They just proved they don't need memory capex to hit 53% gross margins.
Thesis: Street is mispricing ONTO as a memory cycle play waiting for a DRAM recovery. Wrong. This is a bottleneck arbitrage on CoWoS yields. The 43% collapse in advanced nodes masked a structural mix-shift improvement to 53% gross margins. TSMC cannot ship Blackwell volumes without DragonFly inspection. When memory eventually turns, it adds torque to a significantly higher margin baseline, not a recovery to par.
Verdict: LONG — Conviction: HIGH
Catalyst: Shipment of the $90M AI packaging order ramping in Q4/Q1, proving the revenue replacement of legacy memory.
Key Risk: Yield normalization at TSMC reducing inspection intensity before volume expansion compensates.
The Tell: When pressed on the $90M split, Plisinski admitted '60/40... A lot of it is on the logic side.' This confirms extreme concentration on Nvidia/TSMC CoWoS lines rather than a broad-based memory recovery.
Friction Level: MODERATE_FRICTION — Street models a cyclical memory recovery; data shows a structural mix shift to high-margin AI packaging that decouples them from DRAM capex.
Report not found
The report data is no longer available. Please return to the archive.