Endra AI Cuts Electrical Design From 2 Months to Hours

The Money Moves

A $50 million Series A led by Andreessen Horowitz closed June 1, 2026, GlobeNewswire's data shows, lifting Endra's total capital to $75 million in thirteen months. Construction Industry AI reported. Notion Capital and Norrsken VC — both in the $20 million seed round reported by Fortune in December 2025, Construction Industry AI found — doubled down. The GlobeNewswire release notes the funding follows a $3 million pre-seed in May 2025, GlobeNewswire's figures put the pre-seed at $3 million in May 2025.

The cash buys geography. Endra is planting a North American headquarters in New York, a West Coast office in San Francisco, and a UK base in London, targeting 100 employees globally within a year. A July 31 GlobeNewswire release ties the push to demand from enterprise engineering consultancies across North America, Europe, and the Middle East. Co-founder and President Anton Juric runs the U.S. go-to-market; he told the wire service "enterprise trust is won in the room." Hadla Bergman, hired from Einride, will adapt the platform to U.S. and UK codes, standards, and design workflows, a process that starts with the new teams on the ground.

Hiring data from Zero G Talent's job board shows the signal is live: six salaried roles posted with a median band of $235,000, spanning Marketing Lead, multiple Software Engineer–Product positions in New York and San Francisco, a Founding Product Marketing Lead in the Bay Area, and Product Specialist–Electrical roles in New York and London. Two roles appeared in the past week. U.S. bands run $119,000–$250,000; the London electrical specialist sits at £65,000–£75,000.

The logic is deliberate. Stockholm keeps the core engineering talent and early enterprise relationships: Buro Happold, Ramboll, AFRY, Rejlers piloted the platform before the Series A. The U.S. offices target the world's largest construction market and the data-center build-out straining MEP capacity. London anchors European and Middle Eastern accounts. Construction Industry AI called the $75 million "reasonable for a company trying to establish itself as infrastructure for a global profession before the incumbents notice how large the gap has become." The funding and expansion come as legacy CAD vendors embed AI copilots into their software.

The expansion also sets up Epoch, Endra's launch event scheduled for September 14, 2026 in Las Vegas, where the company will unveil its electrical module and outline its vision for AI-native MEP engineering. Chess grandmaster Garry Kasparov is billed as a speaker. CEO Niklas Lindgren described Epoch as "the biggest step forward in our product's history" and said "the industry has spent twenty-five years waiting for software that understands the actual engineering work. That wait is ending."

Speed: From Months to Hours

A code-compliant electrical design for a 500,000-square-foot commercial building — the scale of a mid-size data-center hall — traditionally consumes two months. Endra's platform delivers it in under a day. The system ingests standard BIM files, integrates natively with Revit via API, and reconstructs each project in a detailed 3D environment. From there it automates system design, device placement, routing, model generation, and compliance-ready documentation across 3D and 2D deliverables (IFC models, bills of quantities, riser diagrams, point-to-point drawings), all mapped back into the consultant's custom Revit families through a plugin.

The collapse shows up in early projects. On a 185,000-square-meter logistics hub in Sweden, Endra turned 300 hours of fire-alarm design into 30 minutes. Norrsken VC puts the productivity gain at over 70x for the repetitive, low-value tasks that still consume a huge share of MEP design hours, Construction Industry AI's figures put the gain at over 70x.

Endra's founders describe three layers in the current stack: the drafting layer (Revit), the engineering layer where actual sizing and compliance happen, and the coordination layer (Navisworks, Solibri) for clash detection. Their platform targets the middle. It processes IFC models, automatically annotates room context, and imports HVAC and structural models for clash avoidance. In a fire-alarm demo, setting the regulatory framework — BS 5839, NFPA 72, SBF 110:8, DIN 14675 — and placing the fire-alarm control panel triggers an "optimize" function that instantly places every unit across all floors. Cable routes are computed by evaluating hundreds of parameters, including structural data and voltage drop, using a hybrid of deterministic algorithms, simulation, and large language models that structure unstructured BIM data into Endra's proprietary model.

The company says it gets engineers to 80–90 percent design readiness "in a heartbeat," delivering code-compliant models, schedules, and takeoffs. Documentation that once took weeks generates in hours. Engineers remain in control of final validation; the platform handles the grind. That boundary matters: some early clients have declined to train junior engineers on the AI, preferring they learn the manual workflow first. Juric and Lindgren estimate the technology will meaningfully change 70–80 percent of actual project hours. They also argue it upends the outsourcing model, as firms will shift from offshoring routine design to using technology for that same work, keeping decision-making engineers closer to the product.

Regulatory coverage is live for EU standards; U.S., UK, and German codes are in beta as of the July launch. Pricing is outcome-based, not seat-based. The firms moving first, Endra's leadership says, will gain capacity, faster timelines, and a new way to scale expertise.

How Incumbents Are Fighting Back

Autodesk, Bentley, and Trimble each control a distribution layer spanning millions of seats (Revit, Civil 3D, OpenRoads, Tekla) and they are using that surface to embed AI before a specialist entrant can displace them. The pattern across all three is the same: copilots inside the existing authoring environment, generative engines that propose layout alternatives, and agent frameworks that execute multi-step workflows without the user leaving the model.

Autodesk's push is the broadest. Its Autodesk AI brand now covers predictive risk scoring on construction sites, assistive workflows in Construction Cloud, and a Model Context Protocol (MCP) server layer that lets large language models act directly on project data (creating issues, generating RFI responses, pulling quantities) through standard APIs rather than brittle screen scraping. A Data Connector released in 2025 links Autodesk Construction Cloud to Microsoft Fabric, giving Copilot Studio a live pipe into project information; Secure Service Accounts, shipped shortly after, remove the manual authentication step that previously stalled automated agents. At DevCon 2026 the company signaled an "agentic future" where AI orchestrates across design, documentation, and field execution without hand-offs. In manufacturing, Fusion's generative design engine already suggests topology-optimized bodies from load cases; the same architectural pattern is being ported into the AEC cloud.

Bentley moved earlier on civil-site generative AI. In October 2024 it announced a design copilot for OpenSite+ that produces graded site layouts, optimizes cut-fill balance, and emits construction-ready drawings from a single prompt. The follow-on OpenUtilities Substation+ applies the same logic to electric substation design, a direct adjacency to the data-center and electrification workloads Endra is chasing. Bentley's venture arm, SE Ventures, has backed two startups that flank the MEP problem: Temelion (September 2025 seed) for pre-construction workflow automation in building engineering, and Higharc (June 2026 follow-on) for end-to-end homebuilding design-to-construction. Both investments signal Bentley's intent to own the technical middle between architectural intent and field installation.

Trimble's response is more distributed. Its 2026 0-60 Challenge selected 15 AECO startups building AI, robotics, and interoperability integrations that plug into Trimble Connect and the Tekla ecosystem. Rather than ship a monolithic copilot, Trimble is curating a marketplace where specialist agents (clash resolution, rebar detailing, MEP coordination) can be invoked inside the same data environment its core users already inhabit.

What unifies these moves is the recognition that distribution alone is no longer a moat. Endra's platform, which delivers code-compliant MEP systems in hours instead of weeks, forces the incumbents to make their own environments the default host for that class of automation, using MCP, Copilot Studio, and proprietary agent runtimes to keep the workflow inside their walled gardens. The race is now over whose copilot the engineer trusts to run the next large-scale data-center electrical package unattended.

Where the Rubber Meets the Road

Endra's first production deployments are landing inside the consultancies that design the power and cooling backbone for hyperscale data centers, battery storage farms, and electrified industrial campuses. The company says it is in active engagement with several of the world's largest engineering firms across North America and Europe, a claim backed by a public partnership with AFRY, the Swedish-headquartered consultancy employing 19,000 people across energy, infrastructure, and industrial sectors. AFRY stated in a July 2026 release that the partnership allows it to evaluate the technology from the inside and help shape solutions for clients, people, and the wider industry, language that signals a co-development model rather than a simple license sale.

The demand pull is unmistakable. The June 2026 Nomic newsletter noted that AI-assisted plan review tools in Denver and Jacksonville are cutting first-round permit cycles by 45–60 days, and that LightTable's multimodal AI caught roughly 70 percent of design errors versus 30 percent for human-only review. The same briefing tied the funding surge for construction-AI startups (Endra's $50 million Series A, Xpanner's $18 million bridge, August Robotics' $30 million Series B, Foresight's $25 million Series A) directly to "AI data centers and energy infrastructure driving demand for faster, less labor-constrained delivery." Xpanner's CEO framed it bluntly: "Solar piling gave them the foothold. BESS and AI data centers give them the runway." Endra's July announcement echoed the same drivers: "Data centers, hospitals, electrified buildings, dense urban developments, and net-zero mandates all add engineering complexity" while "the industry is facing a shortage of qualified MEP engineers and cannot train new engineers fast enough to keep up."

No defense-specific pilot has been disclosed in available filings or press releases. The critical-infrastructure traction to date clusters around civilian energy and compute: utility-scale solar, battery energy storage systems, and the hyperscale data-center campuses that now dominate new electrical-load growth in the U.S. and Europe. The electrical module will be unveiled at Endra's Epoch event in Las Vegas in September. If the early benchmarks hold, the bottleneck shifts from design throughput to the speed at which consultants can review, stamp, and release AI-generated packages to authorities having jurisdiction.

Policy, Power, and the Labor Cliff

The SPEED Act (H.R. 4776), which passed the House 221-196 on December 18, 2025 and now sits before the Senate Committee on Environment and Public Works, is the clearest signal yet that Washington intends to unclog the permitting pipeline for AI infrastructure. The bill rewrites the National Environmental Policy Act to narrow the definition of "major federal action," impose a 150-day statute of limitations on NEPA lawsuits, and bar courts from halting projects during litigation, changes sponsors say could compress hyperscale data-center approvals from years to as little as 18 months. The National Association of Manufacturers, which backed the bill, reports that 80 percent of manufacturers view current permitting as harmful to investment and 87 percent would expand operations if the process were streamlined. That bipartisan vote — co-sponsored by Rep. Jared Golden (D-Maine) and Rep. Bruce Westerman (R-Ark.) — reflects a rare consensus: the U.S. cannot win the AI race if the physical plants that run the models cannot get permitted.

The permitting push is not abstract. Data centers consumed roughly 4.4 percent of U.S. electricity as of late 2025, and Lawrence Berkeley National Laboratory projects demand growth of 6.7–12 percent annually through 2028. Deloitte estimates U.S. data-center power load will grow more than fivefold to 176 gigawatts by 2035, with AI-specific facilities accounting for up to 123 GW — a thirtyfold increase from 2024 levels. McKinsey puts the capital required at $6.7 trillion by 2030, of which $5.2 trillion targets AI-ready capacity. The Trump administration's Genesis Mission, announced in November 2025, formalized this urgency: the Department of Energy signed collaboration agreements with 24 firms including Google, AWS, Microsoft, OpenAI, and Nvidia, while AWS committed up to $50 billion for government AI infrastructure and Microsoft spent $11.1 billion on data-center expansion in a single quarter. OpenAI's $38 billion AWS partnership underscores the scale. Every one of those facilities needs mechanical, electrical, and plumbing systems designed, coordinated, and approved, fast.

The market numbers confirm the pull. North America MEP software market size was valued at $5.8 billion in 2024 and is forecast to grow at a CAGR of 9.3% from 2026 to 2033, reaching $12.5 billion by 2033. Broader MEP engineering software market projections reach $184 billion by 2032 at an 11.1% CAGR. But the labor side is tightening in parallel. Construction wages rose 4.2 percent year-over-year as of August 2025. The industry needs 499,000 new workers in 2026, up from 439,000 in 2025, and faces a projected shortfall of over two million skilled craft professionals by 2028. By 2031, 41 percent of the current construction workforce will retire while only 10 percent are under 25. Deloitte calculates the labor gap could cost $124 billion in lost output. Automation is no longer a productivity play; it is a capacity play.

When Endra takes the stage at Epoch in September, the electrical module will face its first public stress test. The permit and the pour come next.

Working in frontier tech? Zero G Talent tracks the openings: see every open Endra role, browse frontier tech jobs, the companies hiring, and the people building the field.