Schneider Electric buys PTC for $22.6B to expand into AI
Industrial automation just wrote the biggest AI check of the day — and a public benchmark is quietly deflating one of the guardrail market's hottest pitches.
Schneider Electric has agreed to buy PTC for about $22.6 billion in cash, its largest acquisition ever and an explicit bet on turning industrial software into an AI business. The French automation giant will pay $205 per share — a 42.3% premium to the last close, 46.1% over the 30-day average — with the deal expected to close in the third quarter of 2027, financed with equity and new debt. Schneider described the combination as building "a scaled industrial software and AI business built around open and interoperable systems," promising €250 million in annual run-rate cost savings by the third year after close and roughly €800 million in revenue synergies; PTC brings more than 30,000 customers, about €2.4 billion in annual revenue, and a near-40% adjusted operating margin. Investors flinched — Schneider shares fell 4.7% in pre-market trading — though a J.P. Morgan note argued the company's deals have "typically proven strategically astute, if debatable from a valuation standpoint." Our read: this is the AI-fication of the factory floor at scale. Schneider already spent $3.1 billion on the industrial data platform Cognite in June, and PTC's product-lifecycle software is where AI agents could eventually touch physical products rather than documents — which is exactly why an industrial automation company is paying a 42% premium for a software logo.
Red Hat benchmarked the new "decision models" against LLM-as-a-judge and ordinary classifiers — and the category didn't win. The test ran roughly 1,300 prompt-injection attempts and about 1,500 content-safety prompts against eight or nine guardrail options: TypeSafe AI's Jev — a model that returns calibrated probabilities over a state instead of generating text — placed fourth on prompt-injection accuracy at 86.35%, behind Qwen3.6-35B's 89.31% and Red Hat's own DeBERTa classifier's 89.01%, and while it topped the content-safety table at 86.20%, an open Nemotron variant sat just 1.13 points back at much lower latency. Red Hat's conclusion was that decision models "did not produce faster, cheaper, or higher-quality answers compared with LLM-as-a-judge" — but keep two caveats in view: Red Hat sells its own competing guardrails, and TypeSafe claims Jev is 193.6 times faster and 444.6 times cheaper than the alternatives, so this is one interested party's read of a benchmark it at least had the honesty to publish in full. It's a reminder that the useful question is never "which architecture wins" but "which option clears the bar at your latency and price" — as we noted in How to — tell a real benchmark from a marketing one.
What to watch: whether Schneider names specific AI products in the combined lineup before the Q3 2027 close, and whether TypeSafe answers Red Hat's numbers with a benchmark of its own.
Is industrial software the right place to underwrite an AI thesis, or is Schneider paying a 42% premium for a rebrand? Tell us in the comments.
