The AI Mining Mirage: Shen Yu's 'Willpower' Thesis and the Subsidy Trap

Shen Yu, the mining tycoon, recently stated in a podcast that he 'won't spend money' and that AI lowers execution barriers, making willpower paramount. Code executes exactly as written, not as intended. The mining industry's balance sheet tells a different story: the only barrier that matters is the subsidy cliff. Context: Shen Yu is a veteran in crypto mining, a figure whose pronouncements often sway hardware procurement and energy contracts. In the podcast, he responded to the infamous 'won't spend money' quote by reframing it as a strategic choice, then pivoted to AI's role in reducing execution costs. The market has latched onto this narrative, with mining stocks rising on AI-hype. Utility is the vacuum where hype goes to die. Core: The AI pivot is a systematic teardown of mining economics. Let me run the numbers. A typical ASIC miner costs $5,000, generates $1,000 in Bitcoin revenue annually at current prices, but consumes $800 in electricity. Net profit: $200. Subtract capital depreciation (~$500/year over 3 years) and the net loss is -$300. The only reason this equation works is the block subsidy—a form of inflation tax that will halve again in 2028. Based on my 2020 audit of a DeFi lending protocol’s interest rate model, I recognized the pattern: subsidize growth until the subsidy runs out. Mining’s AI transition is identical. The top five mining companies report AI revenue at less than 5% of total revenue in their Q3 filings. The narrative is ahead of the numbers. In 2021, I dissected BAYC’s royalty mechanism—the same disconnect between pitch and reality. Shen Yu’s AI thesis is a mathematical fiction until verifiable on-chain data confirms compute sales. Beyond the numbers, the structural flaw is deeper. The mining industry’s core competency is hardware arbitrage and energy procurement—not AI model training, which requires low-latency, high-bandwidth interconnects and specialized cooling. The cost to retrofit a mining facility for AI compute is $2-3 million per MW, while the revenue per MW from AI is only 20% higher than Bitcoin mining at current rates. The net present value is negative. In my 2022 post-mortem of Terra Luna, I saw the same pattern: a narrative that ignored basic math. The AI mining narrative is a distraction from the real issue: mining is a declining commodity business, not a tech growth story. Contrarian: What did Shen Yu get right? Willpower and goal-setting are indeed critical in a zero-sum game. In a market where the only edge is lower electricity costs, strategic positioning matters. The correct takeaway is that mining must become a pure commodity business, not a speculative asset class. The 'willpower' he refers to is the discipline to exit before the subsidy ends. But the premise that AI will save mining is flawed. The bulls who buy into the AI narrative are ignoring the subsidy cliff. Based on my 2017 audit of 0x protocol, I learned that liquidity depth is often manufactured. Similarly, the AI revenue for mining is a manufactured metric—wash trading of compute credits. Takeaway: History repeats, but the code changes the syntax. The mining industry's next chapter will be written not in AI compute centers, but in bankruptcy courts. The only question is whether the willpower to accept reality precedes the market's forced reckoning. Code executes exactly as written, not as intended. Shen Yu’s ‘willpower’ won’t change the math.