Why AI can’t replace the PLM ecosystem...at least not yet.
Artificial Intelligence (AI) is no longer a futuristic concept in product development; it is actively reshaping the tools we use to design, manufacture, and support our next generation of products. This article offers a deeper look into the impact of AI in its many forms (agentic, machine learning, deep reasoning, etc.) on PLM-enabling software solutions (e.g., PDM, CAD, CAM, ERP, and simulation & analysis) that are used to design, develop, manufacture, and support competitive and innovative products and services.
The main issue is how solution providers are moving beyond traditional software features to offer AI-enhanced benefits, new user capabilities, and new revenue models based on usage or results rather than on user “seats” or headcounts. Why? Because due to AI’s potential benefits and apparent rapid adoption, PLM-related toolsets, capabilities, and data models used over the past four decades are at risk of being “hollowed out” in something called “disintermediation.” This is digital transformation on steroids, disrupting the use of traditionally licensed PDM, CAD, and many other PLM-related solutions to an extent never before imagined.
Despite this aggressive outlook, we believe that many predictions of AI’s impact on enterprise software are overblown, especially those related to complex, domain-specific PLM solutions. That said, many of today’s leading PLM solution providers aren’t waiting to be disrupted; rather, they’re setting out to be the disruptors.
Cadence, Dassault Systèmes, PTC, Siemens, and Synopsys are all adopting Nvidia’s Omniverse platform, for example, to develop new applications for “autonomous systems” such as digital twins, robotic simulations, cameras, and self-driving cars. Nvidia terms this Physical AI, noting that it allows systems and objects to “perceive, understand, reason, and perform or orchestrate complex actions in the physical world.” Not to be left out, Aras is also working on transforming PLM and digital engineering with a digital thread optimized for AI integration.
At the heart of the disintermediation debate is whether AI can fully enable the engineering process, and thereby replace the traditional PLM-enabling solution that we have come to know and love. This shift threatens to redefine the very components of a PLM environment—including ‘apps,’ control systems, and APIs—and perhaps even calls into question the future role of humans in production and service.
Believing that AI is a threat to highly sophisticated PLM and its vast and growing set of engineering- and simulation-oriented “apps” is a clear sign of having bought into the AI hype. Given the fears amid the disruption emerging in multiple enterprise software spaces, as well as any talk of disintermediation, we at CIMdata recommend, at least from a PLM perspective, redefining AI as “augmented” intelligence rather than “artificial” intelligence. Here AI focuses on augmenting human intelligence rather than replacing it.
What is happening?
Whenever we delve into AI, we need to separate fact from fiction. A few published examples of the AI hype one reads of in today’s media include:
- AI could replace more than 50% of enterprise software spending as IT systems are rebuilt around AI and move away from traditional SaaS (“Software as a Service”) delivered tools.
- AI-driven workflows will collapse the “stack of apps” used in working with data in IT, human resources, or services.
- Investors and analysts seem convinced that Agentic AI is eclipsing SaaS licensing models and dashboards.
- Rapid AI adoption raises many hurdles, including how new products are launched.
- Gen-AI and AI agents could just as easily disrupt revenues as enhance them.
To this jumble of potential truths and likely half-truths, I must stress that while AI agents are a smarter way to interact with apps, enterprise data platforms are still needed to keep our data organized, legal, and secure. Additionally, it is important to note that SaaS is a software delivery model—one that can be, and is, used to deliver AI-empowered applications.
My colleague Dr. Diego Tamburini, CIMdata’s Director of AI, puts it this way: “The No-UI concept (No User Interface) is overblown; in the complex tasks of engineering or manufacturing, people need to visualize data they’re working with.
We can’t just ‘agent’ our way out of complexity. If we tried to replace a massive system like Teamcenter with a collection of AI agents, for example, we’d have to spend years rebuilding the rules and logic already built into Teamcenter. But we can add AI to tools like Teamcenter.” Other leading software providers are not replacing their SaaS solutions; instead, they are baking AI into their existing tools to make them better than ever.
Dr. Tamburini goes on to state that, “The idea that AI will ‘replace’ SaaS confuses a change in technology with a change in business model. Moving from perpetual licenses to SaaS was a fundamental shift in how we buy and deploy software. AI isn’t a new way to buy software; it’s a new engine under the hood. You can be a SaaS company and an AI company at the same time. The only place where AI might truly displace traditional SaaS applications is at the lower end of the market, where workflows are simple enough that a basic chatbot might do the trick. We aren’t saying goodbye to SaaS; we’re moving into an era where the software does more of the heavy lifting for us. What is dying is the per-seat revenue model. When a single AI agent can do the work that previously required many people, charging ‘per person’ makes less and less sense. We’re heading toward paying for actual results or usage.” This is a clear indication of “back to the future,” when enterprise software usage models were prevalent in the time-shared computer domains!
What the leading software providers are actually doing is outlined in a recent webinar I gave titled, “Unpacking AI-led Disintermediation Risks for Engineering & Manufacturing Software.” We will be talking more about this in upcoming forums and at CIMdata’s PLM Road Map conference in May in the Washington, D.C., area.
For this article, we will briefly look at what Aras, Dassault, PTC, SAP, and Siemens are doing, as well as data “data platform” disruptors such as Snowflake, Databricks Inc., and Palantir Technologies.
What are PLM solution providers doing?
At this point, we need to look into what the solution providers are doing—disintermediation or otherwise. Their fundamental concerns include, of course, primarily whether they can monetize AI enablement and what might AI do to their revenue streams?
The issues they are wrestling with include:
- If we build it, will the users follow?
- Will application architectures change? And if so, in what ways?
- Will the knowledge workers they currently sell software to be replaced and, if so, by what?
- Can complex PLM-related capabilities leave out the user?
- If users are left out, will the solution providers still be able to monetize their offerings without them?
Aras offers an example of implementing AI as a seamless extension of its PLM Environment, focused on leveraging the digital thread to provide the data AI needs to deliver meaningful industrial insights; the underlying insight is that “AI is only as good as the data it can access.” For Aras users, one of the most visible parts of its strategy is Aras AI Assistant, an AI-powered assistant integrated with the Aras Innovator interface, delivering a Gen-AI implementation in its Low-Code architecture. This enables users to build their own AI-driven apps on the Aras platform. It also allows organizations to create specific “agents” tailored to their unique needs.
Dassault Systèmes’ strategy focuses on implementing AI as the “brain” of the Virtual Twin in its 3DEXPERIENCE platform. It combines GenAI Large Language Models and Large Multimodal Models into physics-based simulations. These “Generative Virtual Twins” can be used to generate viable 3D geometry, material configurations, and manufacturing routings, for example. In everyday use, SIMULIA, the Dassault Systèmes simulation solution suite, can be used to run tests as diverse as stress and fluid dynamics; results are fed back into the Dassault Systèmes AI brain to optimize the design further.
At work throughout Dassault Systèmes’ design and data-management solution offerings, AI also:
- Automatically classifies parts, identifies duplicates across global databases, and suggests standardized components.
- Analyzes global market trends and news to predict the supply chain disruptions before they impact production.
PTC has determined that AI’s usefulness in an industrial context required embedding it in the daily workflows of engineers, factory managers, and service technicians—an approach that PTC notes is focused on practicality, connectivity, and prediction. Through its acquisition of Frustum, PTC became an early marketer of Generative Design, which can be used to generate optimal geometry based on weight, material, and manufacturing constraints.
And thanks to its Creo Simulation Live partnership with Ansys, PTC’s AI-enhanced solvers give users instantaneous feedback. PTC has also introduced Gen-AI Copilots in Windchill+ and Codebeamer to analyze requirements, assist with change-impact analyses, and support data migration.
SAP, for example, uses its Business Technology Platform (BTP) to provide its “AI Foundation” with a strategy built on relevance, reliability, and responsibility. The SAP AI strategy manifests as a “Design-to-Operate” lifecycle that assists in sustainable design and provides accurate demand forecasts. SAP uses Joule, its Gen-AI natural-language assistant, as the universal interface across its entire portfolio to:
- Help users find information across S/4HANA, SuccessFactors, and Ariba.
- Draft job descriptions, generate code for SAP BTP extensions, and summarize complex financial reports.
Siemens is moving beyond what it terms “Generative Design” toward “Generative Engineering,” involving AI to automate the entire engineering process, not just geometry creation. Their strategy focuses on Industrial AI as deterministic, transparent, and operating within strict safety and accuracy requirements. Siemens NX, for example, provides an AI-enabled NX Command Prediction that anticipates user actions. Additionally, Siemens’ Simcenter Neural Networks can be used to create Reduced Order Models (ROMs). They also offer an Industrial Copilot developed with Microsoft Corp., and AI is enabled within Teamcenter to manage the Digital Thread, change impact analysis, identify duplicate parts across global systems, and find standardized alternatives.
And now a quick look at those three “data disrupters” I mentioned above:
- Snowflake is a cloud-native SaaS platform designed for data warehousing, analytics, and AI/ML workloads, operating on major public clouds such as AWS, Azure, and GCP.
- Databricks is a unified, cloud-based Data Intelligence Platform for data engineering, analytics, and AI.
- Palantir Technologies builds platforms for big-data analysis, integration, and decision-making in government intelligence organizations and defense agencies as well as large corporations.
Fundamentally, none of this is disintermediation of today’s complex, domain-specific PLM “app stacks”; it is augmentation. Will replacement come? Perhaps, but it is a non-trivial exercise, and with almost all PLM solution providers taking significant action, additional barriers to entry are being rapidly constructed.
High complexity
I began this discussion of AI and disintermediation in hopes that those working in the industrial domains of engineering, manufacturing, and service will see that these tools are too complex for AI to fully take over, now or in the foreseeable future.
This could change, of course, when (if) AI reaches “superintelligence”—sufficiently autonomous to surpass human brainpower in creativity, problem-solving, and everything else. To get there, AI has to progress beyond using today’s retrieval-augmented generation (RAG) capabilities to using tools outside itself to drive accuracy and functional utility.
This means AI must progress onward and upward through:
- Multi-modal LLM workflows, AI-built models that can process and generate content across diverse types of text and images.
- Advanced agent architectures capable of reasoning independently and executing tasks with multiple steps.
- Futuristic, next-generation architectures designed for full autonomy and seamless interaction with any digital or physical environment.
All that is in the future. Today, in the complex and deeply interconnected worlds of engineering and manufacturing, as noted above, we must view AI as “augmented” intelligence. AI is not about replacing human intelligence but about amplifying it…helping humans benefit from the massive volumes of data we’re generating by combining human and machine intelligence.
AI operates on data—no question about it—and that’s why it’s labeled “knowledge-based.” Meanwhile, solution providers are wrestling with vexing questions:
- Can AI be used now as a lifecycle assistant—even if the underlying data is incomplete or not entirely valid?
- Alternatively, could AI be used as a mechanism to tie together an organization’s digital twins?
And, on a larger scale—across the product lifecycle—can we benefit from:
- Gen-AI powered by LLMs that can pull data from documents without the pain of finding and opening them?
- Deep-Learning Models that generate high-quality text, images, and other content-based data on which they have been trained?
AI has gone through many cycles of hype and will go through more.
Proof? Look at the astonishing gain in the value of Nvidia’s publicly traded shares of stock; to many investors, Nvidia and its Omniverse platform are AI. And many skeptics now see the release of ChatGPT and similar tools as marking a significant turning point.
To sum up, we must understand and be prepared to support the incorporation and use of AI to generate additional value in our lifecycle processes, but certainly not to the point of disintermediation. We must address data quality issues, launch tightly focused pilot projects, and prepare the workforce to use AI.
