The Agreement Aiming to Take Energy AI Out of the Lab
A phrase that has echoed in nearly every energy conference for the past three years is: "We are moving out of the pilot phase." The frequency with which this is stated is, in itself, diagnostic. If so many players feel the need to proclaim it, it’s because the issue persists. AI in the energy sector remains trapped in controlled tests that do not scale into operational flows where money is made or lost.
On March 31, 2026, Applied Computing—the British firm behind Orbital, its physics-based AI platform—announced a strategic alliance with Wipro and Databricks. The declared goal: to help energy operators in the Middle East, India, and Southeast Asia embed verifiable and explainable AI directly into their daily workflows. This news circulated like yet another technological agreement. I read it as a deliberate attempt to address an architectural flaw that the sector has been grappling with since it began experimenting with machine learning models.
Why AI Pilots Fail to Transition to Operations
Before evaluating whether this alliance has the right pieces, it’s important to understand why the problem it seeks to address exists. An energy infrastructure operator—a refinery, a gas distribution network, a power generation plant—does not operate under the logic of a tech startup. They operate under physical, regulatory, and safety constraints that leave no room for ambiguity. When an AI model recommends adjusting pressure in a pipe or redistributing load in a substation, the operator needs to know not only what the model recommends but also why, and under what physical assumptions it reached that conclusion.
This is the knot that has blocked mass adoption. Most deep learning models operate like black boxes: they optimize based on statistical patterns without anchoring their recommendations in verifiable physical laws. In an industry where an incorrect decision can cost lives, assets worth hundreds of millions of dollars, or severe regulatory penalties, this is not a minor limitation. It is an almost insurmountable barrier to entry.
Applied Computing positions Orbital as a direct response to that barrier. The so-called physics-informed AI models integrate domain equations—thermodynamics, fluid mechanics, electric network dynamics—within the model architecture. The theoretical outcome is a system whose outputs are auditable: you can trace the recommendation back to the physical principle that supports it. This turns AI from a black box into something more akin to an engineering blueprint with traceability.
The Logic of Specialization in Three Layers
Where I find this agreement structurally sound is in the division of labor it proposes. This is not an alliance between equals competing for the same portion of the contract; it's a three-layer architecture where each piece resolves a different bottleneck.
Applied Computing contributes the modeling layer: Orbital as the foundational platform with models trained on energy operations physics. Databricks provides the data and infrastructure layer: the ability to move, process, and manage the volumes of operational information that a refinery or power grid generates in real time. Wipro adds the implementation and institutional trust layer: decades of relationships with industrial operators in the target geographies, knowledge of local processes, and the ability to translate a model’s recommendation into a standard operational procedure change.
This detail interests me most from a business model perspective. Selling AI to critical infrastructure does not close in a data room; it closes in the operations room, with the shift supervisor who has spent twenty years reading gauges. Wipro has access to that room. Applied Computing, on its own, likely does not. Thus, the alliance is not just a distribution agreement; it is the acquisition of institutional credibility without the need to build it from scratch, which would have taken five to ten years and a network of relationships that cannot be bought with venture capital.
The geographical specialization is also not accidental. The Middle East, India, and Southeast Asia present a specific combination of conditions: aging energy infrastructure with a high need for modernization, increasing regulatory pressure on emissions, and appetite for solutions that do not require replacing physical assets but rather optimizing them. These markets are where the argument for reducing operating costs and extending asset lifespan carries more immediate weight than abstract narratives of digital transformation. This is precisely the type of proposal that a physically grounded model can support with verifiable numbers.
The Risk the Alliance Cannot Ignore
Although the alliance's architecture has internal logic, there’s one variable that no press release can resolve alone: the quality and availability of operational data in the environments where Orbital will be deployed.
Physics-informed models are more robust than purely statistical models in the face of scarce data, but they are not immune to mislabeled data, miscalibrated sensors, or historical gaps in operational records. In the energy infrastructure of emerging markets, these conditions are not exceptions; in many cases, they are the norm. A plant built in the 1980s with control systems updated in successive patches may have a data history that is, in technical terms, a puzzle with missing pieces.
Databricks addresses part of this problem in the data integration and governance layer, but it does not resolve the source quality. A successful implementation in these environments will require prior work on auditing and cleaning data that Wipro will need to execute before Orbital can generate reliable recommendations. This work has a real cost, consumes time, and is where promises of rapid deployment often fracture against operational reality.
This doesn’t invalidate the alliance’s thesis; it makes it more honest. If the three parties have accurately sized that prior effort in their implementation models—and have built pricing structures to absorb it or pass it onto the customer—then the model is viable. If they assumed that the data would be ready for consumption from day one, they are underestimating the cost of the first piece of the system.
The Flaw That No Press Release Mentions
There’s something that strategic agreements of this type rarely declare in their announcements: the cost of changing human behavior. Embedding AI into operational workflows is not a software issue; it is an organizational adoption problem. The operator who receives a recommendation from Orbital needs to trust it enough to act but not so blindly as to stop using their professional judgment when the model is wrong.
That calibration between trust and oversight is the hardest to build, and there is no technology platform that will install it automatically. It requires training, iteration, and time. Companies that have successfully scaled AI in industrial operations did not do so because their model was more accurate; they achieved it because they designed the human adoption process with the same rigor with which they designed the model. Wipro, as an integrator experienced in change management, is the piece that theoretically covers that gap. Execution will tell if it was enough.
The alliance between Applied Computing, Wipro, and Databricks has a coherent layered architecture and addresses a verifiable market problem. Its strength will not be proven in the announcement, but in how many operators, within eighteen months, will have transitioned from a signed contract to a running model in production with auditable metrics. Companies are not distinguished by the quality of their strategic agreements but by their ability to ensure that each part of the system delivers what it promised when the client needs it most.
