PLMish earnings check: what have we learned?

So. Many. Earnings. Reports. Dassault Systèmes, PTC, and Siemens have reported results for their latest quarters, and, not surprisingly, the companies used their earnings call to make basically the same argument: everything old is cranking along — but many things are now AI, and AI is worthless without governed, trustworthy product data. Convenient, since that favors incumbent vendors.

Let’s do a quick recap of what we’ve learned so far. Today, I’m focusing on DS, PTC, and Siemens, and I’ll get to the others in groups as I can. First, some key figures and announcements — see each company’s investor relations website for far more detail:

	Dassault Systèmes (DS)	PTC	Siemens (DI Software only except for guidance)
Period ending 30 June	Q2 FY2026	Q3 FY2026	Q3 FY2026
Revenue/growth	€1.556B, +4%; subscription +8%	$600M; ARR $2.5B, +9%	Software rev. +15%; orders ~€1.7B (+5%)
Key driver	3DEXPERIENCE & Cloud +14% (cloud +60%)	Net new ARR $60M, above guidance	EDA +30%+, Altair/Dotmatics integration
Big news	$1.8 billion acquisition of ArisGlobal to scale AI intelligence and compliance platforms in Life Sciences, countering softer legacy Medidata segments.	Completed sale of Kepware/ThingWorx. Largest AI deal to date (a near seven-figure ServiceMax AI deal), utilizing the narrative that deep asset data governance yields a 50% reduction in technician prep time.	Launched Intelligence Center X
Guidance for FY2026	Total revenue €6.296B–€6.416B (3%–5% growth ex-FX)	Total revenue $2.69B–$2.75B; constant-currency ARR growth (ex-Kepware/ThingWorx) of 9%–9.5% (up from 8.5%)	DI (software + automation) revenue up 7%–10%

Diving a little deeper,

DS’s revenue growth continues to be dragged down by the Life Sciences (Medidata) business, where revenue declined 3% in FQ2 on low 2025 booking volumes and the loss of a large Moderna contract. Add to that a “soft quarter” in Europe (flat in CQ2 after a strong CQ1, per CFO Rouven Bergmann), with automotive particularly challenged. As the table shows, DS’s subscription/cloud revenue is strong — subscription up 8%, 3DEXPERIENCE Cloud up 60% — but the blend still gets diluted by weaker legacy maintenance revenue, a problem shared across our PLMish companies. DS expected this: CEO Pascal Daloz called 2026 “a foundation year, not because we expect less, but because we are building for much more” as DS invests in its agentic platform buildout. The company confirmed its full-year guidance, sees potential improvement in Medidata, and still needs to digest the ArisGlobal acquisition, expected to close late Q3/early Q4 2026.

PTC’s revenue was below expectations; CFO Jen D’Errico said the $600 million figure reflected “only the shortened duration of a single large contract expansion,” with deal durations elsewhere holding steady. ARR (Annual Recurring Revenue — how PTC and others value subscriptions and recurring contracts as if extended for a full year) was above expectations: net new ARR (God, I hate accounting; that’s new bookings plus expansions, minus churn) came in at $60 million, while constant-currency ARR — $2.448 billion, up 9.1% year over year — beat the high end of guidance. In other words, the FQ3 revenue shortfall was due to timing and product mix, not demand, and PTC raised its full-year ARR and revenue guidance.

Siemens Digital Industries Software’s FQ3 revenue was up 15%, driven largely by EDA (+30%). Orders were €1.7 billion, up 5%, helped by a major PLM order from a large automotive OEM modernizing its infrastructure; organic ARR was up 11%. EDA orders were “softer, as expected,” which CFO Veronika Bienert linked to tough comparables: “we see DI orders around the prior year level on very tough comps due to an exceptionally high volume of EDA bookings.” In short, EDA had an unusually strong quarter a year ago, making this year’s comparison look soft even though the underlying business isn’t necessarily weakening. Siemens confirmed its FY2026 guidance, with DI revenue expected up 7%–10%, though Ms. Bienert expects order volume below last year’s record level.

The AI pitch, three ways

Let’s look specifically at what the companies said about AI, since that’s the question I get most often.

Siemens wants to be the operating system for industrial AI across the whole real-to-digital stack. It announced Intelligence Center X — a knowledge graph, industrial ontologies, AI Studio and Mendix — meant to get customers started quickly with AI pilots and then into production. CEO Roland Busch put it plainly: Siemens isn’t tokenizing software yet; it’s still license/SaaS-based while it watches how customers want to use AI. As a sign that customers buy licenses for AI-native tools, he pointed to Eigen, a generative AI assistant connected to TIA Portal that generates PLC code from prompts, configures hardware and automation devices, and builds HMIs (human-machine interfaces) for factory automation — not a PLM product, but an interesting proof point. Mr. Busch said hundreds of customers across 30+ countries have signed on, from small system integrators to global OEMs, paying license fees in “a market where people are used to paying for licenses.” He called it one of the fastest-growing products he’s seen launched, with customers “doubling and tripling” in recent weeks.

Dassault Systèmes went hardest on architecture. Laying out the “building year,” Mr. Daloz described 3DEXPERIENCE going “agentic,” with AI built natively into the platform rather than bolted on through a chatbot or an MCP wrapper. He drew a sharp line against “frontier model + MCP” approaches, arguing that industrial customers don’t want their IP flowing into someone else’s LLM. That’s, however, where most users currently live: using a general-purpose AI model (from OpenAI, Anthropic, Google, etc.) trained on language and general knowledge, connected via MCP (Model Context Protocol) to call external tools, apps, or data sources to complete tasks, essentially orchestrating it against your existing software. (If I use Claude to summarize my emails, that’s Claude+MCP.) Mr. Daloz sees this as a chatbot layered on legacy applications rather than AI built into the platform’s core, and he’s not wrong: a general model doesn’t understand the underlying physics, biology, or engineering constraints, so while it can automate workflows, he argues it’s “useless” for things like advanced design exploration, where domain-specific reasoning matters. Dassault’s alternative — AI built natively into the platform with contextual, physics-based understanding — is, he says, far superior.

PTC told a similar story about being the system of record: CEO Neil Barua said, “AI requires PTC’s systems of record (CAD, PLM, ALM, SLM) and the structured product data inside them to be effective.” AI can’t reliably use unstructured or ungoverned data, he argued — PTC provides the structure, context, and governance that makes AI outputs trustworthy in engineering and manufacturing settings. Like the others, he said customers are wary of handing proprietary data to frontier-model providers, and that PTC keeps that data inside governed enterprise environments, with frontier models treated as “infrastructure” while PTC’s systems handle the data/workflow layer where the actual engineering work happens. Mr. Barua and CFO Jen D’Errico said customers are modernizing and consolidating their product data foundation onto PTC systems (citing several competitive displacements and expansions), with AI-enabled products such as Creo AI, PTC Orbit, Onshape Labs and ServiceMax AI driving some of that activity.

TL; DR. What does this mean for PLMish AI? Right now, each company seems to be staking out a different layer of the stack. Siemens is making an infrastructure and tooling bet: build a horizontal AI platform (Intelligence Center X, knowledge graphs, ontologies, Mendix) that spans the whole portfolio, and keep the license/SaaS business model unchanged for now. Dassault is insisting the AI itself has to be physics-grounded and native to the platform, positioned directly against the frontier-model-plus-MCP approach. PTC sees the frontier models as commodity “infrastructure” and PTC’s systems of record as the layer that makes any model’s output trustworthy. That means PTC doesn’t need to win the argument about whose AI is smartest—it just needs to be where the structured product data lives.

All three are constructing the same moat (governed domain data >> raw frontier-model access), which is probably correct but oh so convenient for these incumbent providers. I’m speaking with a lot startups that want to get in on what may turn out to be a very big pie, with niche solutions that could destabilize all of this. Maybe. Someday.

Still to come this week: Synopsys and Autodesk

Synopsys reports Q3 FY2026 today, after market close. Broker consensus has revenue around $2.44B (up 40% year over year, largely reflecting the Ansys acquisition). Of course, there’ll be a lot of interest in AI’s role in Synopsys’ core business, but also in how Ansys is performing inside the combined company.

Then, on Thursday afternoon, Autodesk reports Q2 FY2027. Analysts expect revenue of roughly $2.01B (up 14% year over year). Worth watching: what it says about AI-for-the-masses, whether AECO/construction momentum continues, and the first quarter of commentary since the MaintainX close.

And then it will be time for a very, very long nap.