The pitch sounds almost too simple: What if you could model a building's airflow and thermal dynamics in real time, adjusting HVAC systems moment-by-moment rather than relying on the slow, static simulations that have governed facility management for decades?
That's the wager Kabir Jain and Ziming Qiu are making with Inviscid AI, a fledgling company that emerged from Y Combinator's Winter 2026 batch. Their timing may be fortunate. Data centers are colliding with the physical limits of electrical grids, and the pressure to wring efficiency from cooling systems—often the second-largest power drain after servers themselves—has never been more acute.
An EPRI analysis from February 2026 projected AI-driven workloads could push data center electricity consumption to 17% of total U.S. demand by 2030. Whether that forecast proves accurate or not, the trajectory is undeniable: facility operators face a reckoning over energy use, and the market for technologies that promise meaningful reductions is heating up accordingly.
Inviscid claims its platform can deliver energy savings of 15–30% across buildings and data centers. Independent validation of those figures hasn't surfaced yet. But in a sector where even marginal efficiency gains translate to millions in avoided costs and expanded grid access, the startup has already secured early deployments in India and Singapore.
Faster Than Physics Used to Be
Traditional computational fluid dynamics—the simulation toolkit engineers have relied on for years to model how air moves through complex spaces—operates on a timeline measured in hours. Eight to twelve, typically, for a single run modeling airflow patterns in a large facility. That's far too sluggish for active control.
Inviscid's approach leans on GPU-accelerated solvers and what it calls physics-informed neural operators, compressing that computational cycle to something closer to real time. The company pegs the speedup at roughly 1,000x versus legacy CFD methods, though the comparison depends heavily on which legacy methods you're measuring against.
The system ingests live data streams from building management systems and IoT sensor arrays, constructing a digital twin that continuously updates as conditions shift. A facility manager can test virtual adjustments—different temperature setpoints, altered airflow configurations—before deploying them in the physical world, sidestepping the trial-and-error approach that still governs much of HVAC operation.
Inviscid says its models achieve better than 95% accuracy against established physics benchmarks. That's a claim worth scrutinizing once third parties have had a chance to kick the tires.
Where the Grid Meets the Wall
Buildings globally consume about 40% of all energy, with HVAC systems accounting for anywhere from 30% to half of that load depending on climate and building design. Data centers face even sharper constraints, particularly as AI workloads proliferate.
The U.S. Department of Energy estimated data centers consumed 4.4% of the nation's electricity in 2023. Projections from various sources suggest that figure could climb to somewhere between 6.7% and 12% by 2028, though forecasting in this space has a tendency to overshoot or undershoot as technology and deployment patterns shift.
Grid capacity is fast becoming the binding constraint. The UK government this past spring announced plans to prioritize data center grid access amid tightening supply. Operators who can demonstrably cut their power draw gain leverage in siting decisions and expansion negotiations—a competitive edge that matters when new capacity can take years to bring online.
Perhaps more to the point, reducing energy waste is no longer just about cost. It's about whether you can build at all.
Pilots and Promises
Jain and Qiu founded the company in 2025 and have moved with speed that suggests urgency—or at least awareness of the crowded field they're entering. In mid-February, Madhya Pradesh State Electronics Development Corporation signed a memorandum of understanding to deploy Inviscid's digital twin technology at a state-run data center.
Around the same time, the startup announced a partnership with The GEAR by Kajima, a Singapore-based experimental facility operated by Kajima Corporation's local subsidiary. The collaboration positions Inviscid's platform as a testbed for smart building optimization in what the industry likes to call "living labs"—real-world environments instrumented for continuous monitoring and iteration.
Inviscid has referenced one customer achieving a doubling of cooling efficiency, though the company hasn't disclosed specifics about the deployment, the baseline, or the methodology. That opacity is common in early-stage enterprise software, where customers often request anonymity. Still, it makes independent assessment difficult.
Jain attended SEMI-THERM 2026, a thermal management conference in San Jose in early March, focusing presentations on real-time optimization applications. It's the kind of circuit-working that signals a startup angling for traction in a technical niche.
A Market That Doesn't Lack for Competition
Inviscid is hardly alone in chasing HVAC optimization through software intelligence. BrainBox AI, which had claimed energy reductions of up to 25% in HVAC operations, was acquired by Trane in early 2025. Phaidra launched a pilot across Khazna Data Centers in the UAE around February, with vendor claims approaching 40% energy savings—a figure that, as with most pilot-stage promises, awaits broader validation.
Johnson Controls markets its OpenBlue platform to enterprise customers at scale. Siemens offers AI-optimized cooling solutions that integrate with its existing building automation infrastructure. NVIDIA's Omniverse platform has become something of a backbone for physics-aware digital twins across industries, with partnerships that include Siemens and Jacobs Engineering. PassiveLogic's autonomous building controls cite pilot savings in the 30% range.
What Inviscid offers, at least in theory, is specificity: a platform purpose-built for real-time airflow physics rather than the broader, more generalized building automation suites that dominate the market. Whether that narrow focus becomes a defensible moat or a limitation depends on execution—and on whether facility operators value best-of-breed point solutions over integrated platforms.
The company lists its target customers as facility operators, data center teams, cold storage owners, and manufacturing engineers. Essentially, anyone managing climate-controlled environments at scale. With grid constraints tightening and energy costs climbing across much of the developed world, the pitch is direct: Optimize continuously, or pay the premium for inefficiency.
The Unanswered Questions
Two people. That's the team behind Inviscid, at least for now. It's both a strength and a vulnerability. Startups at this stage move fast precisely because there's no organizational drag. But scaling enterprise sales, supporting deployments across geographies, and maintaining the kind of simulation accuracy required for mission-critical infrastructure—those are challenges that don't easily bend to ambition alone.
The company will need to prove its models hold up across the messy realities of aging infrastructure, inconsistent sensor data, and facility operators who may not have the technical fluency to extract value from a physics simulation platform, no matter how fast it runs.
And then there's the question of how defensible the technology actually is. Physics-informed neural networks are an active area of research, not a proprietary black box. GPU-accelerated CFD solvers are becoming more accessible. If Inviscid's advantage is speed and specificity, competitors with deeper pockets and existing customer relationships could close that gap faster than a two-person team can build a sales pipeline.
Still, in a market where the cost of doing nothing is rising—literally, in the form of stranded capital and unbuilt data centers—there's room for solutions that work. Whether Inviscid becomes one of them, or simply a datapoint in the larger story of AI-driven infrastructure optimization, will depend on what happens beyond the pilots and the press releases.
For now, the company is making its case one simulation at a time.
