A Wall Street analyst predicts Broadcom will earn $200 billion from AI by 2028. That figure is more than NVIDIA’s total revenue last year, and nearly four times Broadcom’s entire 2024 revenue. As a cryptographer who has spent nearly three decades building bridges between code and human trust, I read predictions like this with a mix of curiosity and caution. The numbers are seductive, but they hide a deeper story about the limits of physical infrastructure, the fragility of centralized bets, and the quiet power of decentralized alternatives.
Let me start with a confession: I have been wrong about predictions before. In 2017, I spent four months auditing the Telegram Open Network whitepaper. I found a critical game-theory flaw that ignored small-holder participation. My 40-page critique was shared across 15 Telegram groups, reaching 50,000 people before the project eventually halted. That experience taught me that technical correctness without social empathy leads to community fragmentation. The same lesson applies here. The $200B prediction is technically possible on paper, but it ignores the human and physical realities that determine whether a vision becomes reality.
Context: The Prediction and the Landscape
Wolfe Research, a respected sell-side firm, released a report suggesting Broadcom could see $200 billion in AI revenue by 2028. This is an extraordinary claim. Broadcom’s current AI semiconductor revenue is around $20-24 billion for fiscal 2025, driven by custom AI accelerators (XPUs) for Google’s TPU, Meta’s MTIA, and possibly Microsoft’s Maia chips. The company also benefits from its Ethernet networking chips (Tomahawk, Jericho) that power large-scale AI clusters. The prediction implies a compound annual growth rate of 70-90% over three years, which is unprecedented in the semiconductor industry. Even NVIDIA, riding the AI boom, grew from $27 billion to $130 billion over two years—a 4.8x increase. Broadcom would need an 8.3x increase from its current base.
Crypto Briefing, a blockchain-focused media outlet, reported this prediction. As someone who founded a Web3 community, I find it interesting that a crypto news site is covering a traditional semiconductor company’s revenue forecast. The crossover reflects how AI infrastructure is becoming a shared concern for both centralized and decentralized computing ecosystems. But the report omitted crucial context: the assumptions behind the prediction, the probability weighting, and the physical constraints that make $200B a tail scenario, not a base case.
Core Analysis: The Physics of $200B
From code audits to community heartbeats, I’ve learned that the most elegant architectures fail if they ignore real-world constraints. Let me walk through the barriers that make $200B a near-impossible target.
1. Silicon and Packaging Constraints
Broadcom’s AI chips rely on TSMC’s most advanced nodes (3nm, soon 2nm) and CoWoS advanced packaging. To achieve $200B in revenue, Broadcom would need to ship approximately 400-500 million custom AI chips annually, assuming an average selling price of $4,000-5,000 per chip. That would require roughly 500,000-600,000 12-inch equivalent wafers per year just for compute chips, plus additional wafers for networking. TSMC’s total 3nm/5nm capacity in 2025 is about 1.5-1.8 million wafers annually. NVIDIA already consumes 30-40% of that, Apple takes 20-30%, and the rest is shared among AMD, Broadcom, and others. For Broadcom to capture $200B, it would need to monopolize TSMC’s advanced capacity, which is impossible without displacing NVIDIA and Apple. TSMC naturally prioritizes customers with higher margins per wafer—NVIDIA’s GPUs yield more value per wafer than Broadcom’s ASICs, which have lower margins due to design service costs.
CoWoS packaging is an even tighter bottleneck. TSMC’s CoWoS capacity in 2025 is about 40,000-60,000 wafers per month. NVIDIA consumes over 60% of that. To support Broadcom’s $200B revenue, CoWoS capacity would need to triple to 150,000 wafers per month by 2028. That is possible if TSMC invests heavily, but it would require massive capital expenditure and time. The supply chain cannot scale overnight.
2. Memory (HBM) Supply
AI chips require High Bandwidth Memory (HBM), currently dominated by SK Hynix, Samsung, and Micron. Total HBM supply in 2025 is about 50-60 billion gigabytes. NVIDIA consumes 70% or more. Broadcom’s ASICs would need an additional 20-30% of global HBM supply to support $200B in revenue. HBM production is also constrained by packaging and advanced process nodes. Expanding HBM capacity takes 2-3 years and requires multi-billion dollar investments. The competition for HBM is already intense, and NVIDIA has long-term agreements with suppliers. Broadcom would need to secure similar commitments, which is possible but not guaranteed.
3. Power and Data Center Constraints
The $200B revenue implies deploying enough AI chips to consume 100-200 gigawatts of power. For perspective, the entire global data center power consumption in 2024 was about 500 terawatt-hours annually, with AI accounting for roughly 100 TWh. Adding 100-200 GW of load would require building dozens of new data centers and upgrading power grids. The grid infrastructure expansion is slow, regulated, and often opposed by local communities. Many data center projects are already facing delays due to power availability. Even if the chips are manufactured, they may not be able to run at full capacity.
4. Customer Concentration
Broadcom’s AI revenue is heavily concentrated in a few hyperscale customers. Google is the largest, contributing over 50% of Broadcom’s AI chip revenue. For Broadcom to reach $200B, Google alone would need to purchase about $100 billion in custom chips annually—that’s 30% of Google’s total 2024 revenue of $350 billion. It is unlikely that any single company would allocate that much of its revenue to one vendor. Even if Google, Meta, Microsoft, Amazon, and Apple all became Broadcom customers, each would need to spend $20-30 billion annually on Broadcom chips. The number of companies with that level of AI compute demand is fewer than ten globally. The customer base is not wide enough to support $200B without a massive expansion into sovereign AI and enterprise markets, which are slower to adopt.
5. Competitive Pressure from NVIDIA
NVIDIA is not standing still. Its next-generation Rubin architecture, expected in 2026-2027, will likely maintain a 1.5-2x performance-per-watt advantage over custom ASICs in training workloads. NVIDIA’s CUDA ecosystem is a powerful moat that ASICs cannot easily replicate. Broadcom’s ASICs are competitive in inference and specific training workloads, but they cannot replace NVIDIA in the general-purpose AI training market that drives the bulk of AI compute spending. If NVIDIA responds by pricing aggressively in the inference segment, Broadcom’s ASIC value proposition erodes.
Building bridges where DeFi once built walls: The same dynamic applies in AI hardware. NVIDIA has built a walled garden with CUDA, while Broadcom offers a more open, customizable alternative. But openness does not always win if the incumbent continues to innovate.
Contrarian Angle: The Tail That Could Wag the Dog
Despite the low probability of $200B, there is a scenario where Broadcom’s AI revenue surprises to the upside. The contrarian view is that the market underestimates the shift from training to inference. As large language models move into production, inference workloads will dominate compute demand. Custom ASICs can be 2-5x more energy-efficient than GPUs for inference. If AI adoption explodes in edge devices, robotics, and vertical industries, the total addressable market for inference chips could be much larger than today’s estimates. Broadcom, with its expertise in custom silicon and networking, is well-positioned to capture that growth.
Another factor is sovereign AI. Countries like India, Saudi Arabia, and those in Southeast Asia are building their own AI compute infrastructure. They want to avoid dependence on a single US supplier (NVIDIA). Broadcom offers a second source that can be customized for specific national needs. This could open up a new revenue stream worth tens of billions.
But even in these optimistic scenarios, the revenue is more likely to be in the $60-100 billion range by 2028, not $200 billion. The $200 billion figure is a “blue sky” scenario that requires everything to go perfectly: no supply chain disruptions, no regulatory hurdles, no slowdown in AI capex, and no competitive response from NVIDIA.
Trust is not a protocol, it is a practice. I have seen too many projects promise moonshots and deliver heartbreak. The 2022 Terra collapse taught me that the industry’s greatest vulnerability is not technical, but emotional. When a prediction like $200B circulates, it creates a FOMO cycle that can lead to irrational investment decisions. As a community founder, I have a responsibility to ground expectations in reality.
Takeaway: The Real Innovation Is Not in the Chip
What excites me about Broadcom’s AI journey is not the revenue number, but the underlying technology. Broadcom’s custom ASIC model is a step toward a more modular, decentralized compute infrastructure. Instead of one-size-fits-all GPUs, we are seeing chips optimized for specific workloads. This is analogous to the shift from monolithic blockchains to modular rollups. The future of AI compute is likely to be a mix of specialized silicon, orchestrated by open networking standards like Ultra Ethernet.
But the real innovation is not in the chip—it is in the community. The 2020 DeFi Summer taught me that trust is built through education, transparency, and empathy. When I translated 50 technical upgrade proposals into simple guides for Indian retail investors, I saw that people are not irrational; they are just uninformed. The same applies to AI hardware. The $200B prediction is a distraction. What matters is whether we can build an AI infrastructure that is resilient, accessible, and ethical.
Auditing the soul behind the smart contract: I apply the same lens to corporate predictions. The soul of Broadcom’s AI business is not its revenue targets; it is the ability to deliver value to real users, not just to Wall Street.
Final Reflection
I have been in this industry long enough to know that predictions are often wrong, but they shape narratives. The $200B narrative will likely boost Broadcom’s stock in the short term, but the real test will come in 2027-2028 when the physical constraints bite. The crypto market has its own version of this phenomenon: hype cycles driven by unrealistic projections. The antidote is the same: focus on fundamentals, community, and sustainable growth.
Digital artifacts that remember who we are: Our predictions should be grounded in the reality of silicon, power, and human trust. Let’s not get lost in the numbers. Let’s build the bridges that connect ambition to practice.
