Tessara Pro members received the full note a day ago, ahead of today’s TSMC print.
ASML gave us one conservative read and one important validation yesterday.
And today: we think TSMC could raise its full-year capex guidance from $52–56B to $56–60B. TSMC reports its quarterly earnings today (Thursday) at ~2am ET
AI is eating TSMC
ASML: the grade
Revenue call: €8.9B Tessara’s forecast versus €9.3B reported, at a 54.0% gross margin.
We were somewhat conservative: system sales came in roughly in line at €6.6B, and the upside came from €2.8B of service and upgrade revenue, which our model does not yet separately forecast. The graded result goes on our public track record either way.
EUV reconstruction: ASML stopped publishing EUV shipments in 2024, so we reconstructed 2026 output ourselves: approximately 69 systems including High-NA, frozen July 14. ASML now puts planned 2026 Low-NA capacity at approximately 65 systems; on the same basis, our estimate is 65–67. Capacity and shipments are not identical measures, but the company’s figure materially validates the reconstruction.
VIEW OUR ASML EARNINGS REPORT → https://tessara.tech/earnings/asml
Key takeaways from ASML’s reporting
Guidance: Full-year outlook rose from €36–40B to €43–45B, a €6B midpoint increase.
Back-end loading: The new range implies approximately €14.4B in Q4, 55% above this quarter.
Shipment timing: EUV units remained flat at 16 systems while capacity expands 30% for 2027. The wave is scheduled, not yet shipped.
Geography: Taiwan rose from 23% to 30% of system sales while China fell to 14%, directionally consistent with TSMC expanding leading-edge capacity.
Our TSMC call
Before the print, we believe TSMC could raise its FY2026 capex to $56–60B, a 8% increase from current guidance: $52–56B). This would be a strong read for the AI infrastructure cycle.
Why we think so:
Revenue is running 31% year over year against guidance set in April, and TSMC’s recent spending trajectory, $29.8B in 2024, $40.9B in 2025 and $41.9B over the trailing 12 months, provides a roughly $58B anchor for our range.
Tight bottlenecks: CoWoS, HBM3E, N3 and N2 all remain tight on our constraint map. When leading-edge wafer and packaging constraints remain tight together, additional spending becomes the most likely response.
Committed demand: We identified 22 credible forward GPU deployment signals in the past 90 days, and five hyperscalers are guiding capex higher.
(Tessara’s model is proprietary, so we explain our reasoning without disclosing the underlying calculations or methodology)
We could be wrong on the timing. It’s quite possible TSMC waits until October to raise guidance, this specific forecast will be wrong, even if the broader thesis proves correct.
Why this number matters: TSMC commits capacity before most downstream suppliers recognize the resulting revenue, making its capex guide one of the earliest reads on the AI infrastructure cycle.
We are making no differentiated revenue call. TSMC’s disclosed April-to-June revenue already totals NT$1.27T, implying approximately $39.7B, effectively in line with consensus. There is no edge in forecasting a number that has already been substantially disclosed.
VIEW OUR TSMC EARNINGS REPORT → https://tessara.tech/earnings/tsm
ASML and TSMC sit at the front of the AI supply chain. Once both have reported, we’ll have one of the clearest views yet of how the AI buildout is progressing.
Teng