Onto Innovation Is the AI Chip Buildout’s Yield Layer

The AI infrastructure trade has been loudest around GPUs, custom accelerators, networking, and power. Onto Innovation is quieter because it sits one layer below the ticker drama. The company does not sell the AI chip. It sells the process-control stack that helps the chip and its package become manufacturable.

That distinction matters. The next bottleneck in AI silicon is not only transistor density or accelerator design. It is yield across advanced packaging, high-bandwidth memory, gate-all-around logic, and increasingly complicated materials. If the package is the system, measurement becomes infrastructure.

That is why ONTO belongs in the same AI-infrastructure conversation as the more obvious chip names, even if it trades like a semiconductor equipment stock. As of the Friday, June 5, 2026 regular close, ONTO was near $253, down about 7.8% on the day. TECHi’s ONTO quote dashboard shows a Buy analyst view but a more restrained forward model: positive, selective, and valuation-sensitive. That is the right framing. This is a real AI supply-chain asset, but the multiple already asks investors to be precise.

What Onto actually sells

Onto Innovation’s language can sound technical because the business is technical. The simpler version is this: Onto helps chipmakers and advanced packaging houses see whether their wafers, interconnects, materials, layers, and packages are good enough before bad yield turns into expensive scrap.

The company describes itself as a process-control business spanning unpatterned wafer quality, 3D metrology, macro defect inspection, metal interconnect composition, factory analytics, and lithography for advanced semiconductor packaging. That puts Onto between design intent and manufacturing reality.

That is also why the EDA label only partly fits. Onto is not Synopsys or Cadence. It is not where engineers draw the chip. It is where the fab and packaging line discover whether the chip, interposer, memory stack, or interconnect structure can be built repeatedly at commercial yield. In advanced AI silicon, that feedback loop is becoming as strategic as design software itself.

The quarter gave the thesis more evidence

Onto’s first-quarter 2026 results were not a vague AI story. Revenue was $291.9 million, up 9.5% year over year, with non-GAAP gross margin of 55.7% and non-GAAP operating income of $77.9 million. The company ended the quarter with $654 million of cash and short-term investments.

The more important detail was where the demand showed up. Onto said its Dragonfly G5 system was qualified at a leading 2.5D logic customer and a high-bandwidth memory customer. It also said its Atlas G6 system was selected by a second logic customer for gate-all-around metrology, helping advanced nodes grow 13% in the quarter and positioning that business for about 25% growth for the full year.

That is the AI-chip read-through. HBM and 2.5D packaging are not side stories anymore. They are where AI accelerator economics are increasingly decided. The GPU can be designed correctly and still lose commercial power if the memory stack, package, interconnect, thermal behavior, or defect profile does not scale.

Onto’s own second-quarter guide sharpened the setup. Management guided for revenue of $320 million to $330 million, gross margin of 56% to 56.5%, and non-GAAP EPS of $1.65 to $1.73. That is a step-up quarter, not a maintenance quarter.

Rigaku is the strategic swing

The clearest sign that Onto is trying to move deeper into next-generation process control is the Rigaku partnership. In April, Onto announced a strategic partnership with Rigaku to combine Onto’s Ai Diffract analysis software and optical critical dimension metrology with Rigaku’s X-ray platforms.

The customer problem is straightforward: as logic and memory structures get deeper and more three-dimensional, optical tools alone do not see everything. X-ray can help measure deeper structures, while optical metrology can add speed and location context. The value is not just another tool sale. It is correlated measurement across different physics.

Onto said the new offering had already been selected by two key customers and that external analysts estimate the addressable market at more than $1 billion within five years. The company also agreed to buy 27% of Rigaku for about $710 million, with the investment expected to be accretive as of December 31, 2026, assuming the transaction closes as planned.

That is a meaningful capital decision. It turns a product collaboration into a strategic alignment move, and it shows Onto wants to own more of the process-control answer for advanced logic and memory.

Semilab makes the portfolio broader

Rigaku is not the only portfolio move. Onto completed the acquisition of selected Semilab product lines in late 2025, adding materials composition and electrical analysis capabilities. In that Semilab closing release, Onto said the acquired product lines were expected to contribute about $120 million of revenue in 2026, likely weighted toward the second half.

That matters because modern process control is not one measurement. It is a portfolio of measurements. Advanced packages need inspection at macro scale and detail at tiny geometry. Gate-all-around devices add new materials and structures. Memory stacks add vertical complexity. Factory analytics needs cleaner data across tools.

Semilab gives Onto more ways to measure the electrical and materials side of that problem. Rigaku gives it a deeper X-ray path. Dragonfly and Atlas are already tied to HBM, 2.5D logic, and gate-all-around wins. The pieces line up with where AI silicon is getting harder.

The balance-sheet move is not free

Onto also used the convertible market. In May, the company priced an upsized $1.3 billion 0.00% convertible senior notes offering, with an initial conversion price of about $381.80 per share and capped call transactions with an initial cap price of $509.06. Net proceeds were expected to be about $1.274 billion before expenses, with part of the proceeds used for capped calls, about $205 million for share repurchases, and the remainder available for general corporate purposes, including financing the Rigaku stake.

That financing is rational. A zero-coupon convertible can be attractive when a company wants strategic flexibility without a conventional cash interest burden. But it still changes the story. Investors now have to watch not just revenue growth and margin, but also dilution mechanics, hedge activity, the Rigaku close, and whether the acquired or partnered assets generate enough growth to justify the complexity.

This is the part of the ONTO thesis that should keep bulls disciplined. The company has a stronger strategic position than a screen for semiconductor equipment might show, but the stock is not priced like a cyclical small-cap equipment name. TECHi’s quote-page valuation context already shows a rich price-to-sales and price-to-book profile. If AI packaging orders slip, the market will not forgive the multiple just because the long-term theme is real.

That puts Onto alongside the names TECHi has been tracking across the wider AI infrastructure war: Nvidia’s AI factory quarter, Broadcom’s AI ASIC bar, and the custom-silicon cycle. Onto’s difference is location: it sits closer to yield than to accelerator sales.

The investor question

The best case for Onto is that AI silicon keeps moving from chip-level competition to system-level manufacturing. In that world, HBM stacks, chiplet packages, advanced substrates, gate-all-around devices, and factory analytics all need more inspection and metrology per dollar of compute capacity.

The bear case is not that Onto lacks relevance. It clearly has relevance. The bear case is that relevance gets capitalized too early. Semiconductor equipment demand still moves in cycles. A pause in advanced packaging capex, a slower HBM digestion period, a delayed Rigaku close, or integration friction from Semilab would all hit a stock trading on the assumption that process control is becoming more valuable.

That leaves ONTO as a watchlist name with a clear trigger set. The company needs to convert the Q2 guide, show that Dragonfly and Atlas demand is durable, prove that Semilab can add the expected 2026 revenue without damaging margins, and turn the Rigaku alignment into customer wins rather than just strategic language.

If that happens, Onto Innovation deserves a bigger place in the AI infrastructure map. It is not the chip. It is the measurement layer that decides whether the chip can be built well enough, often enough, and cheaply enough to matter.