BharatGen, India’s flagship sovereign AI initiative, announced the launch of India’s first multimodal large language model at the AI Impact Summit in New Delhi earlier this month. The new model, which will be applied to education, healthcare, and agriculture, is the cornerstone of BharatGen’s mission to build India-first, sovereign AI systems that fully reflect and capture the nation’s immense linguistic, cultural, and societal diversity.
But an iconic photo from the summit of Indian Prime Minister Narendra Modi triumphantly raising his arms while holding hands with OpenAI’s Sam Altman and Google’s Sundar Pichai illustrates a hard truth: no country can claim complete self-sufficiency.
For India and governments around the world, the challenge is to avoid being entirely dependent on the United States and China, which have emerged as the two poles of AI power. “No country should serve only as a market where foreign companies sell their models and download the citizens’ data,” French President Emmanuel Macron said during his address at the AI Impact Summit.
The trouble is few if any countries are likely to catch up with the U.S. and China, says Keegan McBride, Director of Science and Technology Policy at the Tony Blair Institute for Global Change (TBI). “There’s nobody else who can match the resources that are required to be competitive, so then the question for government leaders is what do you do about that, given that AI is now central to economic competitiveness, national security, and public service delivery.”
New reports on AI sovereignty — one from TBI, another from the Brookings Institution, and a third by the World Economic Forum in collaboration with Bain & Co. — seek to help governments find ways to move forward.
The U.S. and China alone capture around 65% of aggregate global AI investment, according to the World Economic Forum. Their outsized presence in every element of the AI value chain reflects a full-stack approach that few economies can match, given the scale of investment needed. Faced with this reality, many leaders are now asking what steps they should take to stay in control of their country’s future. The instinctive response is often to try to do everything at home — to build fully sovereign AI and treat reliance on partners as a threat, says the TBI report. It argues that while this instinct is understandable, it is wrong. “Full self-sufficiency is too expensive, too slow and, for most countries, simply impossible,” says TBI. “More importantly, it misrepresents what sovereignty really means in a digital, global, and interconnected world.”
Sovereignty in the age of AI is a hybrid construct, says the TBI report. It is a continuum of agency that is defined based on a state’s ability to make deliberate, future-oriented choices about how AI is integrated, governed, and used in ways that protect public interests, create value, build domestic ecosystems, and preserve fallback capacity if external access is disrupted. India is an example of how a country can expand agency by governing key levers of data, identity, and digital infrastructure while leveraging partnerships to fill capability gaps, says the TBI report.
“Most countries trying to build AI sovereignty are starting from scratch. India is starting from a decade of digital public infrastructure that already serves over a billion people. The DPI wasn’t built as a defensive move, but rather was designed to solve domestic problems and to be replicated by other countries. That’s a fundamentally different starting position,” says Vilas Dhar, President of the Patrick J. McGovern Foundation and a keynote speaker at New Delhi Impact Summit events. Other ways for countries to achieve agency, ranging from using digital embassies to contributing to a third AI stack as an alternative to the Chinese and U.S. stacks, are also emerging, he says.
“What we are seeing is that the future of AI can’t be dominated by one country or ideology,” says Dhar. “Countries that are participating in the global AI economy need to have a voice in shared governance and develop their own ways to guide these tools for their own benefit. The New Delhi meeting included heads of state of many countries in the Global South, including Mauritius and the Seychelles. It was clear in the discussions that they have contributions to make. Billions of people in the global majority aren’t waiting for permission to lead on AI. Instead, they are starting with one key question: who is this technology for?”
India’s AI Sovereignty Model
India’s AI sovereignty model leverages its population’s scale, its domestic talent reservoir, and state-backed digital public infrastructure (DPI) to forge a distinctive model of AI capability. Through the India AI Mission and the broader India Stack, the government aspires to assert strategic control over data, digital identity, and foundational digital services serving 1.4 billion people, while partnering with domestic and international firms to bridge gaps in compute, semiconductors, and energy reliability. This produces a layered sovereignty posture: control where the state has strong institutional capacity (data governance, standards, India Stack, and regulatory norms), steering where it can direct investments and innovation (compute, model development, and public-sector applications), and managed dependence where structural constraints remain (chips, energy, and frontier-scale compute), says the TBI report.
What distinguishes India is the depth of its DPI, which gives the state unparalleled leverage over identity, payments, data sharing, and digital public-service delivery, says the Brookings Institution report. While the development of India’s DPI model has had its controversies and criticisms, the scale of deployment to over a billion people has garnered the attention of many countries and regions — notably Brazil, Singapore, the UAE, Nigeria, Kenya, and the EU — as a possible template for their own digital infrastructure.
India is now engaged in a process to integrate AI into its DPI as described in its AI Mission, which includes the Bhashini initiative to build its own models trained on India’s diverse languages and then use these models to improve public services like health and education. BharatGen, a government-built public AI backbone, and Sarvam, a government-backed private AI company — both funded and supported under the IndiaAI Mission — are both targeting Indian-language AI, but with different ownership models (public institution vs. venture-backed startup) and intended use cases (state infrastructure vs. commercial/enterprise products).
In practice, India’s DPI consists of adding domestic applications to an infrastructure that is heavily dependent on foreign — largely U.S. — tech providers, including AWS, Google Cloud, and Microsoft Azure, says the Brookings Institution report. India is also a large user of GPU chips from Nvidia and is now exploring the use of Google’s Tensor Processing Unit (TPU) chips and other accelerators (AMD, Cerebras, SambaNova), as well as China’s DeepSeek, for AI development. While currently dependent on the U.S., India is purposefully diversifying its dependence across suppliers. And, as it rolls out AI applications, the Brookings Institution report notes that India is also exercising managed interdependence, with specific initiatives to develop its own national cloud (GI Cloud/MeghRaj) that includes foreign providers but increasingly seeks to shift to Indian firms like Tata, Reliance, and Airtel to bolster its sovereign position. This extends to chip production through the Shakti Project led by IIT Madras, which is developing a 7-nanometer RISC-V instruction set chip expected to be ready by 2028.
India only has around 3% of global data center capacity, according to the Forum’s report. Recognising this, the IndiaAI Mission aims to establish a compute grid of 10,000 graphics processing units (GPUs) and a subsidised compute marketplace. Notwithstanding these advances and various initiatives, some Indian observers have charged that the pace of substituting foreign for sovereign infrastructure is too slow, underscoring the challenge of competing at the frontier, says the Brookings Institution report.
Building A Third AI Stack
Another strategy for AI stack builders and layer specialists is to cooperatively develop a Global DPI where each partner contributes to various strata of the AI stack. Collectively, this would amount to a third AI stack through interdependence that is distinct from the U.S. and the Chinese stacks, although some elements might still have some degree of dependence, says the Brookings Institution report. Since this infrastructure would be built through a consortium of countries and not independently by one, this collective effort is not “sovereign,” but rather a strategy for mutualizing interdependence to achieve a higher degree of collective sovereignty than what could be obtained individually. Achieving AI sovereignty on an individual basis, even when relying on foreign tech suppliers and attempting to build sovereign capabilities, would be extremely costly and prone to failure and stranded assets. Collectively, the stack builders and layer specialists have select but deep expertise in specific segments of the AI stack: Canada and Australia for critical materials, Brazil for renewable energy, Africa in data curation, the U.K. in chip design, the Netherlands in chip fabrication machine tools, Taiwan in chip fabrication, Japan and South Korea in high-bandwidth memory, France and Italy for compute, and so on. Each country has unique repositories of data in areas like health, mining, agriculture, transportation, and manufacturing.
By concentrating on their relative comparative advantage and entering alliances, these countries and others could assure each other strategic interdependence that promotes innovation and competition while collectively ensuring all members leveraging their respective advantages gain sovereignty. By aligning their efforts collectively, they could also muster the needed financial investments, aggregate demand, and a ready market for the AI models produced.
Building such alliances and orchestrating coordination would be a complex undertaking but can be a feasible strategy — both technically and financially — for most countries to achieve some semblance of AI sovereignty, says the Brookings Institution report. Antecedents for such an approach exist. The report mentions Airbus, the International Space Station, and the CERN particle physics lab.
Digital Embassies
As countries race to secure access to compute and build public infrastructure, it is becoming increasingly clear that not all nations can or should build AI infrastructure within their own borders, Cathy Li, head of the Forum’s AI Excellence Center, said during a session at the Forum’s annual meeting in Davos in January called “Digital Embassies for Sovereign AI.” In the session, the Forum announced a multi-stakeholder effort to draft a global framework for innovative and trusted “digital embassies” that allow data, and increasingly workloads, to be hosted abroad under agreed legal protections and security requirements. The idea is to help countries accelerate AI adoption when capex constraints, energy availability, or deployment timelines would otherwise delay progress.
The new global Digital Embassy Framework, developed with ministries and the private sector, seeks to enable countries to extend sovereign digital infrastructure beyond their borders while maintaining control over data, compute, and governance.
A digital embassy agreement between Estonia and Luxembourg is an example of how this can work. It was created years ago to protect Estonia’s national data from cyberattacks, Samira Gazzane, the Forum’s AI policy lead, explained in an interview with The Innovator. What needs to happen now is identifying global baseline trust principles for creating and using digital embassies in the AI era, she says.
Some countries are already positioning themselves to fulfill that role. For example, Saudi Arabia’s AI Hub law aims to “facilitate the establishment of sovereign data centers that strengthen bilateral relationships with foreign states, hyperscalers, and other digital service providers by offering continuity of service, data sovereignty beyond borders, enhanced security, and future-proofed digital infrastructure.”
The United Arab Emirates (UAE), for its part, has launched a commercial digital embassies initiative through G42, an Emirati company specializing in artificial intelligence and advanced technologies, based in Abu Dhabi, that is primarily owned by Sheikh Tahnoon bin Zayed, a member of the United Arab Emirates royal family and national security advisor. G42’s Digital Embassy framework operates across three distinct implementation models, says the company:
- Static Sovereign Environment (Digital Corridor): This model is a fully air-gapped or logically disconnected cloud environment with sovereign control across all layers of the stack. It can operate under a government-to-government designation. Conceptually, this mirrors the Vienna Convention applied to digital infrastructure. The sovereign retains full jurisdiction, governance, access control, and operational authority. Secret and top-secret workloads are deployed in this model.
- Virtual Sovereign Services (VSS): In this model, a sovereign control platform is implemented over public cloud infrastructure, including Microsoft’s. Sovereign policy enforcement remains external to the hyperscaler. Key elements include confidential compute enabling encryption in use, in addition to encryption at rest and in transit, and full lifecycle data protection; sovereign control of identity, access, policy enforcement, and key management; Core42’s Insight platform provides regulators with real-time technical controls, policy verification, auditability, and classification enforcement. Under this framework, open and confidential workloads can operate with sovereign enablement. Classified workloads requiring air-gap isolation (separation of the computing environment from the network) do not run in this modality.
- AI Compute Digital Embassy: This model applies to AI infrastructure (GPUs and AI accelerators) deployed within the UAE inside a regulated technology environment. Other nations may access this infrastructure through secure, high-bandwidth, low-latency connectivity for inference workloads. This enables countries to leverage UAE compute capacity and energy infrastructure without duplicating capital investment or waiting for domestic data center buildouts. This model allows participating countries to access compute while sovereignty over data, policy, and workload governance remains enforced contractually and technically.
Malaysia, which is becoming a data center powerhouse, is also positioning itself as a hub that can provide digital embassy services.
Such shared infrastructure initiatives extend access to compute, storage, and connectivity under enforceable safeguards so economies can scale capabilities without surrendering control over how critical data and workloads are governed, notes the Forum. But they are not automatically cheaper or safer. They accelerate access and reduce upfront capital needs, helping to reduce the AI divide, but can also introduce legal complexity, operational risk, and new forms of lock-in.
Challenges include what kind of data should be stored outside of a nation’s territory, how to adjudicate existing extraterritorial laws like the U.S. Cloud Act or GDPR, and what would be covered by diplomatic immunity, as it is not clear what that means in a digital world, says Andy Wyckoff, a non-resident Senior Fellow at the Brookings Institution and a co-author of the organization’s AI Sovereignty report.
As an impartial player, the Forum wants to play a central role in developing a shared global framework that offers a common reference point for the technical, legal, and governance requirements that underpin bilateral arrangements, reducing fragmentation and duplication, Gazzane says. As countries navigate the physical and geopolitical limits of compute and data, digital embassies may become essential to how sovereign AI is built and sustained, but only if there is trust, she says.
The Bottom Line
To capture the greatest economic and societal benefits of AI, economies must invest strategically across the AI value chain and not aim to own it entirely, say all three reports. “Sovereignty should not mean independence from all others,” says the TBI report. “It should be viewed as the ability to act strategically — with agency and choice — in a world that is irreversibly interdependent. In the context of AI, that means being able to shape how systems are used in your country, having real options over infrastructure, models, and partners, and retaining the flexibility to adapt as the technology evolves.” Case studies such as India illustrate how deliberate strategy and coordinated public-private action can build competitiveness. For most countries, including the Global South, “full AI sovereignty is out of reach,” says the Forum’s report. “Lasting advantages are not.”
This article is content that would normally only be available to subscribers. Become a subscriber to see what you have been missing
