Representative End User Clients
Representative Automation Clients
Representative Software Clients
Earlier this month, more than 30,000 data professionals gathered at San Francisco's Moscone Center for the Databricks Data + AI Summit (DAIS) 2026. The air was thick with keynotes, technical deep dives, and major platform announcements declaring a permanent architectural shift: the transformation of the data lakehouse into an active, governed control plane for the agentic enterprise.
Yet I made a deliberate choice to skip the Silicon Valley keynotes this year. Instead, I spent mid-June on the uncarpeted front lines, hunkered down with specialized industrial domain experts and operations engineers.
My decision wasn't born of a lack of interest in the cutting-edge frameworks coming out of the hyperscale ecosystem. Rather, it stems from a pragmatic reality we constantly encounter as we build out ARC Advisory Group's 3-Axis Industrial AI Models Taxonomy Research:
horizontal enterprise platforms are built for digital abstractions, but the cyber-physical world operates on physical variables, kinetic assets, and immutable safety rules.
During his opening keynote, Databricks CEO Ali Ghodsi stated a premise that resonates perfectly with our context-first research approach:
"AI does not have an intelligence problem; it has a context problem."
However, while horizontal technology providers are engineering impressive structures to solve enterprise context debt, industrial operations teams need to understand exactly where these tools succeed and where they hit an unyielding wall at the plant boundary.
The Horizontal Marvels: What Rolled Out at DAIS 2026
Make no mistake: Databricks and its immediate peers have established themselves as massive, undeniable centers of gravity for enterprise-level data aggregation and horizontal AIOps infrastructure. The announcements coming out of San Francisco represent an engineering tour de force designed to shift enterprise software from simple AI experimentation to production-grade, governed autonomy.
For information technology (IT) and data science executives looking to control agent sprawl, the new runtime layers offer highly sophisticated capabilities:
LTAP (Lake Transactional/Analytical Processing): By introducing native LTAP alongside its new Lakehouse//RT serving engine, Databricks aims to eliminate the historical latency gap between operational transactions and analytical queries on a single copy of data. For automated workflows, this limits the generation of "decision debt" caused by agents reasoning over stale data copies.
Omnigent and AgentBricks: These services provide standardized developer frameworks to deploy, connect, and share specialized agents across enterprise endpoints while preserving unified context via a living business ontology.
Unity AI Gateway and LakeWatch: Directly addressing the cost volatility of the Tokenpocalypse, the AI Gateway introduces a centralized interface for intelligent model routing, credential management, and token rate limiting. This runs in tandem with LakeWatch, an agentic SIEM framework that monitors runtime traces as security workloads.
Databricks Sandbox: To prevent autonomous code generation from corrupting active business processes, this feature deploys secure virtual machines with downscoped, "on-behalf-of" data permissions. It allows agents to experiment safely without the risk of taking down production databases.
The Analytical Reality: The Strategy of Horizontal Scale
These rollouts are invaluable for automating horizontal business workflows, managing customer data platforms (via the new CustomerLake), or optimizing corporate finance reporting logs. However, an analytical look at the platform's core code reveals a clear strategic choice: Databricks is focusing its immense resources on solving decades-old, horizontal data challenges at scale.
As an industry analyst, it is vital to separate marketing aspirations from core technical IP. Databricks' success is funding grander ambitions, but like the hyperscalers and silicon giants before it, the company is playing a high-volume platform game. It is explicitly counting on strategic alliances with industry domain specialists to deliver the much more challenging, niche, and lower-margin "last mile" solutions required by the physical world.
This is the exact point where a blended audience of IT, OT, and data science executives must align on the architectural boundary. A general-purpose enterprise lakehouse lacks the native, low-level semantic understanding required to govern a physical plant floor:
The Protocol Chasm: Enterprise platforms reason over clean text, tabular data streams, and standardized APIs. They do not natively ingest or normalize sub-millisecond process control loops, raw time-series telemetry, or binary industrial protocol frames (like Modbus, Profinet, or OPC UA) directly from edge machinery.
The Physics Bounding Deficit: A horizontal graph engine can map abstract corporate dependencies, but it has no native concept of process chemistry, fluid dynamics, thermodynamic limits, or plant asset configurations.
The Kinetic Intent Collision: While the Databricks Sandbox handles software containment beautifully, it cannot resolve an operational control loop oscillation. If an asset performance agent tries to throttle down a vibrating turbine for preventive maintenance while a procurement agent triggers a flow increase to satisfy a spot-market delivery window, a horizontal platform cannot arbitrate that physical conflict in real time.
The Industrial Requirement: Assembling the Cyber-Physical Context Layer
To bridge the power of enterprise data lakes with the high-consequence reality of live operations, industrial companies must realize that horizontal platforms are merely a component of a larger architecture. True multi-agent governance across factories and supply chains requires anchoring your data strategy in what our ARC Taxonomy defines as Cyber-Physical Context Engines.
Rather than forcing messy plant telemetry into a massive cloud lakehouse and attempting to sort out the relationships after the fact, pacesetting industrial frameworks process data in motion using a distributed hierarchy.
By placing specialized context architectures between your physical assets and your analytical enterprise core, you establish a structurally sound data foundation:
Edge Abstraction and Event Graphs: Platforms like HighByte handle the initial heavy lifting of edge-level DataOps, structuring chaotic tags into clean payloads at the machine interface. This feeds into modern, event-driven engines like Rhize, which maps legacy Purdue Model structures into real-time, GraphQL-native databases. By treating the operational event as the primary data driver, these tools give agents a fluid, highly extensible graph schema that runs circles around rigid relational databases.
Operational Event Mapping: Engines like XMPro wrap this data into composable digital twins, managing real-time event routing and safety-critical alerting close to the source. Concurrently, vertical data fabrics like Cognite Data Fusion® and AVEVA CONNECT automatically harmonize these multi-vendor streams, cross-referencing historical time-series data with P&IDs and unstructured asset logs into an integrated operational canvas.
Editor’s Note / Breaking Market Realignment: As this research goes to press, Schneider Electric has entered into a definitive agreement to acquire Cognite for $3.1 billion in cash, with plans to align its data platform natively with AVEVA to build a comprehensive data foundation on which AI can be trusted to operate at scale. While Cognite will eventually report within Schneider’s industrial automation division, its core architectural strengths remain highly critical for cross-domain optimization. Read: Schneider Electric to Acquire Cognite to Strengthen Industrial AI Portfolio.
Open Vocabularies: This entire physical pipeline is bound together using open exchange standards like CESMII’s i3X and Smart Manufacturing Profiles. Published across a localized Unified Namespace (UNS), i3X acts as the universal translator, turning raw operational telemetry into structured semantic models that upstream enterprise graphs can instantly parse and leverage without data loss or manual mapping.
The Blueprint for Leadership: Ownership Over Integration
The key takeaway from this month's tech announcements is clear: Databricks can run the analytical brain of your enterprise data lake, but it takes an assembled vertical fabric to manage high-velocity fieldbus telemetry, real-time event brokers, and physical equipment boundaries. Industrial organizations cannot wait around for horizontal software giants to build purpose-built IP for their specific production processes. To achieve sustainable, trusted multi-agent autonomy, industrial buyers must take immediate ownership of their technical foundation:
Build Your Own Context Foundations: Intentionally combine edge DataOps engines, event-driven namespace brokers, and industrial knowledge graphs to secure absolute structural lineage over your operational data.
Enforce Open Interoperability Standards: Mandate that every technology vendor across your ecosystem natively supports open protocols like the Model Context Protocol (MCP) and CESMII i3X profiles to eliminate proprietary lock-in.
Establish Immutable Operational Sandboxes: Implement explicit data quality ingestion firewalls and automated autonomy revocation protocol gates. Ensure that no autonomous agent can execute an operational write-back to an edge device without passing through deterministic safety boundaries and hardcoded Human-in-the-Loop verification controls.
Databricks has delivered an exceptional horizontal foundation for enterprise-wide data engineering and open agentic governance. But on the factory floor, the final mile remains an engineering discipline that belongs entirely to you.
In our upcoming piece, we will dig even deeper into why it is still all about the data. We will lay out the core premise that defines the true leaders of the agentic AI era in the heavy industrial sector: it isn't just about managing data infrastructure pipelines; it is about aggressively driving hard operational and financial outcomes using the petabytes of raw data being generated by oil wells, energy grids, and factories every single day.
Configure the runtime, protect your context, and prepare to monetize the metric. It was, it is, and it always will be all about the data.
Engage with ARC Advisory Group
The Industrial AI (R)Evolution is moving faster than ever. To dive deeper into the frameworks and data shaping the future of the industrial sector, explore my latest research:
Navigating the AI Wars and the escalating Industrial Robot Wars
Closing the Digital Divide by Embracing Industrial AI
Assembling your Industrial-Grade Data Fabric
Charting the new frontier of Physical Intelligence and transitioning to a Cyber-Physical Industrial Architecture (CPIA)
Mapping your maturity and strategy with ARC's 3-Axis Industrial AI Models Taxonomy
Where do you Stand in the Industrial AI (R)Evolution?
Take our Industrial AI Assessment to benchmark your organization's maturity, identify critical gaps in your IT/OT/ET convergence, and get actionable recommendations to accelerate your path to becoming an Industrial AI Pacesetter.
Don't guess what your global operations or prospective customers need. Use empirical data to align your stakeholders and de-hype the market with ARC Advisory Group's Voice of Market Service.