The pitch sounds almost too neat: run building simulations a thousand times faster than the old way, cut energy waste, and do it all with physics-baked neural networks that know how air actually moves through a room. Inviscid AI, fresh from Y Combinator's Winter 2026 batch, is hardly the first startup to promise smarter buildings. But the San Francisco outfit—really just two founders and a whiteboard full of computational fluid dynamics equations—thinks it has an edge.
That edge, they say, is physics. Not just pattern recognition pulled from mountains of historical data, but fundamental laws embedded directly into the machine learning itself.
Whether that claim holds up in practice is what Kabir Jain and Ziming (Qiu) Ziming now need to prove. And they're starting in an unlikely place: a government data center in central India.
When Buildings Are the Problem
The timing, at least, makes sense. Buildings gobble up roughly 30% of global energy and account for 26% of energy-related emissions, according to the International Energy Agency. U.S. data centers alone consumed about 183 terawatt-hours in 2024. Gartner projects that number will more than double by 2030 as AI infrastructure sprawls across the landscape, server rack by server rack.
Someone has to cool all those chips. And cooling, as it happens, is expensive—both financially and thermally.
Inviscid's platform runs physics-informed neural networks, a technical approach that enforces physical constraints (think: conservation of energy, fluid dynamics) while accelerating simulations that would normally take weeks down to seconds. The system pulls data from thousands of IoT sensors and building management telemetry to construct what the company describes as a real-time digital twin—a virtual replica that mirrors a facility's thermal and airflow behavior as conditions shift.
That twin then loops back into building management systems, continuously adjusting HVAC settings, lighting, and energy setpoints. It's closed-loop control, running 24/7, without ripping out existing infrastructure.
The company claims it can optimize HVAC vent placement 240 times faster than conventional methods and improve airflow by 40%. Those figures, though, come from internal benchmarking. Third-party validation remains forthcoming.
A Pilot in Madhya Pradesh
In February 2026, Inviscid signed a memorandum of understanding with the Madhya Pradesh State Electronics Development Corporation to deploy a proof-of-concept digital twin at the Madhya Pradesh State Data Center. The pilot will focus on real-time thermal simulation, hotspot detection, and predictive monitoring—essentially stress-testing whether the technology works outside a controlled demo environment.
"Data centers spend 30%+ of their energy on cooling," Jain said when announcing the partnership. It was a statement of the obvious, perhaps, but also a reminder of where the opportunity sits.
The deployment marks Inviscid's first publicly disclosed customer, arriving roughly a month after the founders emerged from Y Combinator's accelerator program. For a two-person team, moving from demo to government pilot that quickly is notable. Whether it scales beyond one data center in Bhopal is another question entirely.
A Market Already Full of Contenders
Inviscid is hardly alone in chasing building optimization dollars. PassiveLogic raised $74 million in a Series C round in September 2025, backed by NVIDIA nVentures, and offers its own physics-based digital twins with autonomous control. BrainBox AI says it manages thousands of buildings with machine learning-driven HVAC systems, claiming energy cost reductions between 25% and 35% in case studies.
For data center cooling specifically, Vigilent has established deployments showing 20% to 40% cooling energy savings. And the industry-wide benchmark remains Google's DeepMind project, which starting in 2016 achieved up to 40% reductions in cooling energy across multiple data centers—a result that turned heads and validated the broader concept.
Then there are the incumbents. Johnson Controls acquired edge AI specialist FogHorn in 2022, folding those capabilities into its OpenBlue platform. When legacy building automation giants start buying AI startups, the category has arrived.
So what's Inviscid's angle?
The founders lean on their physics-first architecture. While some competitors rely primarily on pattern recognition from historical data, Inviscid embeds physical laws directly into the neural networks—at least in theory, this should make predictions more reliable when conditions shift outside past experience. It's a compelling technical story.
The practical question, though, is whether that architectural choice translates to materially better performance when the system is managing actual airflow in a live building on a hot afternoon. The India pilot may start to answer that. Or it may simply confirm what the market already knows: building optimization is hard, and every approach has trade-offs.
Proving Ground
Inviscid lists operations in San Francisco and Singapore but hasn't disclosed funding beyond its Y Combinator participation. The startup is lean by necessity and design—two founders trying to move fast in a space where energy costs keep climbing and grid constraints are tightening around data center expansion.
There's genuine urgency here. Utilities are balking at new data center connections in Virginia and other AI infrastructure hubs. Power-hungry facilities need to squeeze more efficiency out of every kilowatt-hour, and they need it now.
But urgency doesn't guarantee success. Inviscid's challenge is the same one facing every startup in this market: moving from claims to verified, repeated results at scale. Internal benchmarks are fine for pitch decks. Customers want proof that works in production, under messy real-world conditions, day after day.
The Madhya Pradesh pilot is a start—perhaps more than the founders expected to secure this early. What happens next will determine whether Inviscid becomes another cautionary tale in the crowded building tech graveyard or whether those physics-informed neural networks actually deliver something the market hasn't seen before.
For now, the two-person team has a government data center in India and a lot to prove.
