On 15 September 2026 Nvidia, Palantir Technologies and Cisco announced a joint reference architecture called the Sovereign AI Operating System Reference Architecture. The solution bundles Nvidia’s open‑source Nemotron models, Palantir’s Foundry and Artificial Intelligence Platform (AIP), and Cisco’s Secure AI Factory built on Supermicro rack systems. It is marketed as a complete hardware‑software stack that can be deployed on‑premises or in the cloud for regulated environments such as government agencies and highly regulated enterprises.
The announcement, reported by Heise, emphasizes that the stack is intended to give customers full control over data and AI models, a requirement that has grown as European governments tighten data‑sovereignty rules. By integrating Nemotron models directly into Palantir’s ontology‑driven workflow, the architecture promises a “secure under‑structure for sensitive data” while offering pre‑defined reference components for compute, networking, storage, security, observability and AI workflows.
Why the stack matters for regulated AI
Regulated markets have been forced to choose between on‑premises solutions that are costly to scale and public‑cloud services that raise jurisdictional concerns. The new stack aims to bridge that gap. By delivering a turnkey reference architecture, the partners hope to lower the engineering effort required to build a sovereign‑grade AI platform, potentially accelerating adoption among ministries, defense contractors and regulated financial services.
Nemotron, Nvidia’s open‑source large‑language‑model family, is positioned as the core model family for AI agents. Embedding these models in Palantir Foundry and AIP means that data residing in Palantir’s Ontology can be processed without leaving the customer’s controlled environment. Cisco’s Secure AI Factory supplies the underlying compute, networking, storage and security functions, leveraging Supermicro hardware that is already certified for many government procurement programs.
Financial backdrop of the partners
All three companies are large‑cap U.S. technology firms with substantial balance sheets, which may reassure risk‑averse public‑sector buyers. Their most recent SEC‑filed figures are shown below.
| Company | Revenue | Net income | Total assets |
|---|---|---|---|
| Nvidia (NVDA) | 177.8 bn (2026‑07‑26 to 2026‑01‑26) | 118.0 bn (2026‑07‑26 to 2026‑01‑26) | 320.3 bn (as of 2026‑07‑26) |
| Palantir (PLTR) | 3.57 bn (2026‑06‑30 to 2026‑01‑01) | 1.93 bn (2026‑06‑30 to 2026‑01‑01) | 11.68 bn (as of 2026‑06‑30) |
| Cisco (CSCO) | 49.3 bn (2018‑07‑28 to 2017‑07‑30) | 13.27 bn (2026‑07‑25 to 2025‑07‑27) | 129.6 bn (as of 2026‑07‑25) |
| Sources: Nvidia Form 10‑Q (filed 26 Aug 2026); Palantir Form 10‑Q (filed 4 Aug 2026); Cisco Form 10‑K (filed 2 Sep 2026). | |||
These figures illustrate the scale of the partners: Nvidia’s revenue exceeds $177 bn, Palantir’s is just under $4 bn, and Cisco’s most recent revenue figure dates back to 2018 but still sits above $49 bn. The disparity in size suggests that Nvidia and Cisco bring deep hardware and chip‑scale resources, while Palantir contributes a specialized data‑analytics platform that can be layered on top.
Market implications and unanswered questions
The stack could intensify competition in the emerging sovereign‑AI market, a niche that has attracted interest from traditional cloud providers seeking to offer “data‑local” services. By delivering a pre‑integrated solution, the trio may lower the barrier to entry for mid‑size governments that lack in‑house AI engineering talent.
However, several details remain opaque. The announcement did not disclose pricing, licensing terms for Nemotron models, or the timeline for broader availability beyond the AIPCon 11 showcase mentioned by Heise. Likewise, the extent of Supermicro’s involvement beyond supplying rack hardware is not quantified.
From a regulatory perspective, the stack’s focus on “on‑premises or cloud” deployment aligns with the European Union’s forthcoming AI Act, which will impose strict controls on high‑risk AI systems. If the architecture can demonstrably keep data within national borders while meeting the Act’s transparency and risk‑assessment requirements, it may become a reference point for future procurement standards.
Investors will likely watch how the partnership translates into revenue. Nvidia’s recent 10‑Q shows a net income of $118 bn for the six‑month period ending 26 July 2026, indicating strong cash generation that could fund further AI‑infrastructure development. Palantir’s net income of $1.93 bn for the same half‑year suggests it is already profitable, potentially allowing it to invest in deeper integration with Nvidia’s models. Cisco’s net income of $13.27 bn for the fiscal year ending 25 July 2026 underscores its capacity to support large‑scale hardware roll‑outs.
In sum, the Sovereign AI Operating System Reference Architecture represents a coordinated attempt to meet the growing demand for AI solutions that satisfy data‑sovereignty mandates. Its success will hinge on pricing, ecosystem adoption, and the ability of each partner to deliver on the promised “full‑stack” experience.