Startup Spotlight: Invertix builds energy agents and leaves physical control to operators

Why it matters

Invertix is betting that energy AI wins by connecting operational data and coordinating work, with human approval preserved for physical actions. Its gigawatt figures show reach, while savings and commercial depth remain the tests of value.

When Joseph Perrotta and Kaan Durmaz presented Invertix to Y Combinator on September 18th, they pitched a specific constraint on artificial intelligence: energy infrastructure still depends on people navigating disconnected operational systems. Invertix is building agents to connect those systems, investigate problems and coordinate the work that follows. Invertix's YC profile says its agents are deployed across about 2 GW of energy assets.

Perrotta's path to energy software ran through real estate sales, professional volleyball and Cubbit. The YC profile describes him as a former professional volleyball player who sold more than $17 million of houses at age 18, then worked at Cubbit, an Italian cloud-storage company that YC says raised $25 million. That mix of sales experience and startup exposure sits alongside Durmaz's technical background in computer vision, differential privacy and AI for chemistry, as described in their YC profiles.

The founders' pitch starts with field operations, not a general-purpose chatbot. Invertix aims to connect data from systems such as SCADA, maintenance-management software, work orders, weather feeds and technical documents, then give specialized agents enough shared context to spot anomalies, investigate causes, prepare reports and route tasks to the relevant people. The bet is that energy operators lose time stitching together records and coordinating work across separate tools, even when the underlying information already exists.

A founder's search for a problem operators would pay to solve

Perrotta studied economics at the University of Bologna and started a student venture-capital initiative before taking a role at Cubbit, according to La Repubblica's profile of the founders. He later traveled between startup centers including San Francisco, London, Berlin and Paris. He met Durmaz through Illegible, a fellowship for neurodivergent talent. The two became roommates, built products together and eventually left their master's programs to start Invertix, La Repubblica reported.

The founders spent time testing ideas with prospective users before settling on energy operations. La Repubblica reports that they contacted more than 5,000 energy companies and spoke with about 500 people in the sector. The conversations pointed them toward a problem less glamorous than inventing another AI model: renewable-energy portfolios are expanding, while teams still have to reconcile plant telemetry, maintenance records, procedures and financial information by hand.

Invertix says it can map assets, records, procedures and people into a shared operational model, then run agents against that context. Its website shows examples such as investigating a solar-plant fault, assembling relevant maintenance history and coordinating a work order. Invertix also presents a software-building layer for creating internal tools and workflows on top of the operational data.

Invertix's claim that its models use physics-informed neural networks and differentiable simulators adds a technical distinction to its pitch, though the YC profile does not specify which workflows use those methods or publish comparative performance results. The system is designed to pass work between software and field teams rather than stop at an alert or dashboard.

Perrotta described the boundary in a June interview with pv magazine: detection, analysis and drafting are agent tasks, while decisions and execution remain with people. The report says Invertix agents require human sign-off before action. The founders also say the agents do not directly operate turbines, breakers or other physical equipment. For customers responsible for critical infrastructure, a system that prepares and coordinates an intervention is a different product from one that controls the asset itself.

The gigawatts describe reach, not revenue

Invertix's public traction figures have grown over several months, but they measure different things. In May, investor Vireo Ventures reported that Invertix managed more than 1.8 GW of solar capacity through its AI workers and had commercial opportunities representing more than 10 GW. The September YC profile puts autonomous deployment at about 2 GW and says Invertix is working with partners that have more than 50 GW of energy assets. It also says Invertix is signing two large utilities and has interest from more than 30 energy companies.

Those figures should not be collapsed into one measure of customer adoption. The 2 GW figure is presented as deployment; the 50 GW figure refers to assets associated with partners. The profile does not explain how much of that partner capacity is under contract, active in production or accessible to agents. Interest from 30 companies and discussions with utilities also do not establish paying customer counts. None of these measures, on its own, shows revenue, retention, customer savings or the size of a typical contract.

Vireo said it led a 1.7 million euro pre-seed round announced on May 19th, with participation from Italian Founders Fund and angel investors active in energy and AI. The round predates the YC profile by four months; it is part of Invertix's financing history, not new capital attached to its September batch listing. No valuation was provided in the financing announcement.

The energy bottleneck cuts both ways

Invertix's wider thesis is that AI growth depends on energy, and that energy production itself needs better operational software. The demand figures in Invertix's YC profile have an independent reference point: the International Energy Agency estimates that data centers used about 415 terawatt-hours of electricity in 2024 and projects roughly 945 TWh by 2030 in its base case. The IEA calls that a substantial increase, while also noting uncertainty in future demand and the long lead times involved in building energy infrastructure.

That context supports the premise that electricity supply and infrastructure operations matter to AI's expansion. It does not prove Invertix's more ambitious language about making AI run at zero energy cost. Invertix's operational software may help teams find problems sooner or coordinate maintenance more efficiently; it does not remove the electricity required by data centers, and Invertix's public materials do not quantify energy savings from deployments.

The market already contains software for monitoring renewable assets and managing industrial data. A June comparison in pv magazine names SenseHawk and Envision Digital as platforms focused on monitoring, analytics and workflow. The same report describes Areg.AI's more execution-oriented approach, which includes a robotics layer for field operations. Invertix is positioning itself around the work between data integration and field operations: connect records, reason across them, prepare a response and hand it to the right operator. Customers must trust its data connections, understand why an agent reached a conclusion and know exactly which steps still require approval. Invertix's website says access, hosting and retention are agreed with customers' technical teams, and that workflow permissions define when agents must involve a person.

Perrotta's background is in sales and early-stage company building; Durmaz brings research and applied AI experience. Their early customer conversations pushed Invertix's product toward the practical limits of operator capacity, rather than a promise that a model can independently run an energy company. The founders have also raised a modest pre-seed round from investors with an energy focus, giving them capital to develop and sell into a sector where deployments require integration work and trust.

Invertix is now trying to prove that its agents can earn a place in those workflows. The headline figure of about 2 GW offers a measure of the assets Invertix says it reaches, while the larger partner figure suggests a route to expand. The decisive evidence will be what operators let the system do repeatedly, what work it takes off their hands and whether the resulting time or production gains justify the cost. For now, Invertix's operational case is narrower than its 'superintelligence' label: give energy teams a shared operational picture, automate the investigation and coordination around recurring problems, and leave physical authority with the people responsible for the plant.