- Dangote Refinery Turns to Industrial AI in Push for US$60 Million Annual Gains
Dangote Petroleum Refinery is deploying an artificial intelligence platform developed by Honeywell Technologies with computing capabilities from NVIDIA and Dell Technologies, in a move that could generate more than US$60 million in annual gross operational benefits when the facility reaches full capacity.
The Lekki-based refinery will be the first industrial facility to deploy Honeywell’s Forge Plant Cognition platform, which is designed to analyse operating data, identify production constraints, detect emerging equipment problems and recommend measures to improve throughput, reliability and profitability.
Honeywell estimates that the system could deliver operational benefits exceeding US$0.29 for every barrel processed at the refinery. At full-capacity throughput, that would translate into more than US$60 million annually.
The company has stressed that the estimate is preliminary, facility-specific and not guaranteed. It also excludes the cost of implementing the platform, meaning the eventual net financial benefit could be materially lower. Actual results will depend on data quality, utilisation, crude selection, product margins and the efficiency with which recommended actions are implemented.
Despite those qualifications, the deployment represents an important moment for African industrialisation. Artificial intelligence is frequently discussed on the continent in relation to financial technology, public services and consumer applications. Dangote’s initiative shifts the focus towards the factory floor, where even a small improvement in efficiency can generate substantial value at scale.
A refinery is one of the most complex industrial environments in operation. Thousands of variables covering temperature, pressure, flow rates, energy consumption, equipment condition and product quality must be monitored simultaneously.
A change in one part of the plant can affect several interconnected units. An equipment failure can reduce production, damage other assets or force an unplanned shutdown. Meanwhile, an apparently minor adjustment to operating conditions can alter the quantity and value of products recovered from each barrel of crude.
Forge Plant Cognition seeks to bring these variables into a single analytical system.
Unlike conventional monitoring software that presents operators with data or generates alerts, the platform combines real-time operating information with Honeywell’s engineering knowledge, process calculations and physics-based models.
Its purpose is not merely to tell engineers that something has changed. It is intended to explain the likely operational significance, simulate possible responses and recommend actions.
Honeywell identifies three areas in which the platform is expected to deliver value: capturing hidden refining margins, protecting production and extending specialist knowledge across the workforce.
Capturing hidden margin involves identifying bottlenecks and improving throughput, product yield and value across connected refining processes.
Protecting production means detecting equipment or process problems before they cause significant disruption. Scaling expertise involves embedding specialist engineering knowledge into AI-supported workflows so that operators can make faster and more consistent decisions.
The distinction is significant. The most valuable industrial AI systems will not necessarily replace workers. They will help engineers interpret an overwhelming volume of plant data while leaving safety-critical decisions subject to professional judgement and established control procedures.
The platform combines technologies from three companies with distinct roles.
Honeywell contributes industrial automation, refinery-process knowledge and its UOP engineering expertise. NVIDIA supplies AI models, accelerated computing and simulation tools, while Dell provides the computing, storage and associated infrastructure required to deploy the system.
NVIDIA’s Nemotron open models will power agents capable of interpreting operating data. Omniverse libraries will support the creation of a digital twin—a virtual representation of the refinery and its processes that can be used for simulation, training and planning.
NVIDIA Cosmos models, working with its Metropolis video technology, will support visual monitoring and the detection of anomalies. The PhysicsNeMo framework will enable physics-based AI models that can simulate industrial processes and assess potential changes before they are introduced into live operations.
This is where the project moves beyond the fashionable use of generative AI.
A language model can summarise an operator’s report, but refinery optimisation requires the system to understand physical limits, engineering relationships and the consequences of changing operating conditions.
A recommendation that ignores pressure limits, equipment design or chemical interactions could be ineffective or dangerous. By combining AI with physics-based modelling and established process calculations, Honeywell is attempting to create a system grounded in the actual behaviour of the plant.
The digital twin could allow engineers to test proposed adjustments virtually before applying them to physical equipment. This can lower the risk associated with experimentation and improve training for complex or unusual operating scenarios.
The estimated US$60 million annual benefit is substantial in absolute terms, but Dangote Refinery’s enormous scale explains why a small per-barrel gain can produce such a large figure.
Honeywell calculates the benefit at more than US$0.29 per barrel when the refinery operates at full capacity. Dangote is working with Honeywell to increase capacity from 650,000 barrels per day to 1.4 million barrels per day within three years.
If throughput reaches 1.4 million barrels daily, an improvement of only a few cents per barrel accumulates rapidly. The economics demonstrate why industrial AI may deliver its greatest returns in large, continuous-process facilities.
But the estimate should not be confused with guaranteed profit.
Honeywell describes the US$60 million as a gross operational benefit. Implementation expenses, data preparation, cybersecurity, infrastructure, staff training, licensing and maintenance will reduce the amount retained by the refinery.
The figure also assumes full-capacity operations. If the plant processes fewer barrels because of crude-supply constraints, maintenance, market conditions or logistical bottlenecks, the aggregate benefit will be lower.
The platform’s commercial success should therefore be measured through disclosed results: reductions in unplanned downtime, increases in throughput, improvements in product yield, lower energy consumption and faster resolution of operating problems.
Without measurable baselines, industrial AI risks becoming an expensive layer of technology whose financial value is difficult to separate from broader operational improvements.
The public focus may fall on the US$60 million estimate, but the platform’s greatest value could lie in protecting production.
At a refinery of Dangote’s scale, an unplanned shutdown can interrupt fuel supplies, delay exports and affect prices across several markets. Lost production can quickly cost more than incremental improvements in daily margins.
Predictive systems that identify equipment deterioration before failure can allow maintenance to be scheduled during a controlled period rather than after an emergency.
This has wider implications for Nigeria. The Dangote Refinery has become strategically important to the country’s ambition to reduce dependence on imported petroleum products and earn foreign exchange through refined-fuel exports.
Its reliability therefore affects more than the company’s earnings. It influences fuel availability, shipping flows, import demand and confidence in Nigeria’s ability to operate complex industrial infrastructure.
The same principle applies across Africa. Large industrial projects are often assessed mainly according to their construction cost and nameplate capacity. Far less attention is paid to maintenance systems, operational discipline and the technical workforce required to keep them functioning.
AI cannot substitute for those foundations, but it may help strengthen them.
Dangote says the platform will contribute to its objective of building a workforce that is 70% Nigerian.
David Bird, chief executive and managing director of Dangote Petroleum Refinery and Petrochemicals, said the investment would support operational excellence, technological advancement and the development of Nigerian talent capable of operating at international standards.
“We chose Honeywell Technologies’ Forge Plant Cognition because the platform can deliver unrivalled industrial expertise combined with differentiated AI capabilities to help us optimise performance, unlock greater value from our operations and evolve how we operate our assets,” he said.
This skills dimension could prove more important for Africa than the software itself.
Many complex industrial facilities depend heavily on expatriate expertise during commissioning and early operations. If AI-supported workflows can make specialist knowledge more accessible to local engineers, shorten troubleshooting times and improve training, they could reduce that dependence.
There is, however, a risk that the technology creates a new form of dependency if the underlying models, infrastructure and process knowledge remain controlled externally.
For genuine technology transfer, Nigerian engineers must understand how the system reaches its conclusions, test its recommendations and retain the ability to operate safely when the platform is unavailable.
AI should expand local technical capability, not turn engineers into passive recipients of instructions generated elsewhere.
Connecting refinery data to advanced AI infrastructure also increases the importance of cybersecurity.
A refinery is critical infrastructure. Its operating information, equipment systems and production schedules are commercially sensitive and potentially attractive to cybercriminals or hostile actors.
Industrial AI requires access to large volumes of real-time data. That creates potential entry points which must be secured through network separation, access controls, continuous monitoring and clearly defined responsibility among Dangote, Honeywell, NVIDIA and Dell.
Data governance will also matter. The parties must establish who owns the operational data, where it is processed, how long it is retained and whether it can be used to train models deployed for other customers.
The intelligence generated by the refinery could become a valuable commercial asset. Protecting it should be treated as part of refinery security rather than a conventional information-technology exercise.
The choice of the Lekki refinery as the platform’s first deployment is strategically important for Honeywell.
Dangote offers an unusually large and complex operating environment in which the company can demonstrate whether its combination of industrial expertise, AI and digital twins delivers measurable results.
Success could create a model for deployment across refineries, petrochemical plants and liquefied natural gas facilities elsewhere.
Honeywell’s relationship with Dangote already extends beyond the AI platform. It is supporting the planned expansion of the Lekki refinery and has been selected to assist with Dangote’s proposed refinery in Kenya.
It is also providing technology for the Nigerian group’s expansion into petrochemicals, including additional propylene and linear alkylbenzene production used in plastics, detergents and other manufactured goods.
This creates an opportunity to apply the lessons from Lekki across a wider industrial network. It also concentrates technological dependence on a relatively small group of global suppliers.
Dangote will need to ensure that integration delivers efficiency without limiting its ability to change vendors, upgrade systems or retain control over core operations.
The deployment challenges a common assumption that advanced AI will reach Africa principally through consumer applications imported from overseas.
Some of the continent’s most valuable AI use cases may instead be found in mines, refineries, power plants, ports, farms and manufacturing facilities.
In these sectors, value is created not by generating text or images but by reducing downtime, improving energy efficiency, predicting failures and increasing the output obtained from existing assets.
For Dangote, the immediate objective is commercial: produce more reliably and protect refining margins as the business expands.
For Nigeria, the stakes are broader. The project could demonstrate that African industrial infrastructure can be both large and technologically sophisticated, while providing local engineers with exposure to advanced automation and physical AI.
But the success of the initiative should be judged by outcomes, not the prominence of its technology partners.
If Forge Plant Cognition improves reliability, transfers meaningful skills, protects sensitive data and delivers benefits greater than its cost, it could become a reference point for African industry.
If the promised gains remain projections, it will illustrate another lesson: artificial intelligence cannot rescue operations that lack strong data, disciplined maintenance and skilled people.
Dangote Refinery is not merely installing new software. It is testing whether Africa can use AI to operate industrial assets more intelligently—and retain more of the value those assets create.
