A catalyst-heavy week, and the catalysts all printed in one direction. TSMC raised full-year 2026 revenue growth guidance to >30% YoY, ASML revised 2026 revenue up to €36-40bn (from €34-39bn) and flagged 2027 EUV supply shortage, and Anthropic’s ARR has increased from >$9bn at end-2025 to >$30bn in April. We think the through-line is simple: Agentic AI is real, it is converting into revenue faster than the street modeled three months ago, and the bottleneck is now squarely on the supply side — advanced nodes, fab equipment, memory, CPU, and the interconnect fabric around compute.
We used the week to publish a couple of substantive pieces of the year. The AI Infra 2026 deep dive lays out why networking — not FLOPS — is now the binding constraint for Agent workloads, and why Google’s scale-up dimensionality roadmap (3D torus → 4D+ → 6D+) plus DCN expansion will meaningfully reset OCS, Coherent Lite, and transceiver TAM over the next 2-3 years. The report became the most-read report in our history within just three days of publishing. The SpaceX Part 2 deep dive walks through Starlink unit economics, the Starshield government vertical, and a bottom-up SOTP against the $2T IPO figure being discussed.
Heading into next week, the earnings season kicks off, and Google Cloud Next (Apr 22-24) should be a catalyst for the TPU ecosystem thesis we laid out. We stay constructive on AI semis and hardware, and are watching software carefully as companies provide more color on AI impact during earnings.
Also, we are glad to be the inaugural guest of our friend TMTBreakout’s Slack AMA event. See you at 10:30am ET next Monday!
This Week’s Reports
AI Infra 2026: From “Brain Power” Competition to “Whole-Body” Evolution — Our deepest cut yet on TPU networking economics. We derive the scale-up OCS port ratio moving from 1.5:1 to 2:1~10:1 and scale-out from 0.2:1 to 1:1, implying a step-change in TAM for the optics supply chain. The TPU rack BOM analysis also shows TCO 2x+ better than GB200, freeing CSP CapEx for optics and memory.
SpaceX Deep Dive Part 2 — Our bottoms-up SOTP — Starlink at 10-15x P/S, Launch Services at 10-20x P/S, Starshield at 3-8x — gets to $992B Base / $1.49T Bull. The potentialrumored $2T valuation implies ~$500-1,000B of option premium (Starship platformization, xAI space compute, scarcity). The entire thesis hinges on one variable: Starship production cadence.
TSMC 1Q26 earnings - Clean beat-and-raise with full-year 2026 USD revenue growth raised to >30% YoY on Agentic AI demand. 2026 CapEx, at the high end of $52-56bn, bodes well for overall demand. We raise our earnings forecast accordingly.
ASML 1Q26 earnings— Our first ASML report in January flagged three trends that are now consensus: TSMC revising EUV up, resilient Chinese DUV demand (CXMT), and Intel/Samsung foundry recovery. The earnings validated our thesis, with 2026 revenue guidance revised up. We remain positive on the semicap upcycle.
