The grid's problem isn't a shortage of power. It's a shortage of finished paperwork. Vela Energy, a Y Combinator W26 startup, is building AI agents to do that work - and keeping a licensed engineer on every decision.
Ask most people why the artificial-intelligence boom might stall, and they'll point at chips. The real chokepoint is less glamorous and harder to fix: getting electricity to the buildings full of chips. Across the United States, roughly 2,600 gigawatts of proposed power and storage projects are waiting in interconnection queues, more than double the country's entire operating capacity. The projects exist. The demand exists. What's missing is a way through the line.
Vela Energy, a startup in Y Combinator's Winter 2026 batch, has picked that line as its problem. The company builds what it calls AI execution agents - software that does the procurement, permitting, and engineering-study work behind large-load energy projects, the kind that power data centers, factories, and new substations. The pitch is narrow and specific: not a dashboard that shows you the mess, but agents that work through it.
"We exist to unlock the queue," reads one of the company's stated principles. It's a plain sentence for a genuinely tangled problem.
A large energy project starts with clean data - a site, a load, a set of requirements. Then reality arrives. Information scatters across email threads, chat messages, spreadsheets, and a dozen disconnected tools. Studies get re-run. Permits sit open. A single long-lead component, like a transformer, slips its delivery date, and the schedule quietly rots around it. By the time anyone notices, the project has lost months it can't get back.
The consequences are national in scale. American Electric Power alone has reported 63 GW of new load coming by 2030, with 89% of it tied to data centers. In Texas, grid operator ERCOT tracks more than 438 GW of large-load interconnection requests - nearly 90% from data centers - with wait times stretching from two years to five or more. In June 2026, federal regulators at FERC issued show-cause orders pressing grid operators to justify or fix how they handle large loads. The pressure is real, and it lands on the teams doing the paperwork.
Vela's product is organized around a set of agents, each aimed at a stage that normally eats weeks of an engineer's time. They sit inside an Agent Workspace with roles, approval gates, least-privilege access, and audit trails - the plumbing that lets a serious organization trust software with real decisions.
Teams, roles, and approval gates with least-privilege access and full audit trails.
Monitors brokers, auctions, decommissionings, and OEM channels; flags gaps in nameplate matching and test records.
Maps air, water, NEPA, zoning, and local rules into deterministic checklists with citations and dates.
Runs load-flow, short-circuit, and reliability models in ETAP, PSS®E, and PSCAD - with engineer sign-off.
The clearest way to see the value is a small example the company uses to explain itself. On a hypothetical storage project, a transformer (a "GSU-01") slips its delivery by 28 days. The Procurement Agent catches the change and traces it downstream - four milestones affected, and a decision to pay a $276,000 premium to keep the schedule. No new science; just the difference between finding out today and finding out next quarter.
That integration count matters more than it looks. Vela connects to roughly 25 systems - Gmail, Google Workspace, Autodesk, Procore, SAP - and to the grid operators themselves: ERCOT, PJM, MISO, CAISO. An agent is only as useful as what it can see, and energy work lives in a lot of inboxes.
There's an obvious temptation in agentic software to take the human out. Vela made the opposite call, and it's a deliberate one. Every output is sourced and overrideable. The Studies Agent won't finalize a load-flow or short-circuit model without a licensed engineer's sign-off. "The engineer always decides" is written into the company's principles, and in critical infrastructure that isn't a slogan - it's a liability strategy. A wrong number on a power-system study is not a typo.
This is also where Vela draws its line against competitors. Most software sold into energy is a better way to look at the mess: a dashboard, a tracker, a shared spreadsheet with nicer charts. Vela argues the work itself - the monitoring, the mapping, the modeling - is what should be automated, with people supervising rather than typing. Watching versus doing is the whole distinction the company is betting on.
Vela's two founders arrived from opposite ends of the same problem. Tarun Batchu, the CEO, comes from the policy and community side of energy: a UN Global Goals Ambassador, a former national co-lead for the Citizens' Climate Lobby, an organizer of Bloomberg Philanthropies-backed efficiency programs. His interest is not new. At 12, he built a carbon-capture project, sponsored by Arrow Electronics, that he pitched to ExxonMobil.
Tony Li, the COO, comes from the hardware and research side: published work with the U.S. Department of Energy, bioenergy projects supporting the Department of Defense, and a stint as the youngest intern on Tesla's Thermal Purchasing Team, working on the Megapack platform. He's an Ingram Scholar at Vanderbilt, one of ten chosen each year. One founder spent years watching the grid from the outside; the other spent them inside the supply chain. Vela is what happened when they compared notes.
Vela sells to the organizations that actually move projects: EPCs (the engineering, procurement, and construction firms), owner's engineers, developers, utilities, and data center operators. The model is B2B software - agents that plug into a customer's existing tools and sit alongside their teams. As an early-stage company, Vela hasn't disclosed named customers, and it demonstrates the product with representative projects rather than logos it can't yet share.
The company raised a $1.3 million pre-seed round led by a16z Speedrun and Z Fellows, and joined Y Combinator's W26 batch. The timing is not an accident. The AI compute race has created a demand for power that the connection process was never built to deliver quickly, and every regulator, utility, and hyperscaler is now looking for ways to shave months off the timeline. Software that removes process debt is, suddenly, strategic.
Names carry hints. Vela is a constellation in the southern sky; the Latin means "the sails." The company is small - a young team, a first check, a product still proving itself against projects where mistakes are costly. It has not yet shown named customers or a track record measured in built megawatts, and the interconnection process it targets is thick with incumbents, regulation, and habit.
But the underlying bet is easy to state. If the AI era runs on electricity, and electricity is blocked less by physics than by process, then whoever makes the process move faster helps decide how quickly that era arrives. That's the ground Vela Energy has chosen to build on. Whether its agents can carry the weight of real projects is the question the next few years will answer.
Video interviews and a product demo were not publicly available at the time of writing. Check Vela Energy's website and LinkedIn for the latest.