Unconventional AI Agrees to Acquire Israel’s NeuReality, Deal Expected to Close in Fourth Quarter
Key Facts
Unconventional AI said on October 26, 2026, that it had entered into a definitive agreement to acquire NeuReality, a company specializing in AI inference infrastructure and silicon and based in Caesarea, Israel. The company said the agreement had been approved by both boards and that the deal is expected to close in the fourth quarter of 2026, subject to customary closing conditions. The deal therefore remains pending, and NeuReality has not yet transferred to the buyer. The companies did not disclose financial details. Globes estimated the consideration at between $150 million and $200 million in cash and shares, but described the figure as a market estimate, not a disclosed value.
The announcement focuses on bringing NeuReality’s team and expertise into Unconventional, rather than integrating specific inference chips into an existing infrastructure. The company said the team has experience in systems engineering, silicon design and AI systems software, and that its members spent seven years designing, building and deploying infrastructure. It added that the team worked to reduce the coordination and data-transfer overhead it said dominates the actual cost of AI inference. Unconventional said NeuReality founder Moshe Tanach and the founding team would join it, and that it intends to retain the company’s engineering presence in Israel. Globes reported that about 40 employees in Caesarea and Tel Aviv would join the U.S. company.
The deal comes as Unconventional is already developing AI computing infrastructure and chips. Reuters reported that it had completed an initial test chip and was nearing completion of a second, and that it was seeking engineering expertise to help turn the work into finished products. Founder Naveen Rao told Reuters that when the company raised funding, it expected research and development to take three years and product development to take five years. He added that AI tools had accelerated the work and that the company was now close to completing the second chip. This context frames the deal as adding expertise to accelerate development and production, rather than as the company’s entry into the chip business.
Unconventional presents its technical goal as a challenge involving both systems design and physics. Rao said the company aims to improve AI computing efficiency by 1,000 times, adding: "We are trying to make AI compute 1,000 times more efficient, and that is a systems challenge as much as it is a physics challenge." The goal puts computing efficiency at the heart of the company’s strategy, while NeuReality’s expertise in systems, silicon and software gives it additional resources to work on that problem. Unconventional said adding the team would accelerate its progress toward production systems. The announcement did not specify a launch date for a joint product.
Unconventional announced in December 2025 that it had raised $475 million in seed funding at a disclosed valuation of $4.5 billion. That valuation relates to the earlier funding round and is not a new value resulting from the acquisition. Globes’ estimate of the deal value, at $150 million to $200 million, remains a press estimate; the newspaper said financial details had not been disclosed. The distinction between the two figures makes clear that the company’s valuation and previous funding do not determine the consideration paid for NeuReality. Unconventional’s announcement gave no value for the deal.
NeuReality’s products include an inference operating system called NR-NEXUS and an AI networking chip called AI-SuperNIC; the company also offers an AI-CPU product. The company also has a prior record in chip development: in September 2023, it announced that its 7-nanometer NR1 design had passed final verification and been sent to TSMC for manufacturing. NeuReality said the chip recorded 10 times higher performance at the same cost in its tests, compared with traditional systems based on central processing units. In a later announcement, it said the NR1 system was ready for customers in the first quarter of 2024, after the chip arrived from Taiwan and system integration was completed within 90 days. These results are attributed to the company’s tests and announcements.
NeuReality said tests it conducted on physical hardware in May 2024 showed average cost savings of 90% across the use cases it tested, compared with systems combining central processing units with Nvidia H100 and L40S chips. It also said the system was ready for early trials with cloud providers and financial and healthcare institutions, handling workloads including computer vision, automatic speech recognition and natural language processing. These results and expectations were reported in company statements. The acquisition is expected to close in the fourth quarter of 2026 after customary closing conditions are met. Until then, the companies have not disclosed final financial details.