About
Built inside a video analytics company that hit the CPU wall.
RayoDB is a product of NPAW. This is where it comes from, what we believe, who is building it, and what we are looking for.
The story
From billions of video events to a database.
NPAW has spent more than a decade measuring video: every play, every buffer, every second of watch time across streaming services around the world. That produces billions of events, and the questions people ask of them — concurrency during a live final, funnels, retention minute by minute — are session-level questions over all of it.
We answered them the way everyone does: with the best CPU columnar engines, cubes, pre-aggregation and sampling. It worked, and every year it worked a little less well for a little more money. At some point the honest conclusion was that the ceiling was the hardware, not the software. So we built an engine for different hardware.
RayoDB started as that internal engine. It is now a general-purpose GPU analytics database — video is the first workload it was proven on, not the only one it is for.
NPAW's platform processes more than two trillion events a day for over 200 streaming services in 197 countries — up to 15 billion video plays a month.
What we believe
Three principles we will not trade.
GPU-native, not GPU-assisted.
The GPU is where the data lives and where the query runs. We do not build CPU paths and accelerate them later; we design for the device and hold ourselves to it.
Boring where it should be.
Durability, replication, backup, authentication, observability. A database earns trust with the unglamorous parts, so we shipped those first.
We publish what we can prove.
No performance number without an artifact, a workload manifest and hardware you can look up. Our own repository enforces it mechanically.
Team
Engineers who have run analytics at scale.
The team behind RayoDB built and operated NPAW's analytics platform — ingestion, storage, query engines and the products on top of them — before building this. Systems programming in Rust, CUDA kernels, distributed storage and the unglamorous work of running databases in production are what we do.
RayoDB is led by Ferran Gutiérrez Vilaró, CEO of NPAW, and built by NPAW's engineering and AI teams in Barcelona — the people who run the analytics platform behind NPAW's products.
For investors & partners
The CPU era of analytics has peaked.
The thesis. Analytics workloads are bandwidth problems, CPU bandwidth has plateaued, and GPUs offer an order of magnitude more of it in a footprint one team can run. The first generation of GPU databases proved this and stayed niche because they were proprietary and foreign to the SQL ecosystem. The window to make it accessible is now: GPUs are everywhere and the tooling is mature.
The market. Every team running analytics over billions of events on ClickHouse, Druid, BigQuery or Snowflake and paying in latency, pre-computation or money. Video and streaming is our beachhead — a market NPAW already serves — followed by product analytics, observability and telemetry.
Why us. The product exists and runs on real data; it was born inside a company with a decade of analytics operations, an existing customer base and distribution in video; and we hold ourselves to an evidence standard that most of the category does not.
What we are looking for.We are opening conversations with investors who back infrastructure companies, and with cloud, GPU and data-platform partners who want RayoDB in their ecosystem. If you build for GPUs, sell to teams drowning in events, or fund the companies that do, we would like to talk.
Join us
Work on a database that runs on the GPU.
If you enjoy Rust, CUDA, storage engines, distributed systems, or making a hard thing simple to operate, we would like to hear from you.
We hire deliberately and in small numbers: systems engineers who write Rust or CUDA, people who have run databases in production, and one person who loves explaining hard things simply. Write to us with something you have built.
Evaluate RayoDB with the people who built it.
Early access is a small program by design. Tell us about your workload.