Honeywell Technologies launched Forge Plant Cognition on Monday — an enterprise AI platform purpose-built for refineries, petrochemical plants, and liquefied natural gas operators — with the Dangote Petroleum Refinery in Lekki, Nigeria as its first live deployment, projecting more than $60 million in gross annual operational benefit at the facility's current capacity of 650,000 barrels per day. The platform's optimization target of more than 29 cents per barrel is explicitly described by Honeywell as gross of implementation costs and subject to variability in operating conditions, crude slate, data quality, and market prices.

The Dangote facility is the world's largest single-train petroleum refinery. At 650,000 barrels per day across a full year of operations, Honeywell's per-barrel estimate implies more than $68 million in gross annual impact at the upper bound — making this one of the most specific financial projections attached to an industrial AI platform launch on record, in a sector where up to 95% of AI initiatives historically fail to deliver measurable value, according to analysis of the industrial AI pilot failure rate published by Oil and Gas IQ.

What Makes Physics-Informed AI Different From Conventional Refinery Analytics?

The platform's name is technical, and understanding it explains why Honeywell is positioning Forge Plant Cognition as a qualitative advance rather than a more sophisticated analytics dashboard.

Conventional AI systems applied to refinery operations — predictive maintenance tools, optimization recommenders, process anomaly detectors — learn patterns from historical operational data. When plant conditions stay within the historical envelope, these systems can perform well. When operating conditions move outside that envelope — a new crude feedstock, an equipment change, an extreme ambient event, a market-driven shift in product slate — they extrapolate from patterns that no longer apply, producing recommendations that can be physically implausible or simply wrong. This out-of-distribution failure is the core reason industrial AI pilots collapse when they try to scale.

Physics-informed neural networks (PINNs) address this failure mode through physics-informed neural network training. Instead of optimizing a neural network exclusively against historical data, PINN training adds an additional term to the loss function: the residual of the governing physics equations. For a refinery, this includes fluid dynamics equations (Navier-Stokes for flow behavior), heat and mass transfer equations for distillation and reaction processes, and chemical kinetics equations for catalytic processes such as fluid catalytic cracking — capabilities enabled through governing physics equations and residuals. The network learns to generate outputs that are both consistent with observed operational data and physically plausible — satisfying the same equations that govern the behavior of matter and energy in the physical system, regardless of whether those conditions have appeared in historical data.

The practical consequence is a model that can reason reliably about operating conditions it has never seen, because it is constrained by physics that applies universally. For Honeywell UOP — whose process knowledge reflects a century of refinery design expertise and whose technologies are licensed to more than 6,000 refineries and industrial facilities worldwide — encoding that domain expertise as physical constraints rather than as data patterns is what distinguishes Forge Plant Cognition from generic industrial AI platforms.

The NVIDIA PhysicsNemo open-source framework provides the toolchain for defining governing equations symbolically, computing physical residuals via automatic differentiation, and training neural networks against both data-fitting and physics-consistency objectives simultaneously. Honeywell describes the result as enabling "digital twin integrated agile plant optimization workflows" — the AI proposes, simulates, and validates optimization decisions against a physics-consistent digital representation of the plant before any change is implemented on the physical asset.

Full Technology Stack: PhysicsNemo, Omniverse, Nemotron, Cosmos, and Dell AI Factory

Forge Plant Cognition runs on Dell AI Factory with NVIDIA, supported by Dell Services, integrating four distinct NVIDIA open-model families into a coherent industrial intelligence architecture, as outlined in the NVIDIA and Dell technology stack disclosure accompanying the launch.

NVIDIA Nemotron open models drive the platform's agentic layer — the AI agents that interpret real-time operational data streams and translate analytical outputs into recommended actions for plant operators and engineers.

NVIDIA Omniverse libraries underpin the platform's digital twin capability, providing a full simulation environment for the plant and its processes. The digital twin serves as the validation environment against which PhysicsNemo-trained models are tested before deployment on physical assets.

NVIDIA Cosmos open models, paired with the NVIDIA Metropolis VSS Blueprint, enable a vision AI layer — computer vision agents that process live imagery from plant cameras to detect physical anomalies, generate real-time alerts, and surface equipment issues before they develop into failures.

NVIDIA PhysicsNemo provides the physics-informed training capability described above, enabling construction of models that respect governing physical equations rather than relying exclusively on pattern-matching against historical operational data.

On the infrastructure side, the Dell AI Factory infrastructure components include Dell PowerEdge XE servers and the Dell AI Data Platform for data management and storage, with the system running on NVIDIA RTX PRO GPUs. An expansion to far-edge deployments using NVIDIA DGX Spark is already on the roadmap, meaning refinery operators could eventually run physics-informed inference at the edge of their process control networks.

Dangote's Scale Turns Projection Into Proof Point

The Dangote Petroleum Refinery reached full 650,000 barrels per day capacity — making it the world's largest single-train petroleum refinery by throughput. The Lekki facility was designed to meet 100% of Nigeria's domestic demand for refined petroleum products, with export capacity for surplus output, and currently produces Euro-V quality gasoline, diesel, and jet fuel, along with polypropylene, employing more than 3,000 Nigerian workers.

The Forge Plant Cognition deployment builds on a relationship between Honeywell and Dangote spanning nearly a decade. In April 2026, Honeywell completed its Performance+ Services April 2026 deployment across the refinery's core processing units — combining Honeywell Forge platform tools with UOP engineering specialists to deliver real-time operational insights and operator training simulators built on digital twins of the facility. Forge Plant Cognition represents the next major technological layer in that buildout.

In November 2025, Dangote also selected Honeywell to supply advanced technology, proprietary catalysts, and equipment to achieve a capacity expansion to 1.4 million barrels per day by 2028, up from 650,000 barrels per day today. That agreement includes a license for Honeywell's Oleflex technology, enabling the facility to produce 750,000 metric tons (approximately 826,700 short tons) per year of propylene and expanding total polypropylene output to 2.4 million metric tons (approximately 2.6 million short tons) per year. If the capacity targets are met, the Lekki facility would become the world's largest petroleum refinery by any measure.

David Bird, CEO and Managing Director of Dangote Petroleum Refinery and Petrochemicals FZE, cited operational excellence and Nigerian talent development as the drivers behind the AI deployment. The David Bird Dangote CEO statement issued alongside the announcement read: "This investment reinforces our commitment to operational excellence, advancing technology and developing Nigerian talent to operate at world-class standards."

Physics-Informed AI as Workforce Bridge for Nigeria's Refinery Engineers

Beyond the operational projections, the Forge Plant Cognition deployment addresses a challenge that sets the Dangote facility apart from most comparable global refinery deployments: building a highly skilled engineering workforce largely from the ground up.

Dangote has set a 70% Nigerian workforce development goal — an ambitious target that requires accelerating the development of specialized process engineering expertise that typically takes years of operational experience to accumulate. Forge Plant Cognition functions partly as a knowledge-transfer mechanism: by embedding Honeywell UOP's century of refinery process expertise into AI-powered tools that surface expert guidance in context, the platform enables engineers and operators to make faster, more consistent decisions without first accumulating the years of operational intuition those decisions would otherwise require.

Rajesh Gattupalli, Honeywell UOP president, framed the launch in terms of industry transformation: "By collaborating with Dangote to launch Forge Plant Cognition, we are demonstrating how AI can help transform industrial operations, driving new levels of productivity, reliability and profitability at significant scale."

Forge Plant Cognition Extends Forge Cognition, Honeywell's Broader Agentic Platform

Forge Plant Cognition is a vertical extension of Forge Cognition, a broader agentic AI and simulation platform that Honeywell launched as its Forge Cognition platform twelve days earlier, on October 14, 2026, targeting buildings, campuses, and industrial facilities more broadly. The parent platform supports manual, semi-autonomous, and autonomous operating modes, with operators able to set oversight levels and override the system at any time.

Honeywell is already piloting Forge Cognition across more than 300,000 connected locations globally, including pilot sites including Voyager campus in Santa Clara, the Fairmont Hotel in Dubai, NEX Mall in Singapore, and Vanderbilt University. Early Voyager campus pilot results data include an average reduction in unnecessary alerts of up to 90%, identification of potential energy and utility savings of up to 10%, and an occupant comfort score maintained above 95%.

Forge Plant Cognition applies that architectural foundation with refining-specific physics-informed models layered on top — the additional engineering depth that distinguishes heavy industrial process optimization from building management and campus operations.

What Dangote's Deployment Means for 6,000 Honeywell-Licensed Facilities

The more commercially significant dimension of Monday's announcement may be what the Dangote deployment signals for Honeywell's expansion trajectory.

Honeywell's process technologies and catalysts are licensed to 6,000 licensed refineries worldwide and industrial facilities worldwide. That installed base represents an addressable market of refinery operators already running Honeywell-designed processes — operators currently under mounting pressure to improve margins amid energy transition dynamics, volatile crude markets, and tightening product specifications globally.

For Honeywell Technologies (NASDAQ: HON), which completed the Honeywell Aerospace spinoff June 2026 to become a pure-play automation and process technology company, the Forge platform is central to the company's growth thesis as a standalone entity. Honeywell reported its third-quarter 2026 financial results on October 22, 2026, providing near-term context for how the market evaluates the Forge platform's commercial progress.

For NVIDIA and Dell, the Dangote deployment adds a landmark energy sector reference to the Dell AI Factory and the NVIDIA industrial AI ecosystem — both of which have targeted the multi-trillion-dollar industrial economy as a primary growth market for accelerated computing infrastructure.

Whether physics-informed AI delivers on its projections at the Dangote scale — at 650,000 barrels per day today, and potentially at 1.4 million barrels per day by 2028 — will determine whether the technology becomes the standard approach for refinery optimization globally. At that scale, the Lekki facility will generate the real-world performance data that the rest of the industry will be watching.

Frequently Asked Questions

What is physics-informed AI and how does it differ from conventional industrial analytics?

Physics-informed AI — specifically physics-informed neural networks (PINNs) — trains neural networks to satisfy the governing equations of physical systems as a constraint on their outputs, alongside learning from historical operational data. Governing equations for refinery applications include Navier-Stokes (fluid flow), heat and mass transfer (distillation and reaction), and chemical kinetics (catalytic processes). Conventional industrial AI learns only from historical data; when operating conditions fall outside that historical envelope, those systems can produce physically implausible or simply incorrect recommendations. Physics-informed models are constrained by universal physical laws, making them more reliable in novel operating conditions. NVIDIA's open-source PhysicsNemo framework provides the toolchain for this approach.

What is the $60 million annual benefit projection based on, and what are the caveats?

Honeywell's projection of more than $60 million in gross annual operational benefit at the Dangote facility is based on an optimization estimate of more than 29 cents per barrel at 650,000 barrels per day. Honeywell explicitly describes this figure as gross of implementation costs, forward-looking, and subject to variability in operating conditions, crude slate, data quality, and prevailing market prices. The figure represents a ceiling estimate under favorable conditions, not a guaranteed contractual outcome. Real-world results at scale will depend on factors including actual crude throughput, the product slate being optimized, and how broadly the platform is deployed across the facility's interconnected process units.

Can other refineries and industrial operators access Forge Plant Cognition?

Yes. Forge Plant Cognition is designed as an enterprise platform for refinery, petrochemical, and LNG operators broadly. Honeywell's process technologies are licensed to more than 6,000 refineries and industrial facilities globally, representing the platform's primary near-term addressable market. The broader Forge Cognition platform, of which Forge Plant Cognition is a vertical extension, is already deployed across more than 300,000 connected locations globally.

Why does the Dangote refinery matter as a first deployment?

The Dangote Petroleum Refinery in Lekki, Nigeria is the world's largest single-train petroleum refinery, with current capacity of 650,000 barrels per day and a planned expansion to 1.4 million barrels per day by 2028. Its scale and operational complexity — including the challenge of building a workforce that is 70% Nigerian — make it a credible and demanding first reference deployment for enterprise AI performance claims. A platform that proves viable at Dangote's scale carries substantially more industry credibility than a pilot at a smaller or less complex facility, and Honeywell's decade-long relationship with the site provides the integration depth that AI deployments at this level require.