Sovereign AI is a nation’s ability to build, deploy, and govern AI on its own terms. It is about ensuring that the infrastructure, models, data practices, and institutions behind AI reflect national priorities, security requirements, and local languages and culture.
That is the idea behind Cerebras for Nations. When we launched the initiative this past November, we defined three pillars for sovereign AI: (1) AI supercomputers, (2) model co-development, and (3) local investment and partnership with local institutions to advance AI-related education, workforce development, and policy.
Sovereignty, however, is not enough on its own. National AI systems should also enable rapid innovation, differentiation, and deployment at scale. For example, infrastructure should be of a scale sufficient for local teams to quickly iterate upon and train frontier-class models e.g. on local languages and domain-specific problems, enable fast inference to serve those and other models at scale, and sufficiently easy to deploy-operate-manage to work within national systems and governance.
It is in these areas of scale, speed, and operations where Cerebras’ systems and our approach to partnership enable velocity, differentiation, and value. Success in our view of Cerebras for Nations is putting the right high performance AI infrastructure into the hands of more users, researchers, developers, and builders around the world. In this we see an opportunity not only to help national partners achieve their individual sovereign goals, but we also see an opportunity for our collective broader society to more quickly realize AI's potential benefits. This is why sovereign AI should run on the fastest AI in the world.
When we launched Cerebras for Nations, we established the framework for our approach to partnership and co-development with sovereign government and industry partners. Since then, we have started to see the next phase take shape in production-grade models, scientific infrastructure, and national-scale compute.
Sovereign AI is national capability
We see sovereign AI as more than control. A country can only be truly sovereign in AI if it can do more than consume models built elsewhere. It needs the ability to shape the stack itself: where systems are hosted, how sensitive data is governed, which models get developed, and who inside the country can access and improve them.
Sovereign AI does not mean every nation must build a closed, fully self-contained stack or pretrain every frontier model from scratch. In many cases, the practical starting point is sovereign deployment: domestic capacity for inference, fine-tuning, and regulated workloads, with the option to co-develop or train larger models when national priorities justify it. The point is control over capacity, data flows, deployment, and long-term upgrade paths.
Nor does it mean isolation. Nations should continue to work with trusted partners, open models, and global ecosystems. What matters is that the strategic levers—capacity, governance, access, and local relevance—can remain in the hands of institutions that support nations’ sovereign missions, so that AI priorities and investments can be aligned with and deliver returns for local people and businesses, as well as creating value potential beyond the border.
Why the conversation is accelerating now
AI is becoming a foundational input to science, industry, security, and public services – and is increasingly becoming a source of economic differentiation and competitiveness in international markets. As that happens, countries are starting to view AI infrastructure as a key infrastructure resource, rather than an incidental commodity.
That shift is driving demand for differentiated AI computing infrastructure and industry partnerships that deliver speed and security in the context of national priorities and local culture, e.g. sovereign cloud or domestic on-premises systems that enable secure data handling, domain- or language-specific model development, and high speed inference to serve models at scale, accelerating local education, services, and industry. Nations increasingly want AI that is not only powerful, but also governable, secure, grounded in local context, and backed by capacity they control.
Time is of the essence. As nations’ sovereign AI priorities crystallize and demand within nations grows, international commercial demand is surging in parallel, creating competition for systems and capacity. OECD has warned that many national AI strategies still do not address domestic AI compute capacity. The UK government has said national compute capacity will shape where cutting-edge research happens, where high-growth firms choose to locate, and how quickly new applications reach the public. Stanford HAI reports that compute costs and infrastructure spending are reaching record levels.
Sovereign AI therefore requires more than policy thinking – it requires investment, action, and partnership. It is exactly this kind of partnership we had in mind when we announced the Cerebras for Nations program.
Why speed is a sovereign advantage
We believe sovereign AI only matters if institutions can actually use it. Faster training accelerates model development, reducing development time and compressing the path from basic research to deployment in production. Faster inference with Cerebras Inference correlates with time to insight and productivity, and makes national-language assistants, scientific copilots, and agentic workflows usable at scale.
Cerebras fast inference is not just about interactivity – it's about intelligence. In a fixed amount of time, higher output speed lets institutions allocate more of the response budget to reasoning, verification, and tool use while still meeting real-world SLAs. That often means better answers, not just quicker ones.
It is for this reason that performance is one of the central differential value propositions of Cerebras for Nations. We built our systems for speed because sovereign AI only becomes real when ministries, labs, hospitals, and domestic champions can use AI without waiting in line behind someone else’s priorities. Our systems are up to 15x faster than leading GPU-based solutions for inference, and our AI supercomputers have demonstrated more than 10x faster training time-to-solution than leading GPU systems.
Moreover, sovereign AI has never been just about hardware. Cerebras for Nations combines infrastructure, model development, and local ecosystem investment via partnership with nations’ government and industry sovereign partners. The goal is not simply to install systems in-country. It is to help nations build durable AI capability while keeping trained-model ownership, governance, and long-term value close to home.
Three nations are already turning to Cerebras for sovereign AI
Since launching Cerebras for Nations, three national examples stand out:
- United States with the Genesis Mission and DOE: sovereign AI as national scientific infrastructure.
- UAE with G42 and MBZUAI: sovereign model development and production inference for the Arab world with JAIS 2.
- India with G42, MBZUAI, and C-DAC: national-scale compute governed within India under the India AI Mission.
United States: Genesis and DOE extend sovereign AI into science
On November 24, 2025, The White House launched the Genesis Mission as a national effort to use AI to transform how scientific research is conducted and accelerate discovery. On December 18, 2025, DOE announced collaboration agreements with 24 organizations, and we separately announced an MOU with DOE to explore secure, scalable, and energy-efficient AI infrastructure, converged AI+HPC workflows, and novel AI “co-scientist” capabilities.
This matters because it expands the sovereign AI conversation beyond chat interfaces and citizen-facing services. In the U.S. context, sovereign AI also means scientific capability: national infrastructure that can support researchers, reduce time-to-discovery, and enable science-related workloads.
UAE: G42 and MBZUAI show how sovereign models move from development to production
In December 2025, the JAIS 2 model was launched. Developed with G42's Inception and MBZUAI's Institute of Foundation Models, JAIS 2 was announced as an open Arabic LLM and the first frontier language model both trained and deployed for inference on our systems. It was the first time a frontier-grade LLM had been trained end-to-end and deployed in production for inference on our hardware. The Jais 2 chat application runs at up to 2,000 tokens per second, and open-weight 8B and 70B variants were released publicly.
That matters for more than performance. Built with G42 and MBZUAI for the Arab world, JAIS 2 shows what sovereign AI looks like when it is rooted in language and culture. We described the model as purpose-built for the Arab world, with stronger Arabic linguistic, cultural, and contextual fidelity than general-purpose global models usually provide. For the UAE, it also fits into a broader long-term effort to build governed AI capacity across key sectors—what G42 describes as an “Intelligence Grid.”
Our partnership with G42 also shows how a sovereign AI program matures over time. We began working together under a 2021 MOU to bring high-performance AI compute to the Middle East. That collaboration expanded into Condor Galaxy in 2023, and the original Jais family was trained on infrastructure we built with G42. By the time JAIS 2 arrived, the partnership had evolved into an end-to-end sovereign AI workflow: train locally relevant models, release them openly, and serve them in production at high speed.
India: National-scale compute under Indian governance
In February 2026, G42 announced plans to establish a national-scale AI supercomputer in India with 8 exaflops of compute capacity, delivered by G42 with us in partnership with MBZUAI and C-DAC. The system is intended to be hosted within India, governed under India-defined frameworks, and serve as a foundational asset under the India AI Mission.
The significance is not just size. The project extends the sovereign AI model to national-scale infrastructure intended to support a broad domestic ecosystem: research institutions, startups, enterprises, and government ministries. It shows how sovereign AI moves from concept to capacity when compute, governance, and access are designed together—so more experimentation, deployment, and value creation can happen inside the country.
What these efforts tell us
Taken together, these announcements show that sovereign AI is not a single product category. It is a capability stack. The countries moving fastest are not just asking for better models or more chips. They are building integrated systems that combine compute, model development, governance, access, and local relevance.
That does not mean every country will make the same choices or start at the same layer of the stack. Some will begin with secure domestic inference, data control, and model customization for regulated workloads. Others will invest in frontier model development, national AI laboratories, or AI-for-science systems. Sovereign AI is not a one-size-fits-all prescription. It is the principle that the AI capabilities most important to a nation should sit on nationally governed foundations.
That is why performance matters so much. If sovereign AI is going to support public services, scientific research, industry, and national innovation, it needs high-performance infrastructure. It has to be fast enough to train useful models, serve them interactively, and let institutions spend more of each interaction on reasoning and verification instead of waiting on infrastructure.
The next chapter of Cerebras for Nations
Countries developing AI programs will need to build enduring capability across compute, models, talent, institutions, and policy.
We believe the strongest sovereign AI systems will pair local control with high performance and enduring domestic capacity. Since launching Cerebras for Nations, we have already seen that thesis take shape in the United States, where AI infrastructure is being aimed at faster scientific discovery; UAE, where sovereign models can be trained and served for the Arab world; and in India, where national-scale compute is being governed inside the country for researchers, startups, enterprises, and government. Sovereignty is the foundation. Capacity is the prerequisite. Speed is the differentiator. Together, they let nations build better models, generate higher-quality answers, accelerate discovery, support domestic innovators, and keep more of AI’s economic value close to home.
Connect with our team to find out more.