How AI is modernizing EPCs

Two pieces of conventional wisdom get revised here. The first is that construction is cyclical, so a construction business rides booms and busts. Hershman spent the first two-thirds of his career in commercial construction watching a roughly seven-year cycle, and says renewables has not behaved that way: the business has grown more or less continuously since 2008, with small dips in 2017 and 2018 from a development hangover after the investment tax credit cliff expected in 2016. What is different now is that load growth is real, and his read is that solar and storage are the fastest and cheapest things to deploy against it. He also designs the cyclicality out, pairing construction with operations and maintenance across a 35-year-plus project life for a recurring revenue base spanning more than 20 gigawatts.

The second revision is about labor, and it is the one most likely to get mangled. Hershman does not claim there is no labor shortage. He claims his work is different from a data center’s. A solar project is roughly 85 to 90 percent mechanical labor, putting posts in the ground, bolting racking together and setting modules, with only 10 to 15 percent requiring skilled electricians for terminations and splicing; in a data center that highly skilled electrical work is happening across every square inch. So the mechanical majority can be hired locally and trained, while the genuinely scarce skills sit in a minority of the scope and in his high-voltage and substation groups. Scale compounds the effect: twenty 20-megawatt projects each need their own management team, whereas a single team plus some added management can run a 300 to 400 megawatt project. Megawatts grow much faster than the people managing them. He still says a 20-gigawatt run rate would require more field labor. He just thinks he could recruit and train it.

So what breaks first? His answer is logistics. Asked how you build two gigawatts on a single site in a year, his response is not about finding people but about how to move that many trucks, where to park workers and how to bus them in. The historical evidence is the 2022 and 2023 supply chain disruption, which he calls the biggest hit his business has taken: racking systems erected with no panels to put on them, crews sitting idle, and the expensive cycle of demobilizing and remobilizing people you then have to retrain. The countermeasures are all logistical rather than technological. There is no just-in-time delivery on these projects; he would rather deliver early and double-handle material than risk a gap. Procurement and pre-construction are centralized so the whole fleet of projects is managed as one, and if 80 percent of the cable is identical across jobs it gets bought months to a year ahead, leaving only the job-specific 20 percent to source late.

That is the frame AI drops into, and it explains why the applications he describes are logistics applications. Hershman says SOLV sits on both sides of AI: it drives demand for his product, and the same technology is used to optimize his own operations. Concretely, he runs simulations of site layout, where to park people, how many laydown yards to use, how far break areas and restrooms sit from the work face, chasing a challenge he has set his teams to find 15 percent more productivity, or 15 more productive minutes, out of every person’s eight-hour day. Those questions used to be answerable only by building it and finding out. The data comes from internal platforms, Sunscreen for project production data and Vitals for plant management, with commercially available tools layered on top. Robotics is real but earlier: every site has some form of automation under test, and the payoff he is chasing is speed rather than headcount, since he already thinks labor is expensive enough to be worth optimizing whether or not he can get it. He is candid that the machines do not yet win a straight head-to-head against people, and that the more interesting question is whether they could run a third shift. The gains he describes are incremental by design: 10 or 15 percent off a total build, perhaps 20 percent in certain regions, in a business he says resembles manufacturing more than construction because it is the same activity repeated a million times.