Speedata

Six partners from Israeli academia

Speedata was founded by six partners with backgrounds at Tel Aviv University, the Technion and the Hebrew University: Jonathan Friedmann, Dan Charash, Rafi Shalom, Itai Incze, Yoav Etsion and Dani Voitsechov. According to the company's official stealth-exit announcement, Friedmann, then CEO and a co-founder, described the team this way: "Our amazing team of academic and industry leaders has built a dedicated accelerator that will change the way datacenter analytics are processed."

APU: a dedicated chip for the stage before training

Before a model sees any data, the data has to be cleaned, tables joined to each other, filtered, summarized and converted between formats. This work is done with tools such as Apache Spark on general-purpose processors in whole server farms, and it uses a great deal of time and electricity. Speedata's solution is a component it calls an APU, an analytics processing unit, which performs the common data operations directly in hardware. According to the company, a single server with its APU replaces several entire racks of regular processors, without requiring any change to existing code.

From stealth to $114 million in funding

The company came out of stealth in September 2021 with a $55 million Series A round led by Walden Catalyst Ventures, 83North and Koch Disruptive Technologies, which brought its total funding at the time to $70 million (including an earlier, previously undisclosed $15 million seed round). Among the private investors were Eyal Waldman, a co-founder of Mellanox, and Lip-Bu Tan, who was then a managing partner at Walden Catalyst Ventures and later became Intel's CEO. In June 2025 Calcalist reported a further $44 million Series B round, which brought total funding to about $114 million, with Tan taking part, this time as Intel's CEO.

Who leads the company today

As of 2026, Adi Gelvan is CEO of Speedata. He joined the company in 2025 following a successful chip tape-out and growing business interest. Gelvan came to the role after serving as CEO and co-founder of Speedb, a storage engine acquired by Redis in 2024, and before that he held management roles at SQream, Infinidat, XIV and EMC.

A principle that repeats itself in the hardware industry

Speedata illustrates a principle that repeats across the field of AI hardware: when a particular computing workload becomes common and stable enough, it pays to build a dedicated chip for it instead of running it on a general-purpose processor. What happened to neural network training with graphics processors is now happening, on a smaller scale, to the data preparation stage that comes before training. Another Israeli company, NeuroBlade, worked in the same direction. It developed a dedicated processor to speed up data processing and analytics, and offered it as a server card or as accelerated cloud instances, until Amazon took in its engineering team.