.svg){kind=icon}
Government-owned AI for defense, also called GOTS AI or government-owned reasoning infrastructure, is an approach to fielding artificial intelligence in which the government owns and controls the data, the models, and the reasoning logic that turn its data into decisions, and can inspect, govern, run, and evolve them independently of any single vendor, rather than renting that capability inside a vendor's proprietary platform.
Also called GOTS AI, government-owned reasoning infrastructure, or sovereign AI for defense.
The distinction is not about where the software runs. It is about who controls the mission layer, the place where data, context, and decision support come together. When that layer is government-owned, the customer can open it, audit it, move it, and change it. When it is not, the government's own mission understanding is locked inside a model it cannot inspect and cannot take with it.
This sits at the center of a larger shift: the difference between a system of record (where authoritative data lives) and a system of reason (the layer that turns that data into understanding). Government-owned AI is the argument that the system of reason, not just the data under it, should belong to the government.
Key takeaways
Defense data is a national asset, and the decisions made on it are consequential and contested. When the reasoning that turns that data into understanding lives inside a vendor's proprietary platform, three problems follow. The government cannot fully inspect how a conclusion was reached, which is a problem for accountability and for accreditation. It cannot easily move its mission logic to another environment or vendor, which is lock-in. And it accumulates its hardest-won asset, the institutional understanding of its own data, inside something it does not own.
The Department of Defense has made government ownership an explicit direction, not an aspiration. The Chief Digital and Artificial Intelligence Office's Open DAGIR initiative is named for it: Open Data and Applications Government-owned Interoperable Repositories. It is built to combine data and applications from many providers into repositories the government owns and can interoperate across, so no single vendor's platform becomes the mandatory middle (see the CDAO Open DAGIR fact sheet, 2024, and the Open DAGIR technical paper, 2025). It builds on the Department's broader direction to treat data as a strategic asset and to invest in interoperable, federated infrastructure the government governs rather than depending on any single vendor's environment (see the DoD Data, Analytics, and Artificial Intelligence Adoption Strategy, 2023).
The reason the government cares is straightforward. Ownership of the reasoning layer is what preserves auditability, avoids lock-in, and keeps the compounding value of mission understanding on the government's side of the line. Government-owned AI is the model that delivers those properties by design rather than by negotiation after the fact.
Government-owned is often misread at both extremes, so it is worth stating precisely what it does and does not mean.
It does mean the customer controls the mission layer: the resolved data, the reasoning logic, and the provenance that connects the two. The customer can inspect how the system reached a conclusion, govern who and what can run it, operate it in its own environments including classified and disconnected ones, and evolve it as the mission changes, without depending on a single vendor's roadmap or hosting.
It does not mean the government builds everything from scratch. Government-owned capabilities are frequently built by industry; ownership is about the rights and control the government holds over the delivered capability, not about who wrote the code. It also does not mean the government owns a vendor's underlying intellectual property or general-purpose tooling. The line is drawn at the mission layer, where the government's data, context, and decision support come together; that layer, and the understanding it produces, is what stays government-owned.
Data rights are the mechanism that makes this concrete. In federal acquisition, the government's rights in technical data and software are asserted and negotiated per component (for example under the Defense Federal Acquisition Regulation Supplement), which is why "government-owned" is established item by item in a contract rather than assumed across a whole system.
Government-owned (GOTS) and commercial (COTS) are acquisition and ownership models, not fixed product types. The same underlying capability can often be delivered under either model; the right choice depends on the mission, the data sensitivity, and the control the program needs. The table below contrasts the models at the acquisition level.
| Consideration | COTS (commercial off-the-shelf) | GOTS (government-owned) | Fully custom-built |
|---|---|---|---|
| Who controls the mission logic | The vendor, inside its platform | The government | The government, but built once for one use |
| Data and model ownership | Licensed; often vendor-retained | Government owns and controls the mission layer | Government-owned |
| Portability across vendors/environments | Limited; tied to the vendor | Portable; not locked to one vendor | Portable but bespoke |
| Time to field | Fast | Fast to moderate (built ahead of need, delivered with rights) | Slow; built from zero each time |
| Auditability / accreditation | Varies; can be opaque | Inspectable by design | Inspectable, but re-proven each time |
| Cost profile | Recurring licenses | Owned capability, reused across missions | High up-front, low reuse |
The takeaway is not that commercial software is wrong; COTS is often the right call for well-understood, low-sensitivity functions. It is that for the mission layer, where the government's data becomes decisions, the government-owned model is the one that preserves control, portability, and auditability, and it is the model to evaluate for when those properties matter.
Government-owned reasoning infrastructure is a layer that sits on top of the systems where data already lives, rather than a new platform that replaces them. In practice it works in stages:
Because the layer is owned by the government, each stage is inspectable and governable, and the mission understanding it produces stays with the government rather than accumulating inside a vendor's proprietary model. It complements the systems of record; it does not overwrite them.
These terms overlap but are not interchangeable. The table separates them.
| Term | What it means | Relationship to government-owned AI |
|---|---|---|
| Sovereign AI | A nation controlling AI capability, data, and infrastructure within its own borders and authority | Government-owned AI is how sovereignty is achieved for a specific defense mission layer |
| On-premise AI | AI that runs on the customer's own infrastructure rather than a vendor cloud | A deployment location. Government-owned is about control and rights, which is broader than where it runs |
| Open-source AI | Software released under an open license | A licensing model. Government-owned can use open-source components but is defined by the government's control of the mission layer, not by any one license |
| GOTS (government off-the-shelf) | A capability the government owns and can reuse across programs | The acquisition term for government-owned; the contrast to COTS |
Evaluating an approach? These seven requirements are the backbone of a government-owned AI evaluation you can score against. Bring them to a scoping call and we will walk each one against your environment: request a technical walkthrough.
Torch.AI builds government-owned reasoning infrastructure: a layer the customer can own and govern, offered as a government-owned (GOTS) deployment when a mission requires it. ORCUS connects to multi-source data where it lives and normalizes it into a reusable data layer; NEXUS turns unstructured reporting into semantic representations that preserve meaning; and HALO resolves entities and fuses relationships through graph-based reasoning into an explicit, inspectable graph. Together they turn fragmented, multi-source data into coherent understanding at machine speed. You can see how these are packaged on the Torch.AI software page.
Because the layer reasons on top of and around existing systems of record rather than replacing them, the authoritative sources stay intact and interoperable, not locked inside a proprietary semantic model. This is the systems of record versus systems of reason distinction at the center of Torch.AI's approach: the reasoning layer acts as a system of reason on top of the customer's existing systems of record. In a government-owned deployment, the customer keeps control of the layer where data becomes understanding, and can inspect it, govern it, and evolve it.
Whether a specific deployment is government-owned (GOTS) or commercial (COTS) depends on the system the customer installs and purchases; what Torch.AI makes available is the government-owned model, so ownership, portability, and auditability are properties of the capability rather than concessions negotiated after the fact. To see it in practice, request a technical walkthrough and we will run it against a sample of your mission data, with provenance traced end to end.
What is the difference between GOTS and COTS? COTS (commercial off-the-shelf) is licensed commercial software, typically controlled by the vendor; GOTS (government off-the-shelf) is a capability the government owns and can operate, reuse, and evolve independently of a single vendor. They are acquisition models, and the same capability can often be delivered under either; which one applies depends on the system the customer installs and purchases.
Does government-owned AI mean the government builds it from scratch? No. Government-owned capabilities are frequently built by industry. Ownership is about the rights and control the government holds over the delivered mission layer, its data, models, and reasoning logic, not about who wrote the code.
Is government-owned AI the same as sovereign AI? They are related. Sovereign AI is the broad goal of a nation controlling its AI capability, data, and infrastructure; government-owned AI is how that control is achieved for a specific defense mission layer.
Does government-owned AI have to run on-premise? No. On-premise is a deployment location; government-owned is about control and rights. A government-owned capability can run in government cloud, on-premise, in a classified enclave, or at the disconnected edge, because ownership is defined by control, not by where it runs.
How does government ownership get established in a contract? Through data rights. The government's rights in technical data and software are asserted and negotiated per component (for example under the Defense Federal Acquisition Regulation Supplement), so government ownership of the mission layer is established item by item rather than assumed across a whole system.
Why does the Department of Defense favor government-owned repositories? To avoid vendor lock-in and keep control of its data and the applications that reason over it. The CDAO Open DAGIR initiative (Open Data and Applications Government-owned Interoperable Repositories) is built explicitly on government-owned, interoperable repositories so the mission is not dependent on any single vendor's platform.
Does government-owned AI replace existing systems? It should not. Government-owned reasoning infrastructure reasons on top of the systems of record and preserves the original data with its classification and confidence, so it complements existing systems rather than forcing a rip-and-replace.