The numbers tell a blunt story: buildings consume around 40% of U.S. energy and 74% of the electricity produced annually, and in data centers—those humming warehouses of computation—cooling systems alone devour another 30 to 40% of the power budget. Now multiply that by the surge in AI workloads, and you have what industry veterans call a "compounding nightmare."
Enter Inviscid AI, a two-person startup that emerged from Y Combinator's Winter 2026 cohort with a characteristically bold claim: it can simulate airflow and heat transfer in buildings fast enough to adjust HVAC systems continuously, not just during design reviews. The founders, Kabir Jain and Ziming Qiu, say their physics-informed neural networks run roughly 1,000 times faster than the computational fluid dynamics tools that engineers have relied on for decades.
Whether that speed advantage translates into real energy savings—and whether building managers will trust a startup's algorithms to control their facilities—remains an open question.
Fast Physics, Real-Time Control
Inviscid went public in early March with a pitch that sounded straightforward, perhaps deceptively so. Replace the slow, batch-style simulations that facilities teams run once or twice during a building's design phase with a continuous digital twin. Feed it real-time data from sensors scattered across a structure. Let it adjust thermostats, dampers, and airflow on the fly as occupancy shifts, weather changes, or equipment cycles on and off.
Traditional building energy models can take hours, sometimes days, to simulate how air moves through a conference room or a data hall. That lag has confined them mostly to pre-construction planning or the occasional post-occupancy audit. Inviscid's approach—leaning on what it describes as GPU-accelerated solvers and neural operators—collapses that timeline to near-instantaneous. The system, according to the company's March 5 launch materials, ingests streams from building management systems and continuously models thermal loads and energy flows.
At the heart of it: Navier-Stokes equations, the mathematical bedrock of fluid dynamics, solved not through brute-force computation but through machine learning models trained to respect physical laws. Inviscid claims "state-of-the-art performance on benchmarks," though it hasn't specified which benchmarks or published peer-reviewed validation. The company's website cites 95% accuracy and the ability to process thousands of sensor feeds around the clock.
The promised outcome? Energy reductions of 15 to 30%, per Inviscid's own projections. Those figures haven't been verified by independent auditors. A LinkedIn post from the company mentions one customer—unnamed—seeing "2x cooling efficiency gains" in a data center. No documentation or third-party confirmation accompanies that anecdote, a gap that any facilities manager evaluating the technology will likely notice.
Two Markets, One Urgency
Inviscid is aiming at two energy-intensive sectors. The first is the sprawling world of commercial and industrial buildings: offices, hospitals, warehouses, cleanrooms. HVAC optimization in these environments isn't new, but the startup argues that real-time control informed by continuous simulation could push savings further than the static schedules or rule-based systems many facilities still use.
The second target may prove more pressing. Data centers are facing an electricity crunch. Gartner has projected that demand from these facilities will roughly double by 2030, propelled in large part by AI compute. Cooling already accounts for 30 to 40% of a typical data center's power draw—industry figures that have held stubbornly steady even as efficiency efforts have advanced elsewhere. Real-time thermal modeling, if it works as advertised, could help operators pinpoint hotspots before hardware throttles, optimize airflow across racks, and plan capacity expansions with more precision.
In late February or early March, Inviscid signed a memorandum of understanding with the Madhya Pradesh State Electronics Development Corporation to pilot its digital twin at a government data center in India. The agreement, announced on LinkedIn, outlines a proof-of-concept focused on thermal simulation, hotspot detection, and predictive monitoring. It hasn't moved beyond pilot stage yet, and details on timelines or performance metrics remain sparse.
The Competition Isn't Standing Still
Inviscid didn't invent the idea of applying AI to building operations. The field has gotten crowded, particularly in the past two years.
Trane Technologies acquired BrainBox AI in January 2025, pulling an HVAC AI controls platform—one that has claimed 15 to 25% energy savings—into the portfolio of a major equipment manufacturer. PassiveLogic pitches what it calls a "quantum digital twin" for autonomous building control, a branding choice that raises eyebrows among some engineers but reflects the sector's taste for ambitious rhetoric. Siemens rolled out Building X in late 2025, an AI-based lifecycle twin for operations. Schneider Electric's EcoStruxure suite has long included simulation and digital twin features for both buildings and data centers.
On the computational side, SimScale and Altair have embedded AI into traditional CFD workflows, speeding up design-phase analysis. Phaidra applies reinforcement learning to industrial facilities, including data center cooling loops—a different approach, but one that also promises dynamic optimization. Cadence, after acquiring Future Facilities, has been selling digital twin software for data center design and operations for years.
What Inviscid is wagering on—perhaps more than the founders expected when they started—is that speed alone can carve out a niche. The company's argument: existing digital twins are mostly deployed for offline analysis or require simulation engineers to operate, while its system is fast enough to close the loop in real time. Whether that distinction matters to buyers will depend on how much friction those buyers experience with current tools, and how much they trust a neural network to make split-second decisions about airflow.
Early Days, Big Ambitions
With a team of two and no disclosed funding beyond Y Combinator's standard check, Inviscid is operating at what might generously be called the seed stage. As of early March, the company hadn't posted job openings. Its website offers a handful of case-study metrics—energy reductions, accuracy percentages—without naming customers. The founders are taking meetings via a Calendly link on the YC launch page, offering to build custom digital twins for enterprise clients.
The broader vision, if the launch materials are any indication, stretches beyond HVAC. Inviscid has floated the idea of expanding into electromagnetics, structural mechanics, and other domains where real-time physics simulation could enable smarter control systems. That's a long road from a two-person team debugging airflow models in data centers.
For now, the more immediate test will be whether the company can back up its energy savings claims under scrutiny from customers who have heard bold promises before—and whether building and data center operators, already managing tight margins and risk-averse procurement processes, are ready to hand over control of their cooling systems to a startup's neural network. The physics may be sound, but the business case will require more than simulation.
