TSMC just posted a record Q2 2025 revenue of $40.2 billion. The markets cheered. The AI narrative is vindicated. But look deeper: that number is a red flag for every Proof-of-Work miner on the planet. The revenue surge is almost entirely driven by AI chip demand – NVIDIA, AMD, and the hyperscalers. Crypto mining ASICs? They are now an afterthought in TSMC's earnings reports. The math is simple: TSMC's advanced process capacity (3nm, 5nm) is finite. AI customers pay a premium, lock in multi-year contracts, and demand priority. Miners, by contrast, are volatile, price-sensitive, and low-margin. The result: more expensive chips, longer lead times, and a structural slowdown in mining hardware upgrades. This isn't a temporary squeeze; it's a permanent reordering of the semiconductor hierarchy. Miners are being relegated to second-class citizens in the foundry ecosystem. And they haven't yet priced in the consequences.
Let's establish the dependency chain. Every modern Bitcoin or Litecoin ASIC miner relies on TSMC or Samsung for its core chip. TSMC dominates with over 90% market share in advanced logic. For a miner, the path to profitability is simple: buy the latest-generation ASIC with higher efficiency (J/TH), deploy it, and earn more Bitcoin per watt. But this path now has a toll. The latest generation of mining chips (e.g., 3nm) competes directly with AI accelerators for wafer starts. TSMC's Q2 earnings call confirmed that HPC (high-performance computing, mostly AI) now accounts for over 50% of revenue. Crypto mining falls under "other" – a small, declining bucket. The narrative of "digital gold" doesn't impress TSMC's sales team. They care about long-term, high-margin contracts. Miners, historically, operate in boom-bust cycles. That uncertainty makes them unattractive partners for capacity allocation.
I have spent years auditing the fragility of decentralized systems. The Tezos formal verification debacle taught me that humans rarely verify the assumptions embedded in their infrastructure. Here, the assumption is that the semiconductor supply chain will always be elastic enough to support mining hardware upgrades. That assumption is now invalid.
Let's dissect the fragility systematically. First, the pricing signal. TSMC is raising wafer prices for advanced nodes by 10–20% next year, driven by AI demand. For a mining chip designer like Bitmain or MicroBT, that means the per-chip cost balloons. They pass that on to miners. The result: the breakeven price for a new-generation miner is pushed higher. In a bear market, that reduces ROI and discourages capital expenditure. Second, delivery timelines. AI customers book capacity years in advance. Miners order in batches when Bitcoin price spikes. That mismatch means when miners want to order, capacity is already sold out. Lead times stretch from 6 months to over a year. By the time chips arrive, the market may have turned. This is not a bug; it's a feature of the new semiconductor economy.
Consider the elasticity of mining hardware supply. In the past, when Bitcoin price surged, miners could quickly order new ASICs, and capacity would expand within quarters. That elasticity is now gone. TSMC's capacity is pre-allocated to AI clients with long-term contracts. The notional supply curve for new-generation mining gear has become steeply inelastic. This means that any spike in mining profitability will be quickly arbitraged away by hardware scarcity rather than by hash rate growth. The market mechanism that traditionally restored equilibrium – more miners joining the network – is now impaired.
The math holds, but the humans did not verify it. They assumed that foundries would always serve crypto mining as a priority segment. They did not account for a 500-pound gorilla called AI.
Second-order effects compound the problem. Used mining rigs may appreciate as new rigs become scarce and expensive. That creates a paradox: the CAPEX barrier rises, but the OPEX for efficient rigs may stay competitive. The industry concentrates among large players with long-term contracts. Small miners get squeezed out. Network hash rate growth slows, which could lead to lower difficulty adjustments, extending the life of older gear. But that also means the overall security budget (hash rate) grows slower, potentially making the network more vulnerable to a sustained attack if the price stays low.
Let's run a simple model. Assume TSMC allocates 10% of its 3nm capacity to crypto mining in 2024. In 2025, that drops to 5% due to AI demand. The resulting shortage of new ASICs means the existing fleet must be upgraded more slowly. The network's total energy efficiency improves at a decremental rate. Meanwhile, electricity costs remain high. The net effect is a squeeze on miner margins, especially for those using mid-generation gear.
The bulls argue that miners are resilient and adaptive. They point to the rise of "mining-as-a-service" and the pivot to AI compute. Some are already buying NVIDIA GPUs and renting out compute. This is valid – the nimble operators can diversify. Also, the used mining rig market provides a buffer; older generation miners (7nm, 5nm) become more valuable as new ones are harder to get. Furthermore, the Bitcoin network's security isn't immediately threatened; existing hash rate may stabilize or even grow due to efficiency gains. Value is consensus; truth is optional. The contrarian truth: TSMC's constraint may actually prolong the profitability of older, already-deployed mining hardware, as the pace of obsolescence slows. Miners who already own efficient rigs benefit from reduced competition in new hardware. The "AI squeeze" narrative is real, but it doesn't mean the end of PoW – it means a shift in the competitive landscape.
However, this adaptive narrative has limits. AI compute requires different hardware (GPUs, DPUs) and different customers. Not every miner can pivot. The transition from PoW to AI compute requires capital, expertise, and sales channels. Most small miners will not make that leap. They will be forced to either buy expensive new ASICs or ride old ones into obsolescence. The concentration of mining power among large, institutional players will accelerate.
From my experience analyzing the 2020 Compound liquidity risk, I learned that market efficiency is an illusion during capital influx. Here, the capital influx is into AI, not crypto mining. The market's efficiency in allocating capacity away from mining is a structural shift, not a cyclical one.
Let's also consider the geopolitical layer. TSMC's advanced chip exports are subject to US export controls, especially regarding China. If the US further tightens restrictions on mining chip exports, the supply chain could be severed entirely for certain jurisdictions. That would create a two-tier mining world: one with access to cutting-edge hardware, one without. The regulatory risk amplifies the structural risk.
The takeaway is stark: the next time you see a Bitcoin mining stock report, ask about their supply chain. Not just electricity cost. The real bottleneck is silicon. The AI era has redrawn the semiconductor map, and miners are now at the bottom of the priority list. The assumption that new hardware will always be available at a reasonable price? Assumptions are just risks wearing disguises. Verify the foundry allocation, not the hashpower projection.
This is not the death of PoW, but it is the end of the era where mining hardware upgrades were a sure bet. The industry is maturing, and maturity brings discipline – and higher barriers. The next bull run will test whether miners can adapt to a world where silicon is scarce. I will be watching the wafer allocation reports, not the mining pool hash rates.