The engineering notices went out quietly, and the Israeli tech industry noticed anyway. Google is assembling a new team in Israel to design edge-AI chips, accelerators that run artificial-intelligence workloads on devices rather than in the cloud, and it has hired Guy Kaminitz, a former vice president of research and development at the Israeli AI-chip company Hailo, to lead the effort, according to Israeli media. The team is still hiring, with dozens of engineering positions open.
The move extends a pattern that has made Google’s Israeli offices one of the most important silicon labs outside the United States. The company’s engineers in Israel have played a leading role in developing Axion, Google’s first self-designed server processor built on Arm technology, which the company began deploying in its own data centers. The new edge team pushes the same strategy in the opposite direction, toward the small, power-conscious chips that sit inside robots, cars and other machines.
Kaminitz’s hiring is a signal in itself. Hailo made its name designing chips that give ordinary devices the ability to run AI locally, processors small enough and cheap enough to go into cameras, cars and industrial equipment. The company was recently acquired by Microchip Technology, the American chipmaker, and Kaminitz’s move to Google brings years of edge-AI experience to a team that has none of Hailo’s product lines to worry about, only its engineering methods.
The timing reflects where the industry believes the next growth is. Cloud AI has been Google’s focus for years, with its tensor processing units competing against Nvidia’s accelerators for training and inference work. But a growing share of AI is moving to the edge, into vehicles that must react without waiting for a network, robots that work in environments with unreliable connectivity, and devices whose users will not accept sending every request to a distant server. Chips for that market are the frontier Google is now staffing.
Google has described the target applications in terms that the industry calls physical AI: systems that perceive the real world and act in it, from warehouse robots to self-driving cars. Those systems need inference that is fast, local and power-efficient, qualities that define a different engineering trade-off than data-center chips. A chip that must run for hours on a robot’s battery is designed with constraints a server accelerator never faces.
The data-center side of the Israeli operation is growing too. Ido Bukspan, a former executive at Mellanox and Nvidia and the former chief executive of the storage startup Pliops, has joined Google to lead work on chips for data-center interconnects, the links that let servers and AI accelerators talk to each other at the speeds modern workloads require. The appointment echoes the history of Israel’s chip industry, which grew up around Mellanox, the networking company Nvidia acquired for its high-speed interconnect technology.
The two hires fit a single strategy. Google is trying to reduce its dependence on Nvidia, the supplier that dominates both the accelerators inside its data centers and the networking that connects them. Building its own interconnect chips addresses one half of that dependency, and building its own edge accelerators addresses a market where Nvidia is strong but not entrenched. In both cases, the company is hiring people who have spent their careers building the very products it wants to replace.
Israel has become a natural home for this work. The country produced Mellanox and a generation of networking engineers, Hailo and a generation of edge-AI designers, and a deep pool of chip talent that American companies have raided for years. Google’s offices there already run some of the company’s most sensitive hardware projects, and adding edge and interconnect work consolidates Israel’s role as a center for silicon that Google considers strategic.
The expansion also has a defensive quality. Every major cloud company has concluded that the AI boom’s economics depend on controlling its own silicon, and Google has been among the most aggressive, with its TPU line now in its sixth generation and Axion rolling out across its fleet. The Israeli teams extend that control to the edges of the network, where the next generation of AI devices will run.
For the Israeli tech industry, the hires are a mixed signal. Kaminitz and Bukspan represent a flow of senior talent out of local startups and into the Israeli offices of American giants, a pattern that has accelerated as acquisition activity consolidates the country’s chip sector. The startups that trained these engineers may struggle to keep their successors, even as Google’s presence gives the country’s engineers a reason to stay in Israel rather than move to Silicon Valley.
Google, for its part, is making a long-term bet on a country that has become central to its hardware ambitions despite the region’s volatility. The company’s Israeli engineering center has delivered some of its most important custom silicon, and the new teams give it a footprint that spans the full range of AI hardware, from the chips that run the biggest models to the chips that will run inside the smallest devices. The hiring has just begun, and the dozens of open positions suggest Google intends the Israeli edge team to be more than an experiment.
