Unconventional AI agrees to buy NeuReality to accelerate its AI hardware roadmap

Unconventional AI reached a definitive agreement on October 26, 2026, to acquire Israeli AI-chip startup NeuReality, adding engineers who build both chips and the systems around them. The company says the acquisition will shorten its path to production hardware as it pursues a radically more efficient approach to AI computing.

Both companies’ boards have approved the transaction. Closing is expected in the fourth quarter of 2026, subject to customary conditions, so the agreement is not yet a completed acquisition. Financial terms were not disclosed.

CTech’s Meir Orbach reported that the deal is understood to include cash and Unconventional shares. That structure would give NeuReality’s investors both a payment and an interest in the buyer’s future.

Seven years of building AI infrastructure

NeuReality brings expertise across system architecture, silicon design and AI systems software. Unconventional says the team has spent seven years designing, building and deploying AI infrastructure. Its pitch for the acquisition centers on that experience across complete systems, not simply on adding a single chip design.

Founded in 2019, NeuReality is led by co-founder and CEO Moshe Tanach. Its other founders are Zvika Shmueli and Yossi Kasus. CTech describes the founders as semiconductor veterans whose experience includes Intel, Marvell, Mellanox and Habana Labs—an established engineering background behind the team Unconventional is bringing in.

Tanach and the founding team will join Unconventional. The buyer also intends to maintain NeuReality’s engineering presence in Israel, where the company is headquartered in Caesarea.

The hardest thing in this industry is assembling a team that can build real silicon and the systems around it, and Moshe has built one of the best,

Naveen Rao, co-founder and CEO of Unconventional AI

The work around the AI calculation

NeuReality focuses on inference infrastructure: the hardware and software used to run AI models after training. Its technology targets data movement and system orchestration, the work of moving information and coordinating computing tasks. Those supporting operations can limit the speed and cost efficiency of the system running a model.

Unconventional argues that this overhead dominates inference’s real cost. Rao described the company’s challenge as a systems problem as much as a physics problem.

A 1,000-fold target, not a measured gain

Unconventional says it is developing what it calls a “4-dimensional compute substrate,” targeting a 1,000-fold improvement in AI computing efficiency over conventional architectures.

Reaching the goal requires coordinated design across device physics, chip architecture and the AI models themselves, according to Unconventional.