Silicon as software - hardware-grade acceleration for the data layer nobody upgraded.
A Boston startup turning rewritable cloud chips into a Spark plug-in, so data teams get purpose-built hardware speed without ever touching a box.
For the past few years, the entire technology industry has been buying chips for one job: training and running AI models. GPUs became the most fought-over hardware on earth. But the models are only half the story. Before any model sees a dataset, that data has to be ingested, cleaned, joined and transformed - and that work, the unglamorous plumbing of analytics, still runs almost entirely on general-purpose CPUs.
DualBird was built to fix that mismatch. The company makes a cloud-native engine that accelerates big-data processing at the hardware level, using field-programmable gate arrays - rewritable chips - that already sit inside Amazon's cloud. Rather than shipping accelerator cards to a data center, DualBird delivers the acceleration as software: a lightweight plug-in that slots into a customer's existing Apache Spark pipelines.
The claim is aggressive and specific. DualBird says workloads run 10 to 100 times faster and cost 50 to 90 percent less, with no changes to a customer's code, systems or workflow. On its own site the company publishes tighter figures too - roughly 15 to 20 times the speed for the same cluster size, and 55 to 85 percent lower EC2 cost versus CPU-based Spark.
Whether those numbers hold at scale is a question the market will answer when the engine reaches general availability, planned for early 2026. What is clear is the thesis: data processing deserves purpose-built silicon, and the cloud finally makes that deliverable without a loading dock.
"Data processing is the biggest workload still stuck on general-purpose CPUs. It deserves purpose-built processors just like AI has GPUs."
Amir Gilad - Co-Founder & CEO, DualBird
Your Apache Spark and Iceberg pipelines stay exactly as they are - no rewrites, no re-tuning.
A lightweight DualBird plug-in links your Spark environment to the engine.
Rewritable FPGAs on AWS EC2 F2 instances are wired for each query; idle circuits shut off.
Shuffle data, skews and disk spills shrink - so jobs finish faster and cost far less.
The elegance is in what is not required. Classic hardware acceleration always carried a logistics tax: buy the card, ship it, install it, integrate the drivers. Because the FPGA already lives in the cloud, DualBird collapses that friction into a download. That is why its investors describe the approach as "silicon as a cloud-native software service."
DualBird's pitch lands in a market bracing for enormous spend. McKinsey has projected roughly $7 trillion in new data-center investment by 2030 just to keep pace with demand. Against that backdrop, DualBird is selling the counter-narrative: not more machines, but more work squeezed from each one.
The savings come from attacking Spark's oldest pain points - data skews, shuffle spills and manual tuning - at the level of the silicon itself, where a fixed CPU simply cannot be reshaped per query.
A cloud-native engine that applies FPGA acceleration to data processing, delivering hardware-grade speed as a managed software layer on AWS.
Links an existing Spark environment to DualBird's engine, eliminating manual tuning and reducing shuffle data, skews and disk spills.
Native support for the Iceberg open table format, running acceleration on Amazon EC2 F2 FPGA instances at lakehouse scale.
DualBird sells to enterprise data and platform-engineering teams running large Apache Spark and Iceberg pipelines on AWS - the people who own the analytics and AI-data plumbing and feel every dollar of cloud compute. The company is at the early-access and design-partner stage, with a team of around 30 to 34, heading toward general availability in early 2026.
It is B2B infrastructure software: a cloud-native acceleration engine that monetizes the performance and cost savings it delivers, without pulling customers off the open-source tools they already run.
The default alternative is simply running unaccelerated Spark on CPU instances. Beyond that sit managed lakehouse platforms like Databricks and Amazon EMR, native query engines such as Photon and Velox-based accelerators, warehouses like Snowflake, and a wave of GPU-accelerated data efforts.
DualBird's wedge is its delivery model. Its backers argue it is first to achieve "hardware-style acceleration in a software-only way" - the differentiator is not just speed, but that adopting it requires no code or workflow change at all.
Technion graduate in Electrical Engineering & Physics (summa cum laude), with 14 years in ASIC and software across high-performance networking, AI acceleration and EDA. Previously CEO of TerrainEDA (acquired). Stepped into the CEO role in 2026.
Former chip-design engineer at Amazon and Intel who set DualBird's founding vision. He died unexpectedly in April 2026; the company said his "DNA is in the core of DualBird."
Technion-trained electrical engineer with 15+ years in ASIC and FPGA development for networking, AI and compute.
Technion graduate in Electrical Engineering with 13 years in ASIC and software across high-performance networking and compute.
| Round | Amount | Announced | Investors |
|---|---|---|---|
| Seed | $8M | Prior | Lightspeed, Bessemer, Angular Ventures, Uncork Capital |
| Series A | $17M | Nov 2025 | Lightspeed (lead), Bessemer, Angular Ventures, Uncork Capital |
| Total | $25M | - | Combined Seed + Series A |
Reporting varies on how the round is described: some outlets cite a combined $25M raise, while others break it into an $8M seed and a $17M Series A. Figures above follow the company and CTech reporting.
Four Technion-trained engineers with AWS and semiconductor backgrounds set out to accelerate data processing with rewritable silicon.
An $8M seed round funds development of the cloud-native FPGA acceleration engine.
A $17M Series A led by Lightspeed brings total funding to $25M; the company details 10-100x speedups on Spark.
The engine heads for early-2026 general availability; in April, co-founder and CEO Amir Gilad dies unexpectedly and the team vows to continue his vision.
It provides a cloud-native engine that uses rewritable FPGA silicon on AWS to accelerate big-data and AI data processing, claiming 10-100x faster performance and 50-90% lower cost with no changes to existing systems.
Through a lightweight plug-in for Apache Spark, with support for Apache Iceberg, running on Amazon EC2 F2 FPGA instances - no code or workflow changes required.
$25M total: an $8M seed round and a $17M Series A led by Lightspeed Venture Partners, announced in November 2025, with Bessemer, Angular Ventures and Uncork Capital participating.
Co-founders Amir Gilad, Gilad Tal, Ehud Eliaz and Ohad Gamliel - a team combining semiconductor (ASIC/FPGA) and cloud software expertise from AWS, Intel and the Technion.
DualBird is headquartered in Westborough, Massachusetts, with roots in Israel; general availability of its engine is planned for early 2026.
Sources: DualBird (dualbird.io), SiliconANGLE, CTech / Calcalist, Angular Ventures, Axios Pro, The AI Insider, Pulse 2.0, Crunchbase, Tracxn. Figures are drawn from public reporting and company materials and are approximate where noted. Profile compiled July 2026.