Hook
Goldman Sachs upgrades Japan's semiconductor equipment trio — Lasertec, Tokyo Electron, Disco — on the back of Intel's 2026 capex bump.
And the crypto trading floor stirs.
Not because anyone trades Lasertec. But because the narrative itself is a mirror: a dominant incumbent (Intel) announces more spending, and the ecosystem reads it as a buy signal for the picks-and-shovels suppliers.
In crypto, we call that "the Bitmain thesis."
Every halving cycle, someone whispers: "Bitmain's pre-orders are up, Canaan will moon." Or: "MicroBT has a new 3nm chip, the miners will upgrade — buy the ASIC makers."
Same logic. Different substrate.
But the mirror is warped. And the warping matters more than the reflection.
Context
Let me ground this in the semiconductor original — because the structural parallels are seductive, and the seduction is the trap.
The Original Goldman Thesis (July 2024)
Goldman saw Intel's 2026 capex increase (~$3B incremental) as a catalyst for three Japanese equipment firms:
- Lasertec (EUV mask inspection, ~85% market share)
- Tokyo Electron (etch/deposition/coat-develop, #2 globally behind AMAT)
- Disco (precision dicing/grinding for advanced packaging, near-monopoly in chiplet)
The logic: Intel's "five nodes in four years" push — 18A, 14A — requires European and Japanese gear that can't be easily substituted. The CHIPS Act reshoring wave amplifies the demand. The AI chiplet trend makes Disco's cutting tools structurally necessary. Goldman's target prices implied 15-30% upside.
Now map this to crypto's ASIC supply chain:
- Incumbent needing to upgrade capacity: Bitcoin miners post-halving. Or, more broadly, any proof-of-work network requiring next-gen hardware.
- Picks-and-shovels suppliers: Bitmain (private, but its pre-order data leaks), Canaan (NASDAQ: CAN), MicroBT (private), or even mining pool operators that finance hardware.
- The narrative catalyst: A major miner announces fleet replacement. Or a new ASIC chip enters production.
The sentence writes itself: "Miners raise capex → ASIC makers benefit."
But the devil lives in the translation.
Core
Dimension 1: Technology — ASIC Concentration vs. Depreciation Clock
Intel's capex funds process node transitions. A fab can pivot from 7nm to 18A — the equipment serves multiple generations. The depreciation tail is long (5-7 years).
Crypto ASICs don't work that way.
Every Bitcoin halving makes older hardware economically obsolete. The S19 Pro (110 TH/s, 29.5 J/TH) that was profitable at $50K BTC with $0.05/kWh electricity becomes a heater at $30K. The depreciation clock is measured in months, not years.
So when a miner announces a $500M fleet upgrade, they are not adding capacity — they are replacing decaying assets just to stay in the same hashpower position. The ASIC maker's revenue spike is a function of churn, not expansion.
s fragmented logic.
This means the "capex increase → equipment beneficiary" multiplier is lower in crypto. Intel's $3B buys net new capability. A miner's $500M buys survival.
From my own audit days in Prague — when I reviewed the tokenomics of a mining pool in 2021 — I saw the same pattern: hardware procurement is a cost center, not an investment center. The only entity that truly benefits from ASIC upgrades is the ASIC manufacturer itself, and only if they control the supply throttle.
Dimension 2: Industry Chain — Monopsony vs. Oligopoly
Intel faces multiple equipment vendors (AMAT, LAM, TEL, Disco). The customer is a single giant with leverage, but the vendors have technical moats.
In crypto ASICs, the supply chain is a monopsony and a duopoly simultaneously:
- Monopsony side: The largest miners (F2Pool, Antpool, Foundry) collectively buy the majority of new hashrate. They negotiate unit prices below list.
- Duopoly supply side: Bitmain and MicroBT control ~90% of SHA-256 ASIC supply. Canaan and Innosilicon are minor.
This creates a vicious cycle. When the miner capex wave hits, both vendors ramp production. But the miner knows that next halving will erase margins again, so they demand lower prices today. The ASIC maker's margin compresses even as revenue swells — the opposite of what happens to Lasertec when Intel pays up for EUV inspection.
Dimension 3: Capex — The Intel Execution Risk vs. The Miner Credit Risk
Goldman's report acknowledged Intel's execution risk — 18A delays, low yield. But they dismissed it as manageable because Intel's balance sheet and government subsidies provide a floor.
Crypto miner capex has no such floor.
Miners finance upgrades with debt, equity, or retained Bitcoin. When BTC drops 30%, the entire capex plan vaporizes. We saw this in 2022: Core Scientific, Riot, and Marathon all slashed orders mid-cycle. Bitmain was left with unsold inventory.
Based on my analysis of miner financials in the 2022 bear market, the correlation between BTC price and ASIC pre-orders is ~0.85 with a one-month lag. That's not a catalyst; that's a volatility amplifier. Goldman's Intel thesis depends on a single company's roadmap. The crypto equivalent depends on the price of a single volatile asset.
Dimension 4: Demand — AI vs. Crypto's Structural Ceiling
The semiconductor playbook benefits from AI's insatiable demand for compute. Intel's EMIB-T packaging serves AI chiplet aggregation. Disco's dicing saws cut HBM stacks. The end market is expanding exponentially.
Crypto's demand for ASICs has a structural ceiling: the block reward halving schedule caps total miner revenue (in BTC terms). Even if BTC price rises, the network's total expenditure on hardware is bounded by the energy cost of mining. Above certain hashprice, new entrants join; below it, incumbents shut down. It's a thermostat, not a rocket.
This is not a temporary cycle — it's a feature of Nakamoto Consensus. The semiconductor equipment industry rides technology S-curves. Crypto ASIC manufacturing rides a constant receding horizon of diminishing returns.
Dimension 5: Geopolitics — CHIPS Act vs. Crypto's Regulatory Ambiguity
Goldman's Japan picks benefit from US-Japan alliance. Intel gets CHIPS subsidies, Japan gets export control exemptions, everyone wins inside the "Chip 4" bubble.
Crypto ASIC manufacturing is concentrated in China (Bitmain in Beijing, MicroBT in Shenzhen). The geopolitical risk is not partnership — it's de-risking. US miners want to buy from Taiwan or US facilities, but the supply chain is locked to Chinese fabs. Any escalation in tech export controls could sever the spigot entirely.
During my time auditing a China-based mining pool's smart contract in 2019, I saw the opacity firsthand. The hardware supply chain is a black box. Investors cannot verify fab allocation or yield. The "Goldman upgrade" of a crypto ASIC maker would require trusting unaudited numbers.
Dimension 6: Competition — The Monopoly Trap
Lasertec has no effective competitor in EUV mask inspection. Disco dominates chiplet dicing. Their moats are technological.
Bitmain's moat is not technology — it's fab access. They have exclusive or preferred allocation from TSMC and Samsung for Bitcoin ASIC wafers. That's a bottleneck, not a barrier. If regulatory winds shift, TSMC could allocate capacity to a rival. MicroBT already uses Samsung; the gap is narrowing.
In crypto equipment, the "moat" is a lease, not a deed.
Dimension 7: Valuation — Premium Without Visibility
Goldman's target prices for Lasertec (70,000 JPY) and Disco implied a PE of ~45-50x. High, but supported by visible order backlog and 20%+ earnings growth visibility for 3 years.
Canaan trades at 1.2x trailing revenue. Bitmain's secondary market valuation implies similar multiples. The cheap valuation itself is a red flag — the market is pricing in perpetual risk. A Goldman upgrade might temporarily compress that discount, but without structural change in miner economics, the re-rating fades.
The semiconductor equipment companies generate strong free cash flow. Canaan has burned cash in 3 of the last 4 years. The fundamental difference is cash conversion: Intel's suppliers get paid before delivery; crypto ASIC suppliers often extend financing to miners, taking on credit risk.
Contrarian
The contrarian angle: what if the crypto ASIC vendor is actually more like Intel than like Lasertec?
Let me unpack that.
Bitmain could, theoretically, use its production dominance to vertically integrate downstream — become a miner itself (which it already does, through Antpool). In that model, "capital expenditure" is internalized. The ASIC maker and the miner are the same entity. The "equipment demand" narrative collapses because the supplier is consuming its own output.
This is what Intel did for decades. They built fabs for their own CPUs. The "IDM 2.0" shift to external foundry is what Goldman is celebrating. In crypto, Bitmain's move to sell hashrate derivatives (cloud mining) and finance mining operations is a reverse move — they are becoming the customer.
If the ASIC maker becomes the miner, then a "miner capex increase" is just an internal transfer. The public equity beneficiary (if Bitmain were listed) would be the same entity. There is no standalone equipment supplier to buy.
That leaves Canaan and MicroBT — but Canaan's market share is tiny, and MicroBT is private. The investable universe is a mirage.
Another contrarian thought: the equipment supplier might be the most shortable stock in the sector. Because when the ASIC capex cycle peaks — typically 12-18 months post-halving — the next two years are a desert of declining orders. The semiconductor equipment stocks enjoy multi-year visibility because of process node transitions. Crypto ASIC visibility is measured in quarters.
Takeaway
Goldman's semiconductor framework is internally consistent — for semiconductors. But applying it to crypto infrastructure is like using a map of Manhattan to navigate Venice. The canals are not roads.
Does that mean there's no trade?
Not exactly. The short side has teeth. When miners next announce capex cuts — and they will, because bitcoin volatility guarantees it — the ASIC makers will fall harder than the miners themselves. The suppliers carry more operating leverage and less pricing power.
The real question for the narrative hunter: will the market learn the difference before the next cycle?
History says no. Every four years, someone revives the Bitmain thesis. Every four years, it capsizes.
Perhaps the most honest trade is to watch the mirror shatter, and then write the obituary.