Unconventional AI to Buy Chipmaker NeuReality

Unconventional AI, the chip start-up led by Naveen Rao that launched last December with a $4.5bn valuation, has agreed to buy NeuReality, an Israeli company that designs chips and systems for running AI models. The announcement on Unconventional’s blog on 26 October did not give a price.

The deal is about people as much as technology. Rao told Reuters, in a report carried by CNA, that his company’s research has moved faster than planned and that it now needs engineers who have taken chips all the way to finished products. NeuReality, founded in 2019, has spent seven years building inference chips and the systems around them.

In brief

Unconventional AI has signed a definitive agreement to buy NeuReality, based in Caesarea, Israel. Terms were not disclosed.
Rao told Reuters it is a cash-and-stock deal bringing about 40 people into the company, which will keep a presence in Israel.
The deal is expected to close in the fourth quarter of 2026. Unconventional’s 1,000-fold efficiency goal remains a target.

Unconventional is buying a team that has already shipped silicon

According to Unconventional, NeuReality’s engineers cover system architecture, silicon design and AI systems software. The company says the team has spent “the past seven years designing, building, and deploying AI infrastructure”, aimed at the orchestration and data-movement overhead that it says “dominates the real cost of AI inference”. Inference is the work of running a trained model to answer requests, as opposed to training it.

NeuReality’s chief executive, Moshe Tanach, and its founding team will join Unconventional. The buyer says it intends to keep NeuReality’s engineering presence in Israel. Both boards have approved the deal, which is expected to close in the fourth quarter “subject to customary closing conditions”. Unconventional repeated the news in a post from its official X account.

“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,” Rao said in the announcement. Tanach said NeuReality was founded on the belief that “the economics of AI inference were fundamentally broken”. He called the deal “the next chapter of the mission we started with: fixing the economics of AI, now with the platform to do it at full scale.”

Rao, who sat on NeuReality’s board, told Reuters the deal is paid in cash and stock. Israeli outlet Ctech reported that NeuReality had raised about $105m from investors including Arm, Samsung Ventures and OurCrowd, and that the deal is understood to return more than that to them. NeuReality has not published its own statement.

Rao says research is running ahead of his own timetable

Unconventional says it is building a “4-dimensional compute substrate” that targets “a 1,000x improvement in AI compute efficiency relative to conventional architectures”. Reuters describes the approach as a “dynamical systems” design, in which simple computing elements interact and evolve towards a result rather than following step-by-step instructions, as conventional chips do.

When it launched in December 2025, the company said it had raised $475m in seed funding at a $4.5bn valuation, in a round led by Lightspeed and Andreessen Horowitz, with Sequoia, Lux Capital, DCVC, Future Ventures and Jeff Bezos also taking part. Rao told Reuters the plan at that point was three years of research and five years to a product.

Instead, he said, AI tools have sped up the work. Unconventional has developed its first test chip and is close to finishing a second. “Things have gone a lot faster than anticipated,” Rao told Reuters. “We built up the research bones and now we’ve got to build up the production bones.” Chip-design software makers are pushing the same idea: on Monday Cadence said its new Maestro AI models can lift chip design quality by up to 30%, by its own measure.

Rao has sold two start-ups before. Reuters notes that Intel bought Nervana for more than $400m and Databricks paid $1.3bn for MosaicML.

The deal buys production experience, not proof of the 1,000x goal

The deal tells readers more about Unconventional’s stage than about its physics. A start-up that expected eight years before a product is now buying a ready-made team, which is a clear sign that Rao thinks the hard part has shifted from invention to delivery. That fits his own account, but the only evidence offered so far is his word that two test chips exist. Unconventional has not published results from either.

The 1,000x figure needs the same care. It is a goal, measured against “conventional architectures” that the company does not define, and nothing published so far shows how close the test chips come. Efficiency is the right problem to chase: TSN’s explainer on how much power one AI query uses shows how small gains per question turn into large totals at scale. But incumbents are not standing still either. Dell, for example, has just built a rack-scale inference system around Groq chips, an approach that is already on sale rather than still in the lab.

For NeuReality’s backers, Ctech reads the deal as a better result than recent Israeli chip sales such as Pliops and Hailo, which it says went for less than the capital invested, and the shares keep them in Unconventional’s upside. It follows Rocket Software’s purchase of Geniez AI, another Israeli AI team, earlier this month. TSN tracks deals like these on its AI funding rounds hub.

The test is a working product. Rao now has the production team he asked for, so the next thing to watch is whether Unconventional publishes measured efficiency figures for a chip it can sell.