Every AI chip today is really several chips stacked and stitched together. Each layer has to be checked for microscopic defects before the next goes on, because one bad layer can ruin the whole expensive package. Onto builds the optical machines that do the checking. As packages get more complex, the number of checks rises faster than the number of chips.
Source: Onto Innovation Form 8-K, Exhibit 99.1, 6 Aug 2026, and Q2 2026 conference call; Business Wire, 18 May 2026.
Why I chose it
1. The niche the giants are not built around
Onto's Dragonfly inspection platform and its packaging lithography line target the back end of the process: 2.5D logic packaging, HBM stacks, hybrid-bonding process windows and panel-level packaging. KLA, Applied Materials and Lam Research concentrate their scale on front-end wafer process control, deposition and etch. KLA also sells into packaging, and its own outlook for that business confirms the bottleneck; Onto's edge is focus and breadth inside the niche rather than size.
In plain termsThe big toolmakers sell to the whole factory. Onto specialises in the step where chips are stacked, which is where AI demand is growing fastest.
2. Demand visible in the order book
Backlog passed $1 billion in Q2 2026, and management describes visibility extending into 2027. The second-half 2026 revenue growth framework was raised to 25%+ from 15%, and the Q3 revenue guide of $380 to $400 million sat 8% to 14% above pre-results consensus of $350.5 million. Both reporting lines, Advanced Nodes and Specialty Devices & Advanced Packaging, posted record revenue.
In plain termsCustomers have already ordered more than a billion dollars of equipment that has not yet been delivered.
3. Operating leverage, read on both bases
Non-GAAP gross margin reached 57.0% (+250bps year on year) and non-GAAP operating margin 30.0% (+410bps), with R&D growing 10% against revenue growth of 35%. On a GAAP basis the operating margin was 18.5%; the gap is mostly amortisation from past acquisitions and will persist. I anchor my valuation on GAAP earnings power, not the adjusted figure.
In plain termsSales are growing much faster than costs, so each extra dollar of revenue is more profitable. I use the stricter accounting measure when I value it.
4. The capital structure is the differentiated angle
In May 2026 Onto issued $1.3 billion of 0.00% convertible notes due 2031, with a conversion price of about $381.80 and capped calls that limit dilution up to about $509.06. The proceeds funded a $205 million buyback and the pending ~$710 million purchase of a 27% stake in Rigaku, a Japanese X-ray metrology leader. At 30 June 2026 cash and investments were $1.88 billion against $1.47 billion of notes. Near-free financing is extending the franchise on two fronts at once.
In plain termsOnto borrowed $1.3 billion at zero interest. It repays in 2031, or the debt turns into shares only if the stock rises well above today's level.
What would break it
Each signal below is observable in public data. Two or more appearing together would lead me to close the idea and say so on this page.
| Signal | What I would see | Where it shows first |
|---|---|---|
| Backlog stops converting | Sequential revenue decline, or backlog falling, across two consecutive quarters. The company does not yet disclose book-to-bill, so backlog alone can hide slowing orders. | Quarterly results and call |
| Margin cadence fails | Non-GAAP gross margin misses management's framework of at least 50bps of sequential expansion in Q3 or Q4 2026 (Q3 guide 57.3% to 57.8%). | Q3 2026 results, early Nov 2026 |
| A key customer pauses | Demand is concentrated among a handful of foundry, memory and logic customers [four customers were 57.3% of H1 2026 revenue: verify against 10-Q before publishing]. One capex pause would hit disproportionately. | 10-Q concentration note; customer capex guides |
| Rigaku stake fails to close | Regulatory or execution delay beyond H2 2026 on the 27% stake, removing the hybrid X-ray and optical metrology adjacency. | Company announcements |
| Tariff escalation | Section 301-style tariffs on chips processed at overseas packaging and test houses and reimported to the US, compressing customer margins and packaging capex. | USTR actions; OSAT capex |
| The bar keeps rising | At about 33x forward non-GAAP earnings in August, an in-line quarter is not enough. On 6 Aug 2026 peer Nova fell 8.4% after beating estimates. | Price reaction to results |
Known noise, not a thesis breaker: hedging flows around the convertible notes have already produced one pullback of about 15% this year unrelated to fundamentals. Expect more volatility than the business alone would explain.
How I value it
Dated checkpoints
Glossary
- Metrology
- Precise measurement of film thickness, feature size and alignment on a wafer or package.
- Optical inspection
- Using light-based imaging to find defects such as particles, cracks or misaligned bumps.
- Advanced packaging
- Joining several chips (chiplets) into one package so they behave like a single, larger chip.
- 2.5D / 3D packaging
- Chips placed side by side on a shared base (2.5D) or stacked on top of one another (3D).
- HBM (High-Bandwidth Memory)
- Stacked memory dies placed next to an AI processor to feed it data quickly.
- Backlog
- Orders received but not yet shipped and recognised as revenue.
- Book-to-bill
- New orders divided by revenue in a period; above 1.0 means the order book is growing.
- GAAP / non-GAAP
- Reported accounting results versus company-adjusted figures that exclude items such as acquisition amortisation.
- Convertible note
- Debt that the holder can exchange for shares at a set conversion price.
- Capped call
- An option bought by the issuer that offsets share dilution from conversion, up to a cap price.
- OSAT
- Outsourced Semiconductor Assembly and Test: third-party firms that package and test chips.
- bps
- Basis points; 100bps equals one percentage point.