The industrial automation market doesn't pivot overnight. Still, when Neuron Industries surfaced from Y Combinator's summer batch in early 2026—just two founders, a sharp pitch, and an audacious plan—it posed a question the sector hadn't seriously entertained in decades: What if the programmable logic controller, that workhorse of factory floors and chemical plants, simply went away?
Not augmented. Not upgraded with a bolted-on AI module. Gone.
Dennis Ren and Kenneth Rhee, operating out of a modest El Segundo office, weren't proposing incremental improvement. Their vision: collapse the tangle of PLC, SCADA, and DCS systems into one AI-native platform that accepts instructions in plain English and spits out deterministic Python control logic. Whether Neuron survives its first customer pilots or joins the long list of overpromising automation startups matters, frankly, less than what its existence reveals. The industrial automation industry—long sheltered from software-first disruption by regulatory fortresses and a near-pathological aversion to risk—is entering unfamiliar territory.
For the first time in a generation, the architecture itself is up for debate.
The Landscape: Stable, Profitable, and Enormous
Consider the numbers. The global PLC market reached $13.33 billion in 2026, according to Mordor Intelligence, with projections pointing toward $16.4 billion by 2031—a compound annual growth rate of 4.24 percent. The distributed control system market sits at $21.89 billion as of this year, climbing to a projected $41.13 billion by 2034, per Fortune Business Insights. These aren't hypergrowth software categories. But they're foundational. The unsexy plumbing running chemical refineries, power grids, automotive lines, semiconductor fabs.
Stability here isn't accidental. Siemens, Rockwell Automation, Schneider Electric, Mitsubishi, ABB, Honeywell—they dominate because their systems work. Installed bases stretch into millions of control points. Engineering teams speak ladder logic and structured text the way developers speak Python. Switching costs are punishing. Safety certifications consume years. An hour of unplanned downtime can hemorrhage millions.
Yet manufacturing investment is surging. The CHIPS and Science Act, signed into law in August 2022, has been directing billions in federal funding toward semiconductor manufacturing and research, catalyzing private investment in new fabrication facilities across the United States. McKinsey Global Institute's May 2026 analysis of U.S. manufacturing highlights concentrated spending in electronics, chemicals, and metals. The question hanging over every new fab and battery plant: Do you rebuild the last fifty years of automation architecture, or do you take a chance on something fundamentally different?
Three Converging Forces
Something has shifted, or perhaps several things at once.
First, AI inference got cheap enough and fast enough to run at the edge. NVIDIA's Jetson Orin Industrial platforms and Intel's Atom x7000RE processors deliver edge AI capabilities with power efficiency and low-latency performance suitable for industrial environments. That opens a door previously nailed shut: closed-loop control where AI doesn't just analyze after the fact but makes real-time decisions inside the control loop itself.
Second, virtualization—once the domain of IT departments—has crept into operational technology. Siemens launched its S7-1500V virtual PLC in late 2024; TÜV certified a fail-safe variant by May 2025. CODESYS now ships Virtual Safe Control SL, a containerized safety PLC rated to SIL3. Schneider Electric unveiled its Foxboro Software-Defined Automation DCS in phases between February and September 2026, pitching it as the industry's first genuinely open, software-defined distributed control system. These aren't research projects. Audi's Böllinger Höfe facility is virtualizing shop floor control using Siemens' vPLC stack.
Third, the standards themselves are evolving to accommodate AI workflows. In April 2026, the OPC Foundation announced companion specifications tailored for agentic AI, preparing the OPC UA ecosystem for retrieval-augmented generation and model context protocols. Field-level communications are consolidating around OPC UA FX over time-sensitive networking, with the IEC/IEEE 60802 TSN profile under active development following a series of interoperability trials.
The Incumbents Aren't Sleeping
Siemens introduced its TM NPU 2.0 module, embedding AI inference directly into the S7-1500 and ET200MP controller families. Documentation updated through last year details use cases—visual inspection, predictive maintenance, anomaly detection—but the module operates adjacent to the deterministic control loop. Not inside it.
Rockwell Automation announced FactoryTalk ResilientEdge in June, pairing low-latency edge execution with cloud-based AI orchestration. The company's LogixAI analytics module has shipped since 2018, with version 3.03 arriving in May 2025. The strategy is additive: preserve the installed base, extend it incrementally.
Schneider's move feels more architectural. The Foxboro SDA system decouples control algorithms from hardware, running on virtualized infrastructure while maintaining IEC 61499-based open programming. ARC Advisory Group described this in a February analysis as a shift "toward open control architectures," reducing vendor lock-in without abandoning proven paradigms.
Then Siemens went further. In April, it unveiled the Eigen Engineering Agent, an AI assistant claiming to accelerate workflows by two to five times and boost efficiency by half. ANDRITZ Metals signed on early. At CES 2026, Siemens executives spoke expansively about an "Industrial AI Operating System." These aren't product refreshes. They're acknowledgments that the engineering bottleneck—the human work of designing and configuring control systems—might need rethinking as much as the runtime does.
Where It's Actually Happening
The City of Conroe, Texas, modernized its water operations using Schneider's EcoStruxure Automation Expert, deploying IEC 61499-based control in a municipal utility context. A January case study documented the rollout. In Japan, Sumitomo Chemical deployed Yokogawa's Exasmoc and Exapilot advanced process control tools, cutting temperature variability to roughly a fifth of baseline—meaningful gains in chemical manufacturing stability.
Notice a pattern? Water utilities. Chemical plants. Process industries, not discrete manufacturing. Process control has long pushed beyond basic ladder logic. The open question is whether discrete manufacturing—automotive, electronics, machinery—follows suit, and whether startups can crack markets that historically reward incumbency above all else.
A June preprint described a 5G-connected aerial robot executing autonomous flight in an industrial mine, with edge-offloaded control demonstrating distributed intelligence in subterranean environments. Mining isn't automotive assembly, granted. But it signals that latency-sensitive, mission-critical applications are beginning to trust AI-assisted control loops in environments where failure has consequences.
The Walls Remain High
Regulatory and certification requirements haven't budged. Any controller in a safety-critical role must demonstrate compliance with IEC 61508 (up to SIL3 for process industries), IEC 62061 for machinery, or sector-specific standards like IEC 61511. Nuclear applications face Nuclear Regulatory Commission scrutiny, with updated criteria in IEEE 7-4.3.2-2026 for programmable digital devices in safety systems.
Cybersecurity adds another layer of complexity. The ISA/IEC 62443 series, updated this past December, establishes security baselines for industrial automation and control systems. NERC CIP standards for critical energy infrastructure keep evolving—CIP-003-9 took effect in April. In July, CISA, FBI, and EPA issued joint advisories about Iranian-linked actors targeting Siemens, Schneider, and Rockwell devices through malicious project files uploaded to engineering workstations. The attack vector isn't the controller runtime. It's the engineering environment itself.
For a startup offering AI-generated control logic, this creates what you might call a credibility gap. How do you validate deterministic behavior from a system that writes code from natural language prompts? Academic research on using large language models for PLC programming remains inconclusive at best. A January 2024 paper titled "LLM4PLC" found significant limitations in generating valid, verifiable industrial control code. A May 2026 paper demonstrated ladder logic translation between Rockwell and Siemens platforms, but translation is not the same as generation from scratch.
Brownfield resistance compounds the challenge. Industrial automation forums and Reddit threads from mid-2025 through this year consistently emphasize retrofits and gateways over wholesale controller replacement. Validation, downtime, recertification—the costs make rip-and-replace a last resort. McKinsey's December 2025 survey of manufacturing COOs found that data integration and IT-OT convergence remain the primary AI deployment bottlenecks, not computational horsepower.
The Fork in the Road
The industrial automation market is splitting into two regimes, perhaps inevitably. Greenfield sites—semiconductor fabs, battery plants, other capital-intensive builds—may adopt software-defined, AI-native architectures from day one. Brownfield sites will bolt intelligence onto aging infrastructure through edge gateways, analytics overlays, engineering assistants.
Neuron Industries and startups like it face a narrow window and a steep climb. They need to prove not just technical capability but a credible path to certification, interoperability with legacy ecosystems, and performance guarantees sufficient to convince procurement teams whose careers depend on risk avoidance. That Siemens, Rockwell, and Schneider are moving aggressively to embed AI in their own stacks suggests they see the threat. Or the opportunity. Possibly both.
ARC Advisory Group's 2026 analysis identified what it called a "schism of speed" between industrial AI pacesetters and followers. Global Lighthouse Network sites, per McKinsey, are deploying AI in 90 percent of new use cases. But the gulf between pilot projects and production-scale deployments remains wide, perhaps wider than the optimists admit. Incumbents have distribution networks, certification expertise, decades of operational data. Startups have architectural freedom and faster iteration cycles.
The question isn't whether AI will reshape industrial control. In pockets, it already has. The question is whether that reshaping happens within the existing PLC paradigm or shatters it entirely. For manufacturing executives, the calculus is blunt: the cost of betting wrong on the incumbent is measured in wasted capital expenditure and missed efficiency gains. The cost of betting wrong on the challenger is measured in competitive disadvantage and, potentially, catastrophic system failure.
Both outcomes are expensive. Neither is obvious. And the clock, as they say, is running.
