Tower raises $6.4m to be the Databricks of the Python Era

Brad Heller and I started Tower because we saw a major shift happening in data engineering.

For years, the field had been shaped by overly complex big data platforms. But suddenly, a new generation of data engineers was emerging: building with open-source tools, writing data applications directly in Python, and moving much faster than the infrastructure around them was designed to support. The tooling had changed, but production infrastructure remained complex. There was still no great platform for data engineers who were no longer interested in a long-term relationship with Spark. Too many teams were duct-taping pipelines together on GitHub Actions, AWS Lambda, and assorted internal scripts, spending valuable time on operational problems that should have been invisible.

That was the gap we set out to fill.

Then another shift hit, even faster than we expected: AI-assisted coding.

Tools like Claude made it dramatically easier for data engineers to build their own data applications. But they also made the missing piece much more obvious. Once a data engineer has Claude-coded ("Clauded"?) their app, where does it actually run? How does it get packaged, deployed, instrumented, monitored, and improved when something breaks in production? The "last mile" of AI-assisted data engineering still does not exist.

The Data Backbone Built to Scale

That is the world Tower is built for.

Tower gives builders the core primitives they need to go from idea to production: a Python-native orchestrator, Python execution compute, and analytical storage. In practice, that means a modern data engineer can write code in the tools and libraries they already use, ship it into a secure cloud environment, and operate it as a real production system without getting dragged back into infrastructure pain. We handle the packaging, the runtime, secrets, observability, and the mechanics of scaling and operating the system. Users can bring existing Python code with only a simple config file. They do not need to rewrite everything into a proprietary framework just to get to production.

The Python Data Stack

To understand why this matters, it helps to look at what the modern data stack has actually become.

The modern data stack is increasingly just ... the Python data stack. Polars is Python. dbt is a Python library. dlt is Python. LangChain is Python. Most AI inference and training code is Python. 

The ecosystem has become incredibly productive, but production infrastructure has lagged behind. It is easy to create something useful on your laptop. It is still much harder than it should be to turn that code into a reliable data system that runs every day, emits logs and metrics, stores data correctly, and recovers gracefully.

The Feedback Loop

And that gap becomes even more obvious once software actually meets production.

One of the biggest operational shifts we are seeing in the AI era is that building the first version of a pipeline is no longer the hard part. The hard part is the loop that comes after. You run the software in production, it fails for some reason nobody anticipated, then those runtime failures have to be translated back into code changes, reviewed, approved, and redeployed. That feedback loop between production systems, developers, and coding agents is becoming one of the defining problems of modern software. We think it will be especially important in data engineering, where “vibe quality” is not enough. If the pipeline runs your business, it has to work every time.

That is why Tower is not just a runtime. It is an attempt to build the environment where this new mode of software creation can actually function.

The New Users

That shift is not only changing how software gets built. It is also expanding who gets to build it. It is not just traditional data engineers anymore. We are seeing people build pipelines, agents, notebooks, and dashboards who would not have touched this workflow a year ago. They are "tech-curious" business users: founders, marketers, product managers, operators. They are increasingly capable of defining the requirements for their own data systems and, with the help of trusted technical advisors and AI tools, turning those requirements into something real.

Whether every non-technical business user wants to become an engineer is still an open question. Probably not. But it is already clear that the boundary around who gets to create software is moving. Claude is changing who can create software. Tower is making sure they have a place to run it.

Owning Your Data (Oil)

If more people are going to build software this way, the platform underneath them cannot look like the old world of closed, heavyweight data systems. 

We believe the future is more open, more interoperable, and much more pragmatic. Tower is built around open technologies and a dramatically simpler architecture. Customers can run Python workloads on Tower, store analytical data in an open Apache Iceberg-based lakehouse, and still work alongside tools like Snowflake or Databricks where it makes sense. We want customers to stay in control of their own data and products, not get trapped inside a platform abstraction they did not ask for.

The Middle (Ground)

That belief in openness and simplicity also leads to a clear view of who Tower is for.

We think there is a huge part of the market that is underserved by both ends of the spectrum: too sophisticated for random scripts, too practical to buy a million-dollar big data platform. The midsize company with one strong Python-native data person. The software vendor that needs to ship data products faster. The business that wants a full data platform without hiring a team of Java Spark engineers from 2015. For them, the right answer is not more infrastructure complexity. It is less.

Our Raise and Future Plans

We have reached the point where the foundation is working, and now the opportunity is to scale it.

Today, Tower supports traditional batch jobs, short-running Lambda-style functions, and interactive applications like notebooks, dashboards, and API endpoints. The orchestration layer and runtime have come together well, and we are investing more heavily now in the storage side of the platform and in the collaboration layer between the three roles we believe will define this category going forward: data experts, “tech-curious” business users, and AI assistants.

That work is now accelerating with fresh backing. To date, we have raised $6.4 million to build Tower into the data backbone for this new era of software creation. We are incredibly lucky to be supported by DIG Ventures, Speedinvest, Flyer One Ventures, Roosh Ventures, Celero Ventures, and Angel Invest, as well as many amazing angels, including MotherDuck’s CEO Jordan Tigani and Datadog’s CEO Olivier Pomel, Harvey.ai’s former VP of Engineering Ben Liebald, and Taktile’s CEO Maik Taro Wehmeyer.

We still feel like we are at the beginning of this shift. We are now a team of 12, with people from Greece, Turkey, Sri Lanka, the US, Canada, the UK, Ukraine, and Germany. A mini-United Nations! We are hiring principal and senior engineers in Berlin and London, and GTM talent across London and the US.

If this resonates, come build with us!