He helped make Kubernetes mainstream. His next move is to make it look slow. Welcome to compute that travels to your data, not the other way around.
Bacalhau is Portuguese for salted cod. It is also the name David Aronchick gave to a serious piece of distributed-computing software. The joke is the point. The fish is preserved and shipped without spoiling, and that is roughly what Aronchick wants to do with computation: send a small, durable job out to wherever your data already sits, instead of dragging mountains of data back to some distant cloud.
For most of the last fifteen years, the industry has done it the other way. Collect the data, ship it to a giant warehouse, then rent enormous machines to chew through it. Aronchick spent a good chunk of that era helping build the tools that made it possible. At Google he led product for Kubernetes and helped launch Google Kubernetes Engine, the managed service that turned container orchestration from a science project into a default. He co-founded Kubeflow, the open-source toolkit that bolted machine learning onto that same stack.
Then he looked at the bill. Moving data is expensive, slow, and increasingly illegal in the places where privacy and sovereignty rules have teeth. So at Protocol Labs he ran a research effort called Compute over Data, and out of it came Bacalhau, and out of Bacalhau came Expanso, the Seattle company he co-founded and now runs as CEO. The pitch fits on a sticky note: stop moving the data.
It is a contrarian bet from someone with the resume to make it. Aronchick is a four-time founder who has worked at Microsoft, Google, Amazon, Chef, and Protocol Labs. That is less a career ladder than a guided tour of how modern computing got assembled, and he keeps choosing to stand at the edge of whatever comes next.
The old model treats data like cargo. A factory floor, an oil rig, a hospital, a satellite, each one generates terabytes, and the plan is always the same: haul it across the network to a central cluster, then process it. The transfer is the slow part, the costly part, and often the part regulators object to.
Bacalhau flips the geometry. The data stays put. A tiny container of compute shows up at the source, runs the job locally, and sends back only the answer. Faster, cheaper, and the sensitive bits never leave the building. For edge computing, regulated industries, and multi-region setups, that is not a nicety. It is the whole game.
It is the kind of idea that sounds obvious right up until you realize the entire cloud economy was built on the opposite assumption.
Led Kubernetes product. Launched GKE. Co-founded Kubeflow.
Twice. Early plumbing, then open-source ML strategy at Azure.
Product management for Prime Now restaurants.
Ran Compute over Data. The idea that became Bacalhau.
Senior product marketing in the DevOps trenches.
Co-founder & CEO. The current chapter, written at the edge.
Make it accessible. Strip out the complexity so the people with the problems, not just the people with the GPUs, can use it.
Valuable datasets should be shared for public good, not locked in private vaults. Data democratization, not data hoarding.
Less data shuttled across continents means less energy burned. Efficiency as a climate position, not just a cost line.
Aronchick is a relentless conference speaker, and most of it lives online. A good place to start is his own explainer on bringing the compute to the data.