By Rajesh Beri | July 12, 2026
The enterprise AI conversation has spent three years fixated on knowledge work — chatbots, code assistants, customer service, document summarization. Meanwhile, the $252 billion operational technology market that actually runs the physical world has watched from the sidelines, stuck in what HiveMQ's 2026 industrial AI report calls "permanent pilot purgatory." Sixty-eight percent of manufacturers cannot scale AI past the proof-of-concept stage. Eighty-five percent of industrial AI projects fail due to data quality, not model capability.
On June 30, 2026, Schneider Electric announced a definitive agreement to acquire Cognite — the Norwegian industrial data and AI platform company — for $3.1 billion in all cash. It is the largest software exit in Norwegian history, and the clearest signal yet that the industrial AI race will not be won by whoever has the best model. It will be won by whoever owns the data layer that makes models useful in a factory, refinery, or power grid.
This article unpacks why the data layer is the real bottleneck, how the competitive landscape is consolidating around five industrial AI platform strategies, and provides two frameworks every VP of Operations and CTO at an industrial enterprise should run before their next platform decision.
The $37 Billion Problem: Why Industrial AI Is Fundamentally Different
The global industrial AI market hit $37.45 billion in 2025 and is projected to reach $202.66 billion by 2035 — an 18.4% CAGR that outpaces enterprise SaaS growth. Yet the failure rate remains catastrophic. Gartner reports that only 41% of AI projects transition from prototype to deployment. In manufacturing specifically, 80–95% of AI pilots never reach production.
The reason is not that manufacturing companies picked the wrong model or the wrong cloud. The reason is that operational environments produce data that is fundamentally hostile to AI.
What Makes OT Data Different From IT Data
| Dimension | IT Data (Enterprise AI) | OT Data (Industrial AI) |
|---|---|---|
| Structure | Relational, APIs, documented schemas | Tag-based, proprietary protocols, no standard schema |
| Volume | Millions of records | Billions of time-series data points per facility |
| Latency | Batch acceptable | Real-time required (milliseconds for control loops) |
| Context | Self-describing (column names, metadata) | Meaningless without asset/process context |
| Systems | 5–15 SaaS apps | 200+ siloed systems per plant (SCADA, DCS, historians, ERP) |
| Lifecycle | 3–5 year refresh | 20–40 year equipment lifecycles |
| Governance | GDPR, SOX, clear ownership | No standard governance, fragmented ownership |
A GPT-class model with no access to a refinery's actual sensor history, asset hierarchy, and process context is useless to a plant manager. The model is not the bottleneck. The data is.
The Data Quality Crisis in Numbers
- 85% of industrial AI projects fail due to inadequate data quality or lack of relevant data — Gartner 2026
- 68% of manufacturers remain stuck in pilots, POCs, or research phases — HiveMQ 2026 Report
- 12% of IT/OT budgets now allocated to AI, with 83% planning increases — Cisco 2026 State of Industrial AI
- 465% ROI and payback periods under 6 months when data contextualization is solved first — Forrester TEI study of Cognite customers
- $173 billion in industrial manufacturing M&A over the past year, with mega-deals above $5B accounting for 56% of value — PwC 2026 Mid-Year Industrials Outlook
The gap between "AI works in the lab" and "AI works in my plant" is almost entirely a data contextualization problem. That is precisely what Cognite solved — and what Schneider paid $3.1 billion to own.
What Schneider Actually Bought
Cognite, founded in Oslo in 2017 and spun out of Norwegian industrial conglomerate Aker, built a platform called Cognite Data Fusion that does one thing exceptionally well: it ingests messy, heterogeneous operational data from legacy equipment and transforms it into a structured, AI-ready knowledge graph.
The technical architecture has three layers:
-
Industrial Data Foundation — Cognite Data Fusion connects to hundreds of source systems (SCADA, historians, ERP, maintenance systems) and creates a unified data model with semantic relationships between assets, processes, and events.
-
Knowledge Graph + Contextualization — Raw sensor tags become meaningful. "Tag_4471_PV" becomes "Compressor K-100, Stage 2 Discharge Temperature, Unit: Celsius, Normal Range: 85–110°C." This contextualization is what makes AI queries possible.
-
Agentic AI Workbench (Atlas AI) — Once data is contextualized, Cognite's Atlas AI agents can dynamically model operations, detect anomalies, recommend actions, and execute workflows — all grounded in real operational context rather than hallucinated from generic training data.
Key metrics:
- 2025 annual revenue: $170 million
- Employees: 885 globally
- Customers: ADNOC, Aker BP, TotalEnergies, International Paper, Rockwell Automation
- Total raised: $225 million across 3 rounds
- Exit multiple: roughly 20x Aker's invested capital
The case study that proves the thesis: TotalEnergies signed a three-year enterprise-wide deployment in September 2025, scaling Cognite across 39 upstream assets and 200+ production systems. The company reported a 12% improvement in operational efficiency. ADNOC won Cognite's 2025 AI Innovation Award for deploying AI agents across its offshore operations.
Why Schneider — And Why Now
Schneider Electric CEO Olivier Blum was explicit on the July 1 investor call: "It's extremely important that we create a company that is able to connect the physical and digital world, a company where we can capture data, structure data, contextualize data, and deliver those data across the life cycle."
AVEVA CEO Caspar Herzberg added that Cognite's knowledge graph is "incredibly important because you will be able to dynamically model the changing data, the changing operations data, the sometimes changing asset data for your robots."
The strategic logic is threefold:
-
Closing the generative AI gap. ARC Advisory Group ranked Schneider first on implementation criteria among industrial software vendors but noted it lagged Siemens on generative AI depth. Cognite closes that gap overnight.
-
Vertical integration from hardware to intelligence. Schneider already sells sensors, switchgear, and automation hardware. AVEVA provides engineering design and SCADA. Cognite adds the data contextualization layer that connects hardware to AI. The result: a full-stack industrial AI offering from power infrastructure to autonomous operations.
-
Eliminating third-party dependency. Without owning the data layer, Schneider was dependent on external platforms every time a customer wanted AI-driven workflows on top of Schneider hardware. That dependency is acceptable when AI is a feature. It is fatal when AI becomes the core product.
The acquisition also eliminates the market's most credible vendor-neutral industrial data platform. Cognite deliberately served customers across competing automation ecosystems — Schneider, Siemens, Honeywell. That neutrality is now a competitive asset Schneider must manage carefully, or risk losing the customer trust that justified the premium.
The Industrial AI Platform War: Five Strategies Competing for the Same Budget
The Cognite acquisition accelerates a consolidation wave that CB Insights tracked at 266 AI M&A deals in Q1 2026 alone — a 90% year-over-year increase. Every major automation vendor is now placing bets on how to own the industrial AI stack. The strategies diverge significantly.
Framework #1: Industrial AI Platform Vendor Comparison Matrix
| Vendor | Platform | Data Strategy | AI Approach | Lock-in Risk | Best Fit |
|---|---|---|---|---|---|
| Schneider/AVEVA + Cognite | Cognite Data Fusion + AVEVA | Unified knowledge graph, vendor-agnostic ingestion | Agentic AI (Atlas AI), low-code agents | Medium — historically neutral, now Schneider-owned | Multi-vendor brownfield plants, energy/process industries |
| Siemens | Industrial Copilot + Xcelerator + MindSphere | Digital twin-first, tight integration with Siemens PLCs | GenAI copilot (Microsoft/Azure OpenAI partnership) | High — native to Siemens ecosystem only | Siemens-equipped discrete manufacturing |
| Honeywell | Forge | Process-centric data model, refinery/chemical focus | Predictive analytics, closed-loop optimization | High — Honeywell DCS/instrumentation dependency | Process industries (refineries, chemicals, pharma) |
| Rockwell Automation | FactoryTalk + Plex | Allen-Bradley ecosystem, discrete manufacturing | Analytics-first, acquiring robotics startups | High — Allen-Bradley PLC dependency | Discrete manufacturing, automotive, food & bev |
| Hyperscalers (AWS/Azure/GCP) | IoT + AI services (SageMaker, Vertex, Azure AI) | Bring-your-own-data, cloud-centric | General-purpose ML/AI, no industrial context | Low vendor lock, high integration cost | Tech-forward enterprises with strong internal teams |
The critical insight: Siemens, Honeywell, and Rockwell all have industrial AI platforms — but they are all captive to their own hardware ecosystems. Cognite's historical advantage was serving customers regardless of which automation vendor's equipment was on the floor. Schneider now owns that cross-vendor capability, creating a strategic dilemma for every Cognite customer running non-Schneider equipment.
What About C3 AI, AspenTech, and ABB?
- C3 AI positions as industry-agnostic but has struggled with enterprise adoption (stock down 70% from 2021 highs) and lacks deep OT integration.
- AspenTech (now part of Emerson) dominates process optimization in chemicals and refining but is narrower than a full-stack industrial data platform.
- ABB Ability Genix competes directly with Schneider/Cognite in electrification and process automation, but lacks Cognite's breadth of third-party industrial data ingestion.
Framework #2: Industrial AI Data Readiness Assessment
Before selecting a platform, industrial enterprises must honestly assess where they stand on the data maturity curve. The 68% stuck in pilot purgatory share a common pattern: they bought AI tools before solving the data problem underneath.
The 5-Level Industrial AI Data Maturity Model
| Level | Name | Characteristics | AI Capability | % of Manufacturers (2026) |
|---|---|---|---|---|
| 1 | Siloed | Data locked in proprietary systems (SCADA, historians, spreadsheets). No integration. Manual exports. | None beyond basic dashboards | ~25% |
| 2 | Connected | Point-to-point integrations exist. Data accessible but without context. Tag names meaningless outside one system. | Basic analytics, threshold alerts | ~30% |
| 3 | Contextualized | Unified data model maps assets, processes, and relationships. Tags have semantic meaning. Data queryable by non-experts. | Predictive maintenance, anomaly detection | ~15% |
| 4 | AI-Ready | Knowledge graph operational. Real-time + historical data unified. Governance in place. Data lineage tracked. | Agentic AI, digital twins, autonomous recommendations | ~8% |
| 5 | Autonomous | Closed-loop AI systems make operational decisions. Human oversight for exceptions only. Continuous learning from outcomes. | Self-optimizing operations | ~2% |
How to Score Your Organization
For each dimension, rate 1–5:
- Data Accessibility — Can a data scientist query your operational data without help from a control systems engineer?
- Semantic Context — Do your data tags have machine-readable meaning (asset hierarchy, physical units, normal ranges)?
- Integration Breadth — What percentage of your operational systems feed a unified platform? (<20% = 1, 20–40% = 2, 40–60% = 3, 60–80% = 4, >80% = 5)
- Real-time Availability — Is operational data available to AI systems within seconds of generation?
- Governance Maturity — Do you have clear ownership, quality standards, and lifecycle management for operational data?
- Cross-site Consistency — Is the same asset described the same way across all your facilities?
Scoring:
- 6–12: Level 1–2. Invest in data foundation before AI. Platform selection premature.
- 13–18: Level 2–3. Ready for pilot. Choose platform based on ecosystem fit.
- 19–24: Level 3–4. Ready to scale. Platform selection is strategic (see comparison matrix).
- 25–30: Level 4–5. Ready for autonomous operations. Focus on closed-loop deployment.
The ROI Case for Data-First Investment
The Forrester Total Economic Impact study of Cognite customers quantified the payoff of solving data contextualization first:
- 465% ROI over three years
- $29.4 million in total benefits for a composite organization
- Payback period under 6 months
- Primary value drivers: reduced unplanned downtime (40%), faster engineering decisions (30%), eliminated manual data wrangling (20%), reduced safety incidents (10%)
Compare this to the cost of skipping the data layer: 80–95% of AI pilots fail, each consuming $500K–$2M in resources before being abandoned. A manufacturer running four failed pilots per year burns $2–8 million with zero production value.
The Vendor-Neutrality Dilemma: Schneider's Biggest Risk
Every acquisition creates tension between the buyer's desire for competitive advantage and the target's value as a neutral platform. Cognite's neutrality — serving TotalEnergies (which runs Honeywell DCS alongside Schneider equipment) and ADNOC (which uses multiple automation vendors) — is precisely what made it worth $3.1 billion.
Schneider CEO Blum has indicated that Cognite will continue serving multi-vendor environments. But the market will be watching three signals:
- Pricing parity — Will non-Schneider customers pay more or receive fewer features than Schneider-ecosystem customers?
- Integration priority — Will new Cognite capabilities ship first for Schneider/AVEVA integrations?
- Data access — Will Cognite's knowledge graph become a conduit for Schneider to understand competitor equipment installed at customer sites?
If Schneider manages this correctly, they own the Switzerland of industrial AI — a platform everyone trusts because it serves everyone equally. If they weaponize it, customers defect, and the $3.1 billion premium evaporates.
Siemens and Honeywell are watching. Expect counter-moves within 18 months — either aggressive development of their own open data platforms, or acquisitions of remaining independent players like C3 AI, AspenTech-adjacent startups, or emerging platforms like Opsima.
What This Means for Enterprise Buyers: A 90-Day Action Plan
If You Are Already a Cognite Customer:
- Demand contractual neutrality guarantees before the deal closes. Lock in pricing and feature parity for 3–5 years regardless of which automation vendor's equipment you run.
- Assess data portability. Understand exactly what format your Cognite knowledge graph exports in. If Schneider decides to make the platform less neutral, you need an exit path.
- Accelerate value capture. With Schneider's scale behind it, Cognite's platform will likely improve faster. Get ahead of the feature curve.
If You Are Evaluating Industrial AI Platforms:
- Run the Data Readiness Assessment above. If you score below 13, invest in data foundation (connectors, contextualization, governance) before choosing an AI platform. Buying AI tools for Level 1–2 data is burning money.
- Map your automation ecosystem. If you are 80%+ Siemens, their Industrial Copilot is the path of least resistance. If you run multi-vendor (most large manufacturers), the Schneider/Cognite stack or a hyperscaler approach gives flexibility.
- Calculate the pilot purgatory cost. Total up what you've spent on AI POCs that never reached production. That number is your business case for solving the data layer first.
If You Are Siemens/Honeywell/ABB-Dependent:
- Watch for retaliatory moves. Siemens has MindSphere and Microsoft. Honeywell has Forge. Both will accelerate data platform investments in response.
- Avoid panic switching. Cognite will remain multi-vendor for now. Premature platform migration destroys more value than it protects.
- Negotiate exit clauses into any new platform contract. The market is consolidating fast. Any platform commitment beyond 3 years should include data portability guarantees.
The Bigger Picture: Industrial AI Is the Next Trillion-Dollar Battleground
Enterprise AI has consumed $9 billion in deployment infrastructure in 2026, as we covered in the deployment wars analysis. But that spending was almost entirely aimed at knowledge workers — software engineers, analysts, customer service agents.
The operational technology market — the $252 billion of systems that run factories, refineries, power grids, water treatment plants, mines, and logistics networks — represents a far larger economic surface area. When AI can reliably optimize a chemical plant's throughput by 3–5%, the annual value creation dwarfs what a coding assistant saves on developer productivity.
Schneider's $3.1 billion bet says the same thing Cognite's customers have proven: the model is a commodity. The data layer is the moat. The industrial AI winners will be determined not by who has the smartest model, but by who can make 40-year-old sensor data legible to modern AI systems — at scale, in real time, with governance that regulators accept.
The agentic AI governance crisis we documented yesterday is about to arrive in factories. When AI agents are recommending pressure changes on a reactor vessel or scheduling maintenance windows for a turbine, the stakes are not "wrong email sent." They are "catastrophic equipment failure." The industrial AI stack needs governance baked in from the data layer up — not bolted on after deployment.
Sixty-eight percent of manufacturers are stuck in pilot purgatory. Schneider just paid $3.1 billion for the company that showed the other 32% how to escape it. The question for every industrial enterprise is not whether to invest in the data layer. It is whether you can afford to wait while your competitors do it first.
Rajesh Beri is Head of AI Engineering at Zscaler, where he builds enterprise AI systems for security, compliance, and operational intelligence.
Continue Reading
- The $9B Deployment War: Why Every AI Vendor Now Wants Engineers in Your Office
- 54% Had AI Agent Incidents. The 5-Layer Control Stack.
- $4.2B for 3 Companies: Why Accenture Just Rewrote OT Security
- Xpress Boats Cuts Shipping Costs 50% With Infor AI Agents
- NVIDIA's 2026 State of AI: The Hard ROI Numbers Every CFO Needs
