ThinkLabs AI: physics-informed models collapse a month of grid engineering into minutes

Josh Wong has spent two decades telling utilities their planning tools are obsolete. He ran a distributed-energy management company, then grid orchestration at GE Vernova, and in early 2024 GE spun out his new firm with $6.8 million in seed capital. The pitch that finally landed: train AI on the physics simulators engineers already trust.

TIMELINE: ThinkLabs AI — Physics-Informed Grid Simulation
─────────────────────────────────────────────────────────────
 2014 ──────── 2024 ──────── 2025 ──────── 2026 ──── 🔥 NEXT
 ⚙️          🚀           🧪          💰       ◉ SCALE-UP
 Opus One    GE Vernova   Microsoft    $28M     Utility
 (DERMS)     spinout      SCE pilot    Series A deployments

ThinkLabs builds AI digital twins of the electric grid. A utility models its network in software, and the model learns from the same first-principles simulators the utility already uses, then runs orders of magnitude faster. According to the company, a study that used to take a month now runs in under three minutes, with 10 million scenarios executed in ten minutes at greater than 99.7% accuracy on grid power-flow calculations.

In March 2026 ThinkLabs closed a $28 million Series A led by Energy Impact Partners (EIP), one of the largest energy-transition funds in the world. Nvidia's venture arm NVentures and Edison International, the parent of Southern California Edison (SCE), joined the round. Existing backers GE Vernova, Powerhouse Ventures, Active Impact Investments, Blackhorn Ventures, and Amplify Capital returned.

A utility investor, a chipmaker's fund, and the company's former corporate parent all sat in the same round. The firm is doing, by its own admission, something none of the incumbents do: AI-native grid simulation as the primary product, rather than a feature bolted onto legacy planning software.

Why a month of engineering time collapsed into minutes

Grid planning runs on power-flow studies. When a data center wants to connect to a substation, or a neighborhood fills with EV chargers, engineers simulate how electricity would move through the network. Traditional tools from Siemens, GE, and Schneider Electric can take weeks or months per scenario. Wong describes the bottleneck in blunt terms.

We're not hallucinating the heck out of things. We are talking about engineering calculations here. We do have a source of truth from existing physics-based engineering models.— Josh Wong, founder and CEO, ThinkLabs AI

The source of truth matters because it makes the model explainable and auditable. In an industry where a miscalculation can black out a city, that is not a luxury. The January 2026 collaboration with SCE, built on Microsoft Azure AI Foundry and Nvidia high-performance computing, gave the claim real numbers. Engineers previously spent 30 to 35 days per energization analysis, including six hours of data preparation per project. The AI trained in minutes per circuit and processed a full year of hourly power-flow data across more than 100 circuits in under three minutes, producing bridging-solution recommendations and an engineering report in under 90 seconds.

Scale-up stage, not lab stage

It works with more than ten utilities on AI-native grid simulation for planning and operations, and Wong says the company doubled its customer accounts in the first quarter of 2026 alone. Sales cycles are compressing from the traditional one to two years down to two to three months.

The demand driver is not subtle. U.S. electricity demand is projected to grow 25% by 2030 according to ICF International, pushed by AI data centers, electrified transport, and building electrification. Annual power demand from U.S. data centers is projected to exceed the entire output of Texas by 2028, according to S&P Global. That surge is landing on infrastructure engineered decades ago for a different load profile. As we wrote in July, AI is also being turned on the grid's defence side, with Sandia modelling distributed energy control for grid resilience.

The financing will go toward taking the product to enterprise grade and expanding use cases. Wong frames the opportunity as a land-and-expand play inside each utility account: model a region first, then train across entire states or multi-state territories. He also sees a democratization effect, where smaller utilities that could never staff a planning team can run sophisticated analyses without the engineering headcount.

The turning point was trust, not compute

Utilities are among the most conservative technology buyers on earth. Procurement cycles stretch for years, and regulators watch every decision. The SCE collaboration mattered because it was a real utility, on real networks, with published results, not a vendor demo. When the utility's own senior vice president for system planning describes the technology as "a key enabler" for grid digitalization, that is worth more than any benchmark slide.

SCE's Sergej Mahnovski, managing director of strategy at Edison International, made the urgency explicit in the March announcement: the industry must "rapidly transition from legacy planning tools" to meet demand. Wong goes further, predicting that within five years most traditional utility planning and operational processes will be powered by AI.

What this means for the grid, and for investors

The pattern is recognizable from other capital-intensive industries. The compute exists. The physics is known. What was missing was a software layer that could run the engineering at the speed decisions actually get made. That is the asset class being built here, priced at whatever the market decides a time-compression tool for a $2 trillion physical network is worth.

Wong says the next two years will "dictate and define the next 50 years of the grid." That may be overstated, but the direction is not in dispute. The grid's bottleneck has moved from generation to planning, and planning has moved from spreadsheets to GPU-accelerated simulation.