Updated September 2026.
In November 2025, we announced a collaboration with Databricks. At the time, we described it in general enterprise-AI terms: connecting fragmented data, improving governance, and making AI adoption more reliable. Since then, we have focused that same reasoning specifically on precision manufacturing.
A manufacturing AI system cannot learn from data it cannot see clearly, and it rarely can, because order data sits in the ERP, geometry lives in CAD, execution strategy lives in CAM, actual performance lives on the machine, and inspection results live in a separate quality system. Each speaks its own format. That mismatch, not a lack of AI capability, is the practical barrier to a manufacturing model that actually learns.
Two different layers
Databricks provides a Lakehouse architecture: a way to bring data, analytics and AI workloads into one governed environment. That is one possible foundation for the data and infrastructure layer underneath a system like this.
Subduxion's world model is a different layer built on top: a representation of geometry, requirements, manufacturing processes and production outcomes, purpose-built for manufacturing decisions like estimating, routing and cost calculation. The two are complementary, and neither replaces the other.
We remain vendor-neutral by design. Databricks is one option for the data foundation underneath this kind of system, not a required part of the architecture.
What the collaboration focuses on
The recurring problems this partnership targets are the same ones we see across manufacturing data estates:
disconnected data sources across ERP, CAD, CAM, machines and quality systems,
inconsistent data governance and lineage across those systems,
low trust in reporting and AI output when the underlying data does not agree with itself, and
long project cycles with unclear return, because the data foundation is solved last instead of first.
Subduxion is building a world model for precision manufacturing. Our research focuses on how geometry, requirements, manufacturing processes and production outcomes can be represented and learned computationally, on top of whichever data foundation a manufacturer already runs. Read more in research or see it applied in Blake.