Broadcom would need 50-60% of CoWoS capacity — equivalent to 10-15 million wafers per month. TSMC's 2025 CoWoS capacity is 4-6 million wafers per month. Even with aggressive expansion, tripling capacity by 2028 is optimistic but not impossible. The problem: TSMC will prioritize NVIDIA, Apple, and AMD — customers with higher margin per wafer. Broadcom's ASIC wafers yield lower profit per chip than NVIDIA's GPUs. TSMC's allocation logic is economic, not technical. HBM (High Bandwidth Memory) supply is another hard ceiling. SK Hynix, Samsung, and Micron control the entire HBM market. In 2025, total HBM output is ~50-60 billion GB-equivalents. NVIDIA consumes over 70%. For Broadcom to hit $200 billion in AI revenue, it would need 20-30% of global HBM supply — requiring dedicated capacity expansion that takes 2-3 years. No HBM supplier has announced such a commitment for Broadcom specifically. Power constraints are the ultimate limit. The chip volume implied by $200 billion revenue would require 100-200 GW of data center power. Global data center power consumption in 2024 was ~500 TWh, with AI consuming about 100 TWh. Scaling to 200 GW means a 10x increase in AI power demand. Grid infrastructure cannot expand that fast. The result: AI chips will be underutilized, reducing actual revenue. Contrarian Angle: The $200 billion prediction is not an objective forecast. It is a strategy — a bullish scenario designed to anchor institutional investors' upside expectations. Sell-side analysts often publish extreme numbers to generate attention and maintain relevance in a hot sector. Crypto Briefing, the outlet that amplified this note, has a readership inclined toward optimistic narratives. The omission of risk factors — customer concentration, supply chain bottlenecks, AI capex cycle risk — is a classic case of selective reporting. The real risk is not that Broadcom falls short of $200 billion. It is that the AI infrastructure investment cycle peaks before 2028. The gap between cloud provider AI capital expenditure and AI application revenue is widening. In 2025, the top five hyperscalers spent $200 billion on AI infrastructure but generated only $50 billion in AI-related revenue. That gap is unsustainable. If it closes — either through revenue acceleration or capex cuts — Broadcom's AI revenue will land closer to $60-100 billion, not $200 billion. Takeaway: Wolfe Research's $200 billion prediction is a useful measure of market sentiment, not a reliable forecast. Investors should focus on the physical constraints: CoWoS capacity, HBM supply, and power grid expansion. Those are the real limits. The gap between AI spending and AI revenue is the signal to watch. If it narrows, Broadcom's AI business will grow — but at a humanly achievable rate. If it widens, the sell-side dream will become a memory. Verify everything, trust nothing. Skepticism is the first line of defense. Code is the only law that holds.