DualBird Inc., a startup using customizable field-programmable gate arrays to speed up data processing workloads, today announced that it has raised $25 million in funding.

Lightspeed Venture Partners led the investment. It was joined by Bessemer Venture Partners, Angular Ventures and Uncork Capital. DualBird will use the funds to accelerate its go-to-market efforts.

Companies prepare the information they use in analytics projects with data pipelines. Those are automated workflows that aggregate records from multiple systems, organize them in a consistent format and remove unnecessary items. Data pipelines and the analytics workloads they support can require millions of dollars’ worth of infrastructure to run in some cases.

DualBird is using field-programmable gate arrays, or FPGAs, to reduce enterprises’ hardware bills. The company has developed an FPGA-powered software platform that it claims can lower the costs associated with running data processing workloads by up to 90%. Furthermore, DualBird promises to deliver an up to hundredfold performance boost.

A processor comprises compute and memory modules that are linked together by interconnects. In standard chips, the configuration of the interconnects is fixed. FPGAs, in contrast, have adjustable interconnects, a feature that enables them to provide better performance than off-the-shelf silicon.

If a company plans to use an FPGA to run SQL queries, it could configure the chip’s interconnects in a manner that is conducive to processing such queries. That would give the chip a speed advantage over off-the-shelf processors without SQL-optimized interconnects. Furthermore, engineers can disable the parts of the FPGA that aren’t needed for a given workload. Disconnecting unnecessary circuits reduces power usage and thereby lowers costs.

According to DualBird, its platform optimizes FPGAs to run data-intensive workloads. The software can run on Amazon EC2 F2 instances, which provide up to 8 FPGAs and 192 vCPUs. A vCPU corresponds to a single thread on a physical central processing unit core. 

Many companies use Apache Spark to power their data pipelines and analytics workflows. DualBird provides a plug-in that enables customers to link its platform to their Spark environments. The company’s platform also works with Apache Iceberg, a file format often used to store the information processed by analytics applications. 

Optimizing the performance of a Spark environment usually requires engineers to manually fine-tune its settings. According to DualBird, its platform removes the need for manual tuning. Furthermore, it decreases the amount of data Spark processes during computing operations called shuffles. Reducing data volumes lower storage costs.

DualBird says its platform can also streamline artificial intelligence projects. After deploying an AI model in production, developers periodically retrain the algorithm on fresh train to keep it up-to-date. DualBird claims that its technology makes it possible to retrain models more frequently and using more data than would otherwise be possible.

“Data processing is the biggest workload still stuck on general-purpose CPUs,” said co-founder and Chief Executive Officer Amir Gilad. “It deserves purpose-built processors just like AI has GPUs.” 

DualBird plans to make its platform generally available in early 2026.

Image: Unsplash