ASML just dropped Q2 2025 earnings: €9.33 billion in revenue, €2.92 billion in net profit. Beat. Beat. Beat.
The market is spinning it as a "surprise." I don't buy that. Not for a second. If you've been tracking the on-chain CapEx flows from TSMC, Samsung, and Intel—the three entities that essentially print ASML's quarterly reports before they're released—this was a foregone conclusion. The real story isn't the number itself. It's what drove the number.
For two years, the narrative was a chaotic cocktail: post-COVID inventory correction, geopolitical whiplash, and the crypto winter's chilling effect on GPU demand. Executives hedged every statement with "uncertainty." This quarter, the hedging stopped. The AI chip manufacturing CapEx wave has arrived, and ASML is the sole supplier of the shovels.
The Infrastructure Deconstruction
Let's cut through the official release. The raw data is clean: €9.33B revenue. €3.27B gross profit. Net bookings (new orders) were the critical metric—and they surged significantly quarter-over-quarter, driven overwhelmingly by Logic and Memory customers needing EUV lithography for sub-3nm nodes.
Why does this matter? Because ASML's order book is a 12-to-18-month forward indicator of leading-edge chip production. When TSMC orders a High-NA EUV system (the EXE:5200, at over €350M a pop), they aren't doing it for fun. They are locking in capacity for NVIDIA's Blackwell Ultra, AMD's MI400, and their own 2nm process.
The translation for the crypto native reader is brutal and direct: The bottleneck for the next generation of AI chips—which power the entire DePIN and decentralized compute thesis—is no longer just design. It is lithography. Every millisecond of latency we shave off a model's training cycle is paid for at the wafer level, by a machine that costs more than a small nation's GDP.
Avery's Note: Based on my years of tracking hardware supply chains, the most telling data point isn't the revenue beat itself. It's the mix. The percentage of EUV systems in the total system sales is at an all-time high. This means the "AI offset" narrative is backwards. The AI tailwind isn't just offsetting China uncertainty; it has become the primary propulsion system. China's DUV orders are now the hedged backup.
The Contrarian Angle: The Misreading of "China Uncertainty"
Every single major financial headline this morning frames the story as: "ASML beats despite China headwinds." That's lazy journalism. It's the equivalent of saying a Bitcoin bull run is "despite" a miner capitulation event. The two realities are partially decoupled.
Let's calibrate the risk. China currently represents approximately 20-30% of ASML's backlog. Those orders are for DUV systems (the NXT:1980i and below) used for mature node production (28nm and above). The key insight is that the margin profile on DUV is significantly lower than EUV.
So when you model the impact of a total export ban on DUV to China: - Scenario A (Total Ban): You lose ~25% of revenue but only ~15% of gross profit. - Scenario B (Gradual Escalation, my base case): The Dutch government, under pressure, will continue to allow service and spare parts but block new high-end DUV sales. This actually helps ASML's margin mix by shifting capacity to higher-value EUV lines.
The market is pricing in Scenario A's top-line risk. The on-chain data from TSMC and Samsung's April CapEx announcements suggests the reality is closer to Scenario B with a higher-for-longer AI spend. I don't think the "China risk" premium is warranted at current levels for a 2026 earnings outlook.
The Blind Spot No One Is Talking About
The euphoria around the AI cycle is creating a dangerous complacency regarding a specific operational risk: Power and water capacity for EUV fabs.
A single High-NA EUV scanner consumes roughly 1.5 megawatts of power. That's not the total fab draw—that's just the tool. TSMC's new Fab 21 in Arizona is being built with dedicated natural gas turbines just to power the lithography bay alone.
Based on my on-the-ground due diligence during the 2022 energy crisis, I saw firsthand how fab construction timelines were delayed by 6 to 9 months due to grid interconnection issues. The real bottleneck for ASML's growth isn't demand—it's the global infrastructure to host its products.
If the US, Europe, or Taiwan face another grid reliability shock (which we are highly likely to see in a prolonged bear market for base energy commodities), the CapEx deployment from TSMC will slow down. The orders will be signed. The deposit will be paid. But the tool will sit in a warehouse, not producing revenue.
This is the #1 tail risk to the forward guidance that the sell-side analysts are missing. They are modeling utilization rates at 95%. I'd be comfortable at 85% for the next 18 months given the physical constraints of building next-gen fabs.
How This Directly Impacts Your Crypto Portfolio
You're a crypto trader. You don't care about Taiwan's grid stability. You should.
The marginal cost of AI compute is set by the cost of the semiconductor manufacturing cycle. If ASML's tools are delayed for 6 months, the supply of next-gen H100/B200-equivalent chips gets crimped.
That means: - GPU cloud pricing will remain elevated. Good for RNDR, AKT, and any DePIN compute plays. - The price of used H100s will stay high. This props up the entire narrative for tokenized physical asset networks. - Capital rotation: If ASML's stock dips on a macro concern (like a rate hike), it's a buy. Because the structural AI demand curve is not broken. It's just being delayed by concrete and copper wire.
I don't say this lightly, but: Sell your layer-1s that have no AI thesis. Buy the ASML-correlated tokens. Not because ASML is "bullish." Because the fundamental asymmetry between demand and physical manufacturing capacity is the widest it has been since the 1999 fiber optic buildout.
Takeaway
Don't get distracted by the headline "beat." Watch the Key Signals: - Next week: TSMC's July revenue report. - August: The US BIS's proposed rulemaking on semiconductor equipment. - Q3: ASML's net bookings number.
If Q3 bookings hold above €6B, the AI cycle is not a bubble. It is an infrastructure buildout. If they dip, we'll see the first real test of the demand thesis. Until then, protect your liquidity, monitor the fab power outages, and remember: the fastest way to get return in this market is to own the shovels.
