There is a telling detail buried inside India's IndiaAI Mission. The Cabinet approved ₹10,300 crore — roughly $1.2 billion — to build AI computing infrastructure over five years. But the original goal was modest: 10,000 GPUs. India now operates 38,000, offered to startups and researchers at subsidised rates of ₹65 per hour. A government programme overshooting its own GPU targets isn't a footnote. It's a signal of how urgently the calculus has shifted.
AI used to be a product governments bought or regulated. Now they want to own it — the models, the chips, the cloud infrastructure, all of it. And India is far from alone.
When the Lights Go Out on Someone Else's Server
The core anxiety driving sovereign AI isn't philosophical. It's operational.
Nations are building domestic AI compute capacity not because it is the most economically efficient way to access AI capability, but because they have concluded that relying on foreign-owned hyperscaler infrastructure for AI workloads that touch sensitive government data and economic policy represents an unacceptable strategic dependency. Put simply: if the compute that runs your hospitals, your welfare systems, and your defence logistics sits on a server you don't control, in a jurisdiction whose courts you can't access — that's a vulnerability, not just an inconvenience.
The legal concern is not merely economic but constitutional. A state whose public agencies depend upon commercially provided compute from entities beyond its jurisdictional reach is structurally exposed to supply-chain weaponisation — the exercise of contractual termination rights as a form of strategic coercion.
This is the framework through which governments everywhere are now reading AI.
From Buzzword to Building Site
The race has moved off whiteboards and into concrete. Japan committed to a national AI infrastructure initiative backed by significant public capital. Canada launched a Sovereign AI Compute Strategy with dedicated funding for a nationally owned supercomputing system. In February 2025, France's President Macron announced €109 billion in total investments dedicated to AI infrastructure.
India's approach is arguably the most layered. A white paper from the Principal Scientific Adviser proposed treating AI compute, datasets, and models as Digital Public Goods. Despite generating nearly one-fifth of the world's data, India currently hosts only a small share of global data centre capacity. The mission tries to correct that from both ends — building state-owned GPU pools while simultaneously attracting private capital.
That private capital has arrived in size. Microsoft's $17.5 billion commitment over four years represents its largest investment anywhere in Asia. Google's $15 billion commitment over five years will build the largest AI hub outside the United States in Visakhapatnam, Andhra Pradesh. Crucially, both investments come packaged with sovereign cloud options — data that stays within Indian borders, processed under Indian law.
Countries are increasing investments in local AI stacks as they look to meet sovereignty goals and diversify beyond US companies, including for "computing power, data centres, infrastructure and models aligned with local laws, culture and region."
The Contradiction at the Heart of It All
Here is the uncomfortable tension that most government press releases gloss over: most sovereign AI is being built on infrastructure owned by the very companies governments are trying to reduce dependence on.
Nations build domestic large language models yet run them on foreign clouds. They achieve training sovereignty without operational sovereignty. India's own sovereign LLM — Sarvam AI's multilingual foundational model, showcased in December 2025 as India's first sovereign large language model designed specifically for Indian languages — runs on NVIDIA silicon. The chips are American. The data centres are partly foreign-funded. Sovereignty, in practice, is a spectrum, not a switch.
Even sovereign compute projects that go forward don't appear large enough to replace private-sector infrastructure for large model training. In just the first six months of 2024, Microsoft, Amazon, Google, and Meta collectively spent more than $100 billion on AI and cloud infrastructure. No single national programme comes close to that firepower.
What "Sovereign AI" Is Actually Buying
So if full independence is out of reach, what are governments actually purchasing with these billions?
Leverage. The ability to negotiate from a position of some strength. The assurance that critical workloads — census data, financial surveillance, defence logistics — are not a service agreement away from disruption. And, for countries like India, something harder to quantify: the capacity to build AI that actually reflects its users.
Platforms such as IndiaAIKosh and Bhashini are being developed as shared repositories, hosting thousands of datasets and models across sectors including healthcare, agriculture, and Indian languages. An AI trained on Indian data, in Indian languages, for Indian problems, is not something a Silicon Valley hyperscaler will build by default.
That is ultimately what the sovereign AI movement is wagering on — that the countries who own the infrastructure will shape the intelligence it produces. And that the countries who don't will simply consume whatever someone else decides to build.
Sources
- India's AI infrastructure boom: $50 billion and counting | Introl Blog
- Sovereign AI Infrastructure Is Now a Nation-State Race
- Sovereign AI: Why India Wants Its Own AI Infrastructure and Models | Tech Policy Law
- Sovereign AI: Why Nations are Treating Compute as Critical Infrastructure.
- New plan outlines how India will democratise AI infrastructure | Digital Watch Observatory
- Microsoft AI investments raise questions about long-term strategy | CIO Dive
- Why Sovereign AI Requires Sovereign Clouds | Infrastructure Guide
- India’s sovereign AI model strategy delivering results: Ashwini Vaishnaw | DD News
- Sovereign AI in a Hybrid World: National Strategies and Policy Responses | Lawfare