Speedata, a Tel Aviv-based startup designed to accelerate big data analytics and AI workloads, has developed an Analytical Processing Unit (APU) to raise a $44 million Series B funding round and raises its total capital to $114 million.
The Series B round was led by existing investors, including strategic investors such as Walden Catalyst Ventures, 83North, Koch Anructive Technologies, Pitango First, Viola Ventures and Manage Partner, as well as strategic investors including Eyal Waldman and former CEO of Mellanox Technologies.
Unlike the Graphic Processing Unit (GPU), which was originally designed for graphics and later changed for AI and data-related tasks, the APU architecture focuses on addressing specific analytic bottlenecks at the computing level, according to the startup.
“For decades, data analytics has relied on standard processing units, and more recently companies like Nvidia have invested in pushing GPUs for their analytical workloads,” Speedata CEO Adi Gelvan said in an interview with TechCrunch. “These are either general purpose processors or processors designed for other workloads, not chips built from scratch for data analysis. Our APUs are dedicated to data processing, and a single APU can replace a rack of servers, providing dramatic performance.”
Speedata was founded in 2019 by six founders. Some of them were the first researchers to develop coarse-grain reconfigurable architecture (CGRA) technology. The founder worked with ASIC design experts to address basic issues. Data analysis was performed by a general purpose processor. If your workload is too complex, you will need to take advantage of hundreds of servers. The founders believed that a single dedicated processor could be developed to use less energy to accomplish tasks faster.
“We saw this as an opportunity to put decades of research into silicon and change how the industry processes data,” Gerban said.
APU is currently targeting Apache Spark Workloads, but its roadmap includes support for all major data analytics platforms, according to the CEO.
“We aim to be the standard processor for data processing. We hope that APU is the default for data analysis for all databases and analytics platforms, just as GPUs become the default for AI training,” Gelvan told TechCrunch.
The startup says there are many large companies testing APUs, but they refused to name it. Official product launches are set for Databricks Data & AI Summit in the second week of June. Gerban said it will be releasing the APU for the first time at the event.
Speeddata claims a specific case where APU completed its drug workload in 19 minutes. This is significantly faster than the 90 hours it took with a special processing unit, resulting in a speed improvement of 280 times.
The startup said it has achieved several milestones since its last funding, including final decisions to design and manufacture the first APU in late 2024.
“We have moved from concept to testing field programmable gate arrays (FPGAs) and are proud to have the working hardware currently on sale. With an increasing number of enterprise customers already waiting for this technology, we were ready to expand our off-the-shelf operations,” Gerban said.