Focus Area
AI & Machine Learning
Applied intelligence rather than general-purpose models. The fund backs systems trained on data a competitor cannot rent, deployed where being wrong has a physical cost — and where the hard engineering sits in the data pipeline, the evaluation harness and the deployment constraint rather than in the model itself.
What we back.
- Industrial autonomy and perception
- Perception and control stacks for machines that operate in the physical world, where the model runs against a latency and power budget at the edge and a safety case sits behind every decision it makes.
- Inspection and condition monitoring
- Vision, acoustic and vibration models that read physical assets — welds, pipe wall, rotating equipment, transmission lines, structures — where labelled failures are rare and expensive and the customer is judged on the ones the system misses.
- Decision and scheduling systems
- Optimisation against real constraints in logistics, dispatch, routing and yard or berth scheduling, where the output has to be explainable to an operator who carries the consequence of accepting it.
- The layer underneath
- Data versioning, evaluation, inference serving and model monitoring — the infrastructure that turns a working prototype into something a regulated buyer will run unattended and audit afterwards.
Why now
Why this sector, in this corridor, now.
Raw model capability is increasingly rented by the token, and the part of the stack that a funded competitor can reproduce in a quarter is growing. What does not commoditise is the data collected by operating a process, the evaluation set that proves the system works on a customer's edge cases, and the integration cost someone has already paid. That is where the fund looks.
The Gulf is building domestic compute and data infrastructure at the same time as its heaviest industries — energy, logistics, ports, utilities — finish instrumenting themselves. Those operators now have telemetry and no decision layer on top of it, and a strong preference for systems that run in-country on data that does not leave. That is a buyer, not a theme.
What we look for
What a company in this area has to show.
The five criteria in thethesisapply to every opportunity. These are the questions this area adds on top of them.
A data asset that cannot be rented
The training and evaluation data comes from operating something, owning a sensor fleet, or a contract that grants usage rights. The test is simple: with the same public weights and a comparable budget, what could a competitor not reproduce in two quarters?
An evaluation harness before a benchmark score
The team can name its failure modes, measure them on held-out data drawn from the deployment rather than from a public set, and say who absorbs the cost of a false positive versus a false negative.
Margins that survive inference
Gross margin calculated after serving cost, with a stated route — distillation, quantisation, caching, smaller task-specific models — for what happens to that margin as usage grows.
An integration someone has already paid for
Running inside a customer's environment, with its data-residency, connectivity and security constraints met, and a signature behind it. A pilot that lives in a sandbox has not tested the part that is hard.
Across the corridor
Why the corridor matters here.
Data residency is the mechanism here. A system engineered to run on infrastructure inside the region, tuned on regional operating data and supported by people in the same time zone, can be bought by industrial and public-sector customers that a US-hosted product cannot serve at all — and that constraint is architectural, so it is decided early or it is expensive. Running the other way, a regional operator with a hard problem and the data to describe it is one of the few things that reliably attracts strong US engineering teams. The fund is built to hold both ends of that: a US company with the deployment discipline to meet the region's constraints, and a regional buyer willing to put a real problem and a real budget behind it.
What that support consists of in practice is set out underCapital + Capability, and how the two markets are covered underPlatform.
Next
Building in AI & Machine Learning?
Founders in the United States and the GCC can submit an opportunity for review, or read the five areas together.