Much of the global conversation around artificial intelligence focuses on models, interfaces, and applications. But Amir Fishelov, managing partner at Square One Labs and, before that, co-founder of SolarEdge, believes that narrative is incomplete. What’s more, it is potentially misleading about the current state of high tech, especially in Israel.
“It doesn’t matter if Google will win the race or OpenAI,” Fishelov says. “They all need the same infrastructure.”
I spoke to Fishelov about Square One Labs, a venture creation and investment platform that helps founders build transformative tech companies from the earliest stages. Founded in early 2024, it blends early-stage funding with operational support, providing labs, facilities, mentoring, and strategic guidance to turn big ideas into real companies.
It is deliberately focused on what Fishelov calls “physical infrastructure for AI”: energy, semiconductors, robotics, and the hardware systems that make large-scale AI possible. While software continues to move quickly, he argues that the real bottlenecks sit much deeper in the stack.
“Looking into AI, we figured out there’s lots of opportunity in the software space,” he says. “But there’s a huge opportunity on the actual infrastructure for AI.”
The logic makes sense. Training and running advanced AI systems requires enormous amounts of energy, compute, and physical reliability. Regardless of which models dominate, the same foundational systems must exist underneath them. “All the basic software layers, all of these are critical infrastructure that any solution for AI needs,” Fishelov says.
This focus is informed by experience. At SolarEdge, Fishelov helped scale energy and semiconductor systems globally, selling more than 200 million units of proprietary semiconductors. That operational background shapes Square One Labs’ investment thesis today. “We just came from these worlds,” he said. “Energy is our bread and butter. We also saw what happens in manufacturing.”
Unlike consumer software, infrastructure businesses operate on long timelines and demand precision early. “Deeptech companies were hard 20 years ago,” Fishelov says. “They’re still hard today.” In these sectors, missing early accuracy can be fatal. “Not hitting a target fast can cost the company three to four years in terms of capital and runway.”
That difficulty, however, is exactly what creates defensible moats. “The moat is very high,” he says. “Once they penetrate into the market and you have a differentiated solution… you can be in the game for the long run.”
Fishelov points to energy as a clear example. New systems don’t just need to be cleaner—they must be cheaper and more reliable than technologies built over a century. “You’re competing with gas solutions, oil solutions, whatever, which were built here for the past 100 plus years,” he says. “Now you need to build a new solution which is more cost-competitive than that.”
One Square One Labs investment in geothermal energy reflects this approach. By developing more efficient drilling methods, the company aims to deliver “cheaper energy, green energy, and available 24-7 energy,” Fishelov says—an infrastructure-level improvement that benefits any AI system layered on top.
AI itself, he adds, is accelerating deep-tech development rather than replacing it. “AI can help you understand a much broader range of topics,” he says, speeding up research, simulation, and material discovery. But it doesn’t remove the need for physical systems that work reliably at scale.
I enjoyed speaking to Fishelov, who emphasized that the future of AI will depend less on who writes the best model, but who builds the systems that keep it running.
You can learn more in the video above.
