When AI Ambitions Meet the Geopolitics of Compute

News

1. Blackstone to Take Majority Stake in Neysa with $1.2 Billion Backing

2. When AI Ambitions Meet the Geopolitics of Compute

3. AI Governance Debate Sharpens at Impact Summit

4. AI Policy Debate Turns to Power and Enforcement

5. India AI Summit Flags Urgent Reset in Skills Model

6. India Looks to Orbit for Its Next Data Centre

Control over chips, data and infrastructure is emerging as the central question in India’s push for sovereign artificial intelligence, as officials and technology executives acknowledge that talent alone will not secure autonomy.

That view shaped a discussion on ‘Sovereign AI and National Security’ at the India AI Impact Summit 2026, where speakers examined how dependence on foreign semiconductor supply chains complicates strategic ambitions.

Artificial intelligence is often described in abstract terms: a model, a chatbot, a chip. In the room, it was discussed instead as leverage and vulnerability. The panel also examined how autonomy can be defined when compute, data and models sit inside global supply chains.

Referring to the Russia-Ukraine war, Brijesh Singh, principal secretary in the government of Maharashtra, argued that digital systems have become instruments of power. “We will have to create a cognitive public infrastructure, however difficult it is,” he said. “It is visceral, it is existential, and we will have to do it.”

He described the effort as the “architecture of independence.” India, he argued, does not lack expertise. “There’s no dearth of experts in India, there’s no dearth of talent in India,” he said. The constraint is compute. “Only thing is compute, which I think geopolitically we should solve,” he added, warning that “people are using GPUs to control geopolitics.”

India relies heavily on semiconductor imports from China, Taiwan, South Korea and Singapore, with China accounting for about 30%, the largest share. That dependence underscores the vulnerability embedded in any sovereign AI strategy.

Security at Machine Speed

AI is also reshaping the threat landscape. Singh described it as a force multiplier for law enforcement and criminals alike, citing polymorphic malware that mutates to evade detection and a case in which a chief executive was impersonated on a live video call, resulting in a $30 million loss.

The challenge, he suggested, is velocity. “Do we have real time policing?” he asked. Traditional policing structures were not designed for instantaneous digital threats. In this framing, sovereign AI means building systems that can detect, analyse and respond at machine speed.

The Data Question

Several speakers argued that sovereignty begins below the model layer.

Martin Willcox, global head of analytics at Teradata, called the data foundation decisive. “There is no good AI without good data,” he said.

His prescription was practical: capture granular data, integrate it across domains and simplify its presentation so operators under pressure can interpret it and models are less prone to hallucination.

In defense contexts, integration spans air, maritime, land and cyber domains, producing vast streams of sensor data. Even smaller European nations anticipate storing multiple petabytes for next generation AI platforms. For India, the scale would be significantly larger.

Cost effective storage is therefore strategic. Willcox argued that open table formats enable reliable storage at scale while preserving interoperability. Governments cannot predict which models will dominate in a few years. A flexible data foundation allows multiple tools to operate on the same datasets.

Beyond the Flagship Model

The sovereign AI debate often centres on training domestic large language models. Willcox questioned that focus. “We should focus less on training our own models and think more about inference,” he said.

Deployable systems, he suggested, may matter more than frontier benchmarks. Open source models frequently trail proprietary systems by only a few months and offer transparency because their architecture and weights can be inspected. For many use cases, that may be sufficient.

He also highlighted the role of smaller, task specific models. Research suggests such systems can be “better, cheaper, faster, and more reliable.” Rather than rely on a single large model, organisations can deploy specialised systems tailored to defined tasks. In defence and security settings, multimodal systems analysing text, images and audio together are already used for surveillance and asset protection.

Governance Layer

Preet Saxena, director of global data and analytics at Concentrix, emphasized governance. Data residency under India’s data protection law addresses one dimension of sovereignty by ensuring data physically resides within the country.

But residency alone does not guarantee integrity. “Do we have any biasness in the data? Are we capturing in the right formats, right shape?” she asked. Poorly structured or biased data undermines performance and introduces systemic risk.

She supported open source technologies paired with “all the relevant guardrails” and suggested structured public-private collaboration, including challenge based programmes and hackathons to develop prototypes that can later scale.

Sovereignty also carries environmental trade offs. Building domestic data centres and expanding cloud infrastructure increases energy demand and carbon emissions. As India scales sovereign compute, the environmental cost becomes part of the strategic calculation.

Sovereignty as a Spectrum

Mandar Kulkarni, National Security Officer for India and South Asia at Microsoft, described sovereignty as layered.

Data sovereignty protects national data from external jurisdictional access. Operational sovereignty ensures systems cannot be shut down by outside actors. Technological sovereignty enables domestic innovation on top of infrastructure rather than mere consumption. AI sovereignty requires models that reflect local languages, ethics and context.

The third layer introduces trade offs. Full domestic development across the stack may delay deployment, particularly in sectors such as defence where capability gaps cannot wait.

He cautioned against framing sovereignty as binary. “Sovereignty is not zero or one. It is a spectrum,” he said. Different sectors and applications require different levels of control.