PDF Solutions unveils Exensio Aurora on a 25x speed target of its own, with no named customers and no outside benchmark

by Thandi Mbeki · September 10, 2026

AI Companies Reporter, Frontier Tech Wire

PDF Solutions, the semiconductor manufacturing analytics company listed on Nasdaq under PDFS, announced a re-architected version of its flagship platform on Sept. 9, 2026. Exensio Aurora, according to the company's release, is a highly scalable analytics architecture designed to handle semiconductor manufacturing data at petabyte scale and to deploy agentic AI across manufacturing operations and supply chains.

The headline performance number in the announcement is carefully worded, and the wording matters more than the number. The release says the new Exensio Aurora distributed engine "is designed to deliver approximately 25X faster performance at comparable hardware cost." That is a stated design target, not a reported measurement. The release does not say the figure was measured, does not describe a test, does not identify in the announcement itself the competing system or prior Exensio release the comparison is drawn against, and cites no third-party benchmark or independent laboratory. Anyone quoting 25x should quote it as the company's own design goal.

Coverage ahead of the formal announcement used the same framing. Writing on SemiWiki on Aug. 18, Kalar Rajendiran reported that PDF Solutions "cites roughly 25x faster performance at comparable hardware cost for large data sets" - again the company citing its own figure, relayed rather than tested. Rajendiran offers no independent assessment of the number, and neither document points to a benchmark run by anyone other than PDF Solutions.

A second claim in the release is scale rather than speed, and it is phrased in exactly the same conditional register: data size, the company says, "is designed to be limited only by the analytics cluster, and performance is designed to scale linearly as volumes grow, in width or in length." "Designed to" is engineering intent, and it governs both the scaling curve and the 25x figure. Neither is offered in the release as a result the company has measured and is now reporting.

Technically, Aurora is described as a distributed analytics engine built on Kubernetes. The release says it "is designed to enable central training with deployment to multiple edge locations like outsourced assembly and test houses (OSATs), test floors, and edge boxes." Its named components are scalable analytics, a Manufacturing Data House, model lifecycle management, agentic AI, AI-first interactivity and workflows, with no-code creation of new analysis types and a natural-language interface among the described capabilities. The release also cites enhanced support for 3D and higher-dimensional data.

On availability, the release is clear and appropriately modest: Exensio Aurora "will first be available in September 2026 as part of a beta release program for a small number of early adopters." This is not a generally available product, and no customer names are disclosed. Both the release and SemiWiki point to PDF Solutions CONNECT 2026 in San Francisco on Oct. 15 and 16; SemiWiki says the first public demonstrations of Aurora will take place there. That event is still more than a month away.

John Kibarian, PDF Solutions' chief executive, president and co-founder, said in the release that delivering the operational efficiency the industry needs "requires a new kind of AI-driven collaboration." Said Akar, vice president of enterprise development, said that "AI is the foundational guiding principle for the new Exensio Aurora architecture."

The market problem Aurora is aimed at is real and independently observable. As Rajendiran puts it on SemiWiki, "the semiconductor industry doesn't have a shortage of data. It has a shortage of ways to turn that data into useful manufacturing insights quickly enough to matter." He adds that advanced packaging, globally distributed manufacturing and growing product complexity "are generating enormous amounts of increasingly multidimensional data."

For scale, PDF Solutions is a genuinely small company relative to the fabs it sells into. In results reported on Aug. 6, 2026, it posted second-quarter revenue of $61.5 million, up 19 percent from $51.7 million a year earlier, split between $49.1 million of platform revenue and $12.4 million of volume-based revenue. GAAP net income was $4.3 million, or $0.10 per diluted share; on a non-GAAP basis the company reported $11.4 million, or $0.27 per diluted share.

Margins and balance sheet from the same release: GAAP gross margin of 69 percent and non-GAAP gross margin of 73 percent, cash and equivalents of $114.9 million as of June 30, 2026, and ending backlog of $270.7 million. The company reaffirmed a 20 percent annual revenue growth target for 2026. That target was set before Aurora was announced and does not depend on it.

Kibarian's commentary alongside those results pointed to breadth rather than a single product: "In the quarter, we had major customer wins for our secureWISE system, for our DirectScan system, and for our Exensio product and services." Exensio is one of three named lines, which is worth holding in mind when weighing how much a re-architecture of it changes the whole business.

Context on the tape: the Aurora announcement landed on a soft session for small caps. The Russell 2000 ETF dropped 1.36 percent on Wednesday, Sept. 9, according to 404K Research's Sept. 10 morning brief, matching the index itself, which closed that session at 2,919.86, down 1.36 percent. The U.S. market was open as this piece was filed at approximately 11:20 a.m. Eastern on Thursday, Sept. 10, so no closing level exists for the current session.

What would convert Aurora from an announcement into a result is straightforward and checkable: named beta customers, a benchmark someone other than PDF Solutions ran, a general-availability date, and eventually a line in a quarterly release that attributes revenue to it. Until those arrive, 25x is a design target for a product a small number of unnamed customers are only now beginning to test.