Opinion
AI’s next big winners will build what the giants depend on
"Israeli startups can build big businesses by solving the hardware problems that hold back AI," writes Doron Zauer, General Partner at Earth & Beyond Ventures.
The scale of AI investment can make the opportunity look out of reach for a young company. Competing at the frontier of AI models, or building a vast data center, requires billions. But a startup that solves a critical problem inside those systems can grow into a major business without financing the whole machine. For Israeli founders and early-stage investors, the opportunity lies in silicon architectures, advanced materials and the connections that keep AI infrastructure running.
At the application layer, competition is intense and products built around today’s models are exposed to rapid commoditization. In physical infrastructure, lasting competitive advantage comes from science and engineering that competitors cannot easily replicate. Model architectures change, but the need to process, store, power and move data remains.
As AI workloads grow, the challenge extends from the electricity grid to the chip. Grid capacity, energy costs and regulatory approvals constrain where data centers can expand. Inside them, power delivery, cooling and connectivity determine how much of the installed computing capacity can actually be used. Chip architectures and materials set further limits on performance and efficiency.
Adding more processors does little good if they sit idle waiting for data or if the system cannot remove the heat they generate. A startup that improves optical connectivity, for example, can enable more data to move for the same amount of power, improving the performance of the wider computing system.
Israel has an unusually strong base for building these companies. Decades of work in chip design, photonics, high-speed communications and advanced materials have produced engineers who understand both the underlying science and the demands of large systems. At companies such as Nvidia, Apple, Intel and Google, they see where architectures run into limits. When they leave to build startups around those gaps, they are not competing with Anthropic or building another wrapper app; they are engineering the foundations of future computing.
Commercial success depends on meeting the customer’s requirements. A customer will not risk a major product launch on a component that performs brilliantly in a lab but cannot be manufactured reliably or integrated at an acceptable cost. Manufacturing yield, reliability and system integration are part of the value of the technology. Founders need early engagement with industrial buyers and design partners who can validate components under real operating conditions. November’s DeepTech Week TLV will bring chip designers and system builders into the same discussion about what AI infrastructure needs to deliver.
Once a component has been qualified and designed into a larger product, replacing it may require redesign and renewed testing. That gives a startup a chance to turn an engineering advantage into an enduring commercial position. The same demanding requirements that make the first sale difficult can make a proven supplier valuable for years.
The longer-term investment question is what else the technology can do. An initial customer gives a startup a concrete problem to solve; a core technology with applications across different systems and markets gives it room to grow. Founders need that first commercial use without allowing one customer’s deployment to define the limits of the company.
AI’s infrastructure giants will keep spending to expand what their systems can deliver. Some of the largest venture returns will come from the deeptech companies that make that expansion possible. Israel’s opportunity is to build businesses whose science and engineering make them essential to systems far larger than themselves.
Doron Zauer is a General Partner at Earth & Beyond Ventures.
