Predictive maintenance has become the default example whenever manufacturers are asked how they use AI. It is easy to understand and easy to measure: watch a machine, identify the pattern that precedes failure and intervene before downtime becomes expensive.
A newer group of industrial AI companies is making a broader claim. Instead of modelling one asset or one failure mode, they want software to reason across the plant. Applied Computing, a London startup founded in 2023, raised a $20m Series A in July led by engineering group KBR, with Databricks Ventures participating. Its target market includes oil, gas, refining and petrochemicals, where a facility may generate thousands of sensor streams alongside years of engineering documentation.
The bottleneck is not a lack of sensors
Modern industrial sites already collect large amounts of temperature, pressure, flow, vibration and equipment data. Applied Computing's chief executive has argued that operators make decisions using less than 8% of the information available to them. That figure is a company claim, but the underlying problem is familiar across manufacturing: data exists in different systems, formats and teams, and turning it into one operational picture is difficult.
Traditional predictive-maintenance systems narrow the problem deliberately. They may monitor a pump, compressor or production line and look for anomalies associated with failure. A whole-plant model attempts to connect those signals with process chemistry, engineering documents and the physical relationships between pieces of equipment.
Why physics matters in industrial AI
A consumer AI product can sometimes tolerate a plausible but imperfect answer. A refinery cannot. Industrial operators need to know whether a recommendation respects pressure limits, process dependencies and engineering constraints before anyone acts on it.
That is why Applied Computing describes its model as physics-grounded. The commercial challenge is not simply to generate a convincing explanation of plant behaviour. It is to produce outputs that engineers can verify against the real system and trust enough to incorporate into operational decisions.
Predictive maintenance is still the entry point
The broader ambition does not make predictive maintenance obsolete. In practice, maintenance remains one of the clearest ways for a factory to justify an AI investment because avoided downtime has a visible financial value. British Business Review's reporting on UK manufacturers adopting AI found predictive-maintenance usage rising quickly while skills remained a major constraint.
Whole-plant models may expand from that foothold. If a company can prove that a system detects equipment degradation reliably, it earns the credibility to address energy optimisation, throughput, process stability and operating decisions that involve several assets at once.
The KBR relationship matters more than the funding headline
KBR is not only an investor. Reporting around the round describes a multi-year development relationship around industrial AI. For a young company selling into energy and chemicals, access to engineering expertise and real operating environments can matter more than the branding value of a venture round.
Industrial software usually fails less spectacularly than consumer apps and more expensively. Integrations take time, historical plant data is messy, and customers expect systems to work alongside existing controls rather than replace them overnight. A credible engineering partner can shorten the distance between a promising model and something an operator will actually use.
What UK manufacturers should learn from the shift
The immediate lesson is not that every factory needs a foundation model. It is that manufacturers should avoid treating AI projects as isolated demonstrations. The strongest use cases increasingly depend on common data definitions, access to engineering records and a clear way to connect operational signals across systems.
A plant that cannot combine its existing maintenance, sensor and process information will struggle to benefit from more ambitious AI later. Predictive maintenance is therefore useful for two reasons: it can save money now, and it can expose whether the underlying industrial data architecture is ready for the next generation of models.