Digital Rails for AI
What India's Infrastructure Model Means for UK Governance
Why it matters: India's digital public infrastructure – Aadhaar and UPI – has processed trillions in payments while dramatically reducing benefit leakage. As organisations struggle to make AI investments deliver measurable outcomes, this model of government-owned, interoperable "rails" offers a framework that sidesteps both over-regulation and unchecked deployment. UK boards should understand why this matters before the Impact Summit shapes the next phase of global AI governance.
The infrastructure question nobody is asking
When AI governance conversations happen in London, they typically land in one of two camps. The first wants more regulation – impact assessments, algorithmic audits, mandatory disclosures. The second wants innovation space – sandboxes, light-touch oversight, market-led solutions. Both camps assume the same underlying question: how do we control AI?
India’s High Commissioner to the UK, Vikram Doraiswami, posed a different question at TechUK recently: what rails should AI run on?
The distinction matters. Control focuses on constraining bad outcomes. Rails focus on enabling good ones at scale. And if you have been watching India’s digital transformation over the past decade, the rails approach has produced outcomes that make most UK digital government programmes look rather quaint.
What “digital rails” actually means
In the late 2000s, India began building what it now calls digital public infrastructure. The foundation is Aadhaar – a national identity system that, according to official figures, now covers 1.3 billion people. On top of this sits the Unified Payments Interface (UPI), a government-owned payment railroad where private providers can plug and play for free.
The numbers are striking. UPI processes approximately 20 billion transactions monthly – an order of magnitude larger than comparable systems elsewhere. The annual value flowing through these rails sits around $3 trillion. For context, that is roughly equivalent to the entire UK GDP moving through a single interoperable payment system every year.
The architecture is deliberately simple. Government owns the railroad. Everyone else builds trains.
“The government owns the railroad, but everybody can plug and play on it for free.” – H.E. Vikram Doraiswami [1]
This is not a payments company masquerading as infrastructure. It is infrastructure that enables payments companies – and much else besides.
The leakage problem, solved
Here is where the model becomes interesting for AI governance. In the 1980s, an Indian Prime Minister observed that 85% of every rupee sent out in government benefits was lost before reaching intended recipients. Intermediaries, paperwork, corruption, and simple inefficiency meant citizens received roughly 15 pence of every pound allocated to help them.
Today, according to Doraiswami, that ratio has effectively inverted. Digital rails – identity verification through Aadhaar, payment delivery through UPI – enable precise targeting that bypasses traditional leakage channels.
During the COVID period (2019–2022), India disbursed emergency support through these rails and reportedly saved $21 billion in fiscal headroom through improved targeting alone. That is not efficiency gains from automation. That is money that would have disappeared into the system now reaching actual humans.
Analysis: If these figures represent one of the most significant governance improvements of the digital age. Not through AI, but through infrastructure that AI could dramatically enhance.
The SMTP lesson
Doraiswami drew an analogy that deserves attention. Forty years ago, the SMTP protocol for email created an open, common railroad. Different providers, different interfaces, but universal interoperability. Everyone could email everyone else.
Then came WhatsApp, Signal, iMessage – walled gardens where interoperability declined. The convenience was real, but so was the fragmentation.
“How do you get business benefits,” Doraiswami asked, “and how do you encourage people to communicate freely?” [1]
The question applies directly to AI. Are we building SMTP – shared infrastructure that enables competition and innovation? Or are we building walled gardens where whoever owns the model owns the ecosystem?
India’s answer, at least philosophically, is clear: build the rails, open them to everyone, and let a thousand applications bloom.
What this means for UK organisations
The UK faces a different context. We lack India’s scale, its developmental imperatives, and its political window for building national digital infrastructure from scratch. But the principles transfer.
Consider the typical organisation attempting to implement AI governance. They face a choice between enterprise platforms (expensive, rigid, often US-controlled) and DIY approaches (cheap, fragmented, difficult to scale). Neither option resembles a railroad.
The India model suggests a third path: shared infrastructure that enables rather than constrains. Industry consortia building common data standards. Sector-specific interoperability frameworks. Governance rails that organisations can plug into rather than build from scratch.
Analysis: The UK-India joint AI centre announced during Prime Minister Starmer’s October 2024 visit [1] could prototype exactly this approach. Testing sandboxes for interoperable AI governance – not as regulatory compliance theatre, but as shared infrastructure that reduces friction for everyone.
The multilingual dimension
There is another aspect of India’s approach that UK organisations should notice. India has 29 officially recognised languages. English speakers number around 140 million – impressive in absolute terms, but only 10% of the population [1].
If AI models learn principally in English, they serve that 10%. The other 90% get left behind.
India’s response has been to train AI models across languages. According to Doraiswami, 22 Indian languages now have functional AI solutions [1]. The Bhashini platform provides translation of government documents and court judgements, making legal outcomes accessible regardless of which language citizens speak.
One example: Marathi-language AI farming software reportedly serves 16 million users in a state of 85 million people [1]. Specialised, accessible, and relevant to daily life.
Analysis: This is not about linguistic diversity for its own sake. It is about ensuring AI deployment does not create two classes of citizens – those who can access it, and those who cannot. UK organisations serving diverse populations should take note.
Risks and constraints
The India model is not without significant caveats.
Privacy concerns are real. A national identity database covering 1.3 billion people creates surveillance capabilities that would make any civil liberties organisation nervous. The same rails that enable efficient benefit delivery also enable tracking and control. How these capabilities are governed matters enormously.
Context matters. India’s developmental trajectory, political system, and population scale create conditions that do not directly replicate elsewhere. What works for a country of 1.4 billion people building digital infrastructure from scratch may not work for a mature economy with legacy systems and different privacy expectations.
Interoperability can be mandated more easily in some contexts than others. Government-owned rails work when government has the authority and will to build them. In market-led economies, coordination problems are harder to solve.
Energy constraints apply here too. Doraiswami acknowledged that renewable energy has limits, particularly around storage [1]. India is pursuing green hydrogen and nuclear expansion to support AI compute growth – challenges the UK shares.
What to do next
For boards and executives:
Challenge your teams to articulate what “rails” your AI investments run on. If the answer is “each project builds its own,” you have an infrastructure problem masquerading as an AI strategy.
Consider whether industry consortia or sector bodies could develop shared governance infrastructure that reduces compliance friction for everyone.
For technical leaders:
Study the UPI architecture. Not to replicate it, but to understand how interoperability at the infrastructure layer enables innovation at the application layer.
Evaluate whether your AI governance approach creates walled gardens or shared standards.
For mid-market organisations:
Watch the UK-India AI centre developments. Early participation in testing sandboxes could provide governance frameworks without enterprise-scale investment.
Identify sector peers willing to collaborate on shared data standards and governance infrastructure.
Disclaimer: This article represents analysis based on publicly available statements from a TechUK event in January 2025. Statistics cited are attributed to the speaker and require independent verification. This does not constitute legal, financial, or professional advice.
If your organisation needs support developing AI governance frameworks that deliver measurable outcomes, Arkava helps mid-market enterprises build infrastructure that serves human purpose.
Contact: engage@arkava.ai