Arborum Labs Bets on AI World Models to Understand India’s Power and Telecom Networks

FutureIsNow Editorial
8 Min Read
Photo by Matthew Henry / Unsplash.

AI is beginning to model the systems that keep economies running

The next frontier in artificial intelligence may be less visible than chatbots and copilots. It sits inside the complex physical networks that power factories, connect people and move essential services. Electricity grids and telecom networks are dynamic systems: demand changes, assets age, weather disrupts operations and failures in one part can cascade into another. Understanding those interactions is a difficult problem—and one that Bengaluru-based Arborum Labs wants to address with AI world models.

The startup has raised an undisclosed pre-seed round co-led by Speciale Invest and Ideaspring Capital. The company was founded by Indian Institute of Science associate professor Himanshu Tyagi and former BCG AI leader Pooja Rajdev. Its initial focus is power and telecom infrastructure, with applications including demand forecasting, operational optimisation and anticipating disruptions.

What is an infrastructure world model?

The term world model is used across AI research to describe systems that learn representations of how an environment behaves and how it may change under different actions or conditions. Applied to infrastructure, the ambition is to represent relationships among assets, demand, operating constraints and external events. Instead of analysing each sensor or asset in isolation, a model could help operators reason about the behaviour of a connected network.

For a power network, that might involve understanding load patterns, asset constraints and the effect of weather or equipment outages. In telecom, it could mean modelling traffic demand, network capacity, equipment performance and the impact of local disruptions. If the model is accurate and calibrated, it may help operators compare scenarios before making decisions.

That is the promise, not yet a proven result. The label world model alone does not establish the underlying architecture, level of physical understanding or predictive performance. A rigorous evaluation must show how the approach differs from existing forecasting, optimisation and simulation tools.

Why power and telecom are a logical starting point

Both sectors operate networks where local events can have system-wide consequences. Power operators need to balance supply and demand while managing equipment, grid constraints and reliability. Telecom operators must allocate capacity, maintain uptime and anticipate changes in traffic and network conditions. In each case, better forecasts and earlier warnings can have economic value if they improve decisions.

The sectors also generate large volumes of operational data. The challenge is that data is distributed across legacy systems, equipment vendors, geographies and business units. A model that performs well on a clean dataset may struggle with missing readings, inconsistent asset records, rare events or changes in operating conditions. Integration with existing control rooms and workflows can be as difficult as model development.

What is known about Arborum—and what remains undisclosed

Arborum says it is already working with multiple nationwide networks, including a steel manufacturer and telecom operators, and plans to release a public version within a year. These are company-reported claims. The organisations have not been named in the available announcement, and the size of the funding round has not been disclosed.

There is no public independent benchmark yet showing prediction accuracy, forecast horizon, uncertainty calibration, production deployment scope or financial impact. Those details matter. A proof-of-concept that identifies patterns after an event is not equivalent to a production system that predicts a disruption early enough for an operator to act.

The validation questions that matter

For infrastructure AI, model performance should be measured against the methods operators already use. Does the system outperform existing statistical forecasts or domain-specific simulation? How far ahead can it predict? Does it communicate uncertainty, especially during rare or extreme events? How often does it generate false alarms, and what is the cost of missing a real failure?

Deployment evidence is equally important. Does the system integrate with operational technology and existing enterprise platforms? Can engineers understand why it flags a risk? Does it recommend actions or execute them? Who approves an intervention? A useful model must fit the safety, regulatory and accountability requirements of critical infrastructure.

The strongest commercial proof would be measurable improvements: fewer unplanned outages, lower maintenance costs, better asset utilisation, reduced energy losses, improved service availability or faster restoration after disruption. These outcomes need to be compared with a credible baseline and measured over enough time to account for seasonality and network changes.

Why investors should pay attention

If world models can represent complex industrial and infrastructure systems reliably, the opportunity extends beyond power and telecom. Manufacturing, logistics, transport, water and energy-intensive facilities all involve networks of assets and dependencies. The addressable problem is large, but so are the barriers: domain expertise, data access, integration, long sales cycles and the need to earn operator trust.

The founding combination of academic research and consulting experience gives Arborum a potentially useful starting point. But the company will need to demonstrate that its approach can move from research capability to repeatable product deployments. Investors should look for named customers, paid contracts, repeat usage and evidence that the model produces decisions operators would not otherwise make.

The FutureIsNow View: the real test is operational impact

Arborum Labs is part of a broader shift from language-centric AI towards systems that attempt to model the behaviour of the physical world. The ambition is strategically relevant because infrastructure networks are too complex to manage through isolated dashboards and static forecasts alone. But the category is still early, and the phrase world model should not be mistaken for proof of a new technical breakthrough.

Watch for a public product release, named customer deployments, independent accuracy benchmarks and measurable operational outcomes. If Arborum can show that its models help operators anticipate and prevent costly disruptions, it could establish a meaningful position in industrial AI. If it cannot demonstrate an advantage over existing tools, the world-model framing will remain more compelling than the product evidence.

The infrastructure thesis connects with FutureIsNow’s analysis of why AI’s next constraint may be the power grid and our coverage of Karnataka’s AI infrastructure ambitions. The next generation of AI will have to understand not only digital information, but the physical systems on which the economy depends.

Sources: ETEntrepreneur’s funding report and ET Enterprise AI’s coverage.

FutureIsNow | Signals over noise. Updated 18 November 2026.

Share This Article