One chart. Two tickers. $TSM and $ASML. The market has priced a binary outcome: either the earnings beat and AI keeps soaring, or they miss and the entire tech complex corrects 20%. That is the surface-level narrative. But the on-chain data—or rather, the deep structural data of the semiconductor supply chain—tells a different story.
Follow the gas, not the hype. The gas here is the physical capital expenditure flowing into advanced nodes and High-NA EUV tools. And the data shows a convergence of two monopolies that are not just reporting earnings, but are revealing the heartbeat of the entire AI civilization.
Context: The Duopoly of Bottlenecks
TSMC is the only foundry that can mass-produce 3nm chips with acceptable yield for AI training. ASML is the only supplier of the High-NA EUV lithography machines required to print those chips. Together, they form a dual chokehold on the entire AI compute stack. Their quarterly earnings are not mere financial updates; they are the most granular public dataset available for inferring the real health of AI demand, the pace of technological scaling, and the geopolitical risks embedded in the supply chain.
Core: The On-Chain Evidence Chain of Earnings
Let us deconstruct the key data points that will emerge from their reports, translating factory floor metrics into actionable on-chain signals.
1. Wafers Start Data: The True Leading Indicator
TSMC will not disclose exact wafer starts per node in its earnings call, but it will report revenue by process node (3nm, 5nm, 7nm) and end-market segments (HPC, smartphone, IoT). The ratio of HPC revenue to total wafer revenue is the closest proxy for AI compute demand. In Q2 2025, HPC already accounted for over 60% of TSMC's revenue, up from 40% two years ago. A sequential increase of 5% or more would signal accelerating AI buildout. A flat or declining ratio would be a red flag suggesting inventory digestion at hyperscalers.
But the deeper signal lies in the capital expenditure guidance. TSMC's CapEx is the single most important number for the entire semiconductor ecosystem. In 2024, TSMC spent ~$32B. The street expects $34-36B for 2025. If management raises the midpoint to $38B+, that is a direct statement: we see demand for 3nm and CoWoS continuing to outstrip supply for the next 24 months. If it cuts to $30B, that is a warning that either yields are too low to justify expansion or that customers are cancelling orders.
2. ASML's Order Book: The Future of Silicon
ASML reports quarterly net bookings of lithography systems. In Q1 2025, net bookings were €4.8B, but the critical number is the composition: how many of those are High-NA EUV systems? Each High-NA EUV machine costs ~€400M. If TSMC, Samsung, or Intel order more than 10 of these in a quarter, it means they are committing to 2nm and 1.6nm production at scale. If orders slow, it suggests the industry is hitting a cost wall.
During the 2021 bubble, ASML's inventory buildup preceded a 12-month correction in chip stocks. I audited the lead times of immersion DUV tools in 2022—delays of 18 months signaled demand overhang that later evaporated. History does not repeat, but it rhymes. In a bull market, investors extrapolate every ASML order as confirmation of infinite AI demand. The on-chain evidence: watch the ratio of pull-in requests versus push-out requests from customers. Anecdotal data from equipment supply chain checks will surface on earnings call Q&A. If TSMC asks to accelerate High-NA deliveries, bullish. If Intel postpones, beware.
3. Gross Margin as a Yield Proxy
TSMC's gross margin has been hovering around 53-55%. If it rises above 57%, it implies 3nm yields have improved significantly, lowering unit costs and boosting profit per wafer. If it drops below 50%, it signals either aggressive pricing pressure from Nvidia/AMD or yield issues on N3E or N2 test runs. In my forensic analysis of the 2020 DeFi yield aggregation, I learned that margins tell the truth when narratives lie. A high gross margin with flat CapEx indicates pricing power without capacity expansion—a short-term cash cow but a long-term constraint. A lower margin with rising CapEx indicates investment for future dominance.
4. The Geopolitical Tail Risk Embedded in Geographic Revenue
Both companies will break down revenue by geography. TSMC's China revenue share has dropped from 22% in 2020 to barely 10% in 2024 due to export controls. ASML's China revenue spiked to 49% in Q1 2025 as Chinese customers scrambled to buy older DUV machines before tighter restrictions. This is a dangerous dichotomy. ASML's China boom is a front-loaded pull-in that will reverse in 2026. TSMC's China decline means it loses a growth engine. The market may cheer ASML's current numbers but ignore the cliff ahead.
Contrarian: Correlation ≠ Causation—Why These Earnings Don't Determine AI's Fate
The prevailing narrative: TSMC and ASML earnings are the ultimate judge of AI market health. I disagree. These companies are lagging indicators of investment decisions made 18 months ago. The capital equipment orders they report today were placed when sentiment was even more exuberant. A miss in orders would not reflect today's AI demand but a correction of past over-optimism. A beat would just confirm what we already knew from on-chain GPU utilization data: that compute demand remains high.
Moreover, the bull market euphoria masks a critical technical flaw: the semiconductor industry operates on a 3-5 year investment cycle, while crypto/AI narrative cycles are measured in months. If Nvidia's next-generation Rubin architecture launches in 2026, the wafer starts for that chip are being negotiated now. A single earnings miss from TSMC in Q3 2025 will not change the 2026 trajectory. It will only cause a 10% dip that smart money will buy.
The real blind spot is financial engineering. Both TSMC and ASML have become proxies for passive ETF flows. When the market sells off in June 2025, it was not because of fundamentals; it was because macro fears triggered a rebalancing of risk parity funds. The correlation between these stocks and the broader market has risen above 0.8. That means their earnings are now a reflection of monetary policy expectations, not just chip demand.
Takeaway: The Signal to Watch Next Week
Ignore the headline EPS beats or misses. Focus on two numbers: TSMC's 2025 CapEx guidance midpoint, and ASML's net bookings of High-NA EUV systems. If TSMC guides above $36B and ASML reports more than 12 High-NA bookings, the AI infrastructure buildout is accelerating, and every dip is a buying opportunity. If TSMC reduces CapEx and ASML highlights delivery delays, then the physical bottleneck has shifted from coherency to capital discipline. In that case, rotate into software and away from picks-and-shovels hardware.
Code is law; logic is leverage. The chain of causality in this sector is not linear. The data must be read against the grain of the consensus. Will the whales be accumulating TSMC into the earnings dip, or are they front-running a CapEx cut? The on-chain truth does not sleep—and neither should your analysis.
Whales don't care about your feelings. They care about the next two numbers in the earnings release. Watch them, and you will see the future before the headlines arrive.