The number landed like a depth charge. Wolfe Research, a respected sell-side firm, projected Broadcom's AI revenue could hit $200 billion by 2028. That's eight times the consensus estimate for 2025. It's more than NVIDIA's entire revenue in 2024. It's a number that demands a forensic audit, not a press release.
I've seen this pattern before. In 2017, Ethereum's transaction throughput was the bottleneck. Everyone projected it would scale to Visa levels. It didn't. In 2021, NFT wash trading was the norm. The $200B Broadcom prediction smells the same. The infrastructure is real, but the growth expectations are detached from physical reality.
Broadcom is the silent partner in the AI arms race. Its custom ASICs power Google's TPU, Meta's MTIA, and Microsoft's Maia. Its Tomahawk and Jericho switches route traffic in the world's largest AI clusters. Current AI revenue: ~$24B. To reach $200B, Broadcom would need to capture 60-80% of the entire AI chip market by 2028. That's not growth. That's a monopoly.
Let's examine the physics. The first constraint is wafer capacity. TSMC's 3nm and 5nm lines are already oversubscribed. NVIDIA, Apple, AMD, and Qualcomm are fighting for allocations. Broadcom's custom chips are large dies โ 800mmยฒ or more. At $200B revenue, assuming an average selling price of $5,000 per chip, that's 40 million units per year. Each 300mm wafer yields about 50 such dies. That's 800,000 wafers per year. TSMC's total 3nm capacity in 2025 is about 1.5 million wafers. Broadcom would need half of it. Not happening.
Code doesn't confuse volume with value. It's a hard limit.
CoWoS packaging is the next bottleneck. TSMC's CoWoS capacity is expected to reach 60,000 wafers per month by 2026. NVIDIA already consumes 70%. Broadcom's $200B scenario would require 100,000 wafers per month just for their chips. That's a 67% increase in TSMC's entire capacity. Impossible without multi-year investments that haven't been announced.
HBM memory is the third wall. SK Hynix and Samsung are the only suppliers. Their 2026 HBM capacity is about 100 billion GB. NVIDIA's demand alone will eat 80%. Broadcom's chips need HBM too. At $200B revenue, they'd need 30% of global HBM supply. That's a 50% increase in total HBM production. Physical constraints don't care about analyst forecasts.
Then there's power. The electricity required to run 40 million custom AI chips โ assuming 700W per chip โ is 28 GW of continuous load. That's more than the entire data center power consumption of the United States in 2024. The grid infrastructure to support that doesn't exist. It takes 10 years to build a new nuclear plant. The timeline doesn't match.
Let's look at the commercial side. Customer concentration is the silent killer. Google alone accounts for 50% of Broadcom's AI revenue. To reach $200B, Google would need to spend $100B on Broadcom chips by 2028. That's 30% of Google's total 2024 revenue. No CFO would approve that. The other customers โ Meta, Microsoft, Amazon โ are also building their own chips. They're not going to hand Broadcom a 60% margin.
History rhymes. This isn't recycled.
The competitive landscape is brutal. NVIDIA's CUDA ecosystem is a moat that Broadcom cannot cross. Custom ASICs are good for inference, but training remains NVIDIA's playground. Even if Broadcom captures 30% of the inference market, the total addressable market for inference chips in 2028 is estimated at $120B. That gives Broadcom $36B, not $200B.
And then there's the geopolitical risk. Export controls on AI chips to China are tightening. Broadcom's networking chips are already restricted. If the U.S. government expands controls to include custom ASICs, Broadcom loses a significant portion of the sovereign AI market. The prediction assumes a frictionless regulatory environment. That's naive.
Based on my experience auditing blockchain infrastructure, I've learned that the market always overestimates the speed of exponential growth. The same thing happened with Ethereum's scalability, with DeFi's total value locked, with NFT trading volumes. The narrative runs ahead of the cap table. The $200B Broadcom prediction is the largest example yet.
The contrarian view is not to bet against Broadcom, but to bet against the narrative. The market is pricing in a continuation of exponential AI capex. But the gap between AI investment and AI revenue is widening. Cloud providers are spending billions on NVIDIA GPUs but struggling to monetize them. When that gap closes, the correction will be brutal.
This isn't just about Broadcom. It's about the entire AI infrastructure stack, including crypto mining chips and GPU cloud providers. The same capital flows that drive AI hardware also drive crypto mining. When the macro tide turns, both sectors will feel the ebb. The smart money is watching the capex-to-revenue ratio. When it inverts, rotate out of picks-and-shovels plays.
The macro is the only micro.
Let's put the $200B in context. Global AI semiconductor market in 2028 is estimated at $250-300B. Broadcom capturing 67-80% of that is unprecedented. No semiconductor company has ever held that kind of market share. Intel's peak in CPUs was 80%, but that was after decades of dominance. Broadcom is a design house, not a manufacturer. Its margins are thinner, its customer relationships are transactional, not entrenched.
The risk is clear: the market is pricing Broadcom for perfection. At 40x forward earnings, any disappointment triggers a multiple contraction. If AI revenue comes in at $80B in 2028 โ which is still a heroic 3x from 2025 โ the stock could drop 50% as the premium evaporates. The $200B prediction is a catalyst for FOMO, not a fundamental analysis.
I've been through the 2017 ICO boom, the 2020 DeFi summer, and the 2021 NFT mania. The pattern is always the same: a compelling narrative, a supply constraint, and a price that defies logic. Broadcom's $200B is the latest chapter. The code is the only truth. And the code says the math doesn't work.
What should investors do? The cycle is peaking. The gap between AI infrastructure spending and AI revenue is widening. When that gap closes, the picks-and-shovels plays will be the first to fall. Broadcom will be a great company, but at $200B, it's a sell. The smart money is waiting for the correction.
Follow the money, not the memes.
The takeaway is simple: Wolfe Research is paid to generate excitement. The $200B number is a bull case, not a base case. The real question is not whether Broadcom can reach $200B, but whether the market will realize it's impossible before or after the stock crashes. Based on my analysis, the crash comes first. The narrative breaks when the first cloud provider cuts capex. That signal is already flashing.
Monitor the quarterly earnings calls. When Google and Microsoft start talking about ROI on AI spending, the end is near. The infrastructure buildout is real, but it's not infinite. The $200B prediction is a fantasy that will end in tears. I've seen this movie before. The ending is always the same.