I grew up in a predominantly tribal, underdeveloped, Naxal-infected district. I did not have a lot of people around me I could look up to. When I was in Class 7, my family got me a Reliance Netconnect internet dongle, which was considered high speed back then. And that opened up an entire new world for me.

I discovered John and Hank Green on YouTube, discussing the personal and the political, reassuring me that being nerdy isn’t something to hide. I would spend hours falling down rabbit holes of random YouTube videos, some profound but mostly puerile. I wrote on Blogspot, happy just to see my words sitting beautifully on a digital real estate that was entirely mine. Later, when college gave me real broadband, I devoured books from Genesis Library and movies from the Pirate Bay. I did umpteen free MOOCs, squeezing Coursera’s financial aid like the last bit of toothpaste from the tube. The internet gave me the cultural vocabulary to pass off as someone who belonged in serious places. It let me live without my pincode tattooed on my forehead. The Internet is a beautiful thing. Everyone should have it.

Like a Router in the Sky

Last week, the Government of Meghalaya signed a letter of intent with Starlink India to pilot satellite-based internet across the poorly connected state. Meghalaya is not the first state to sign such an agreement (Gujarat, Goa and Maharashtra have done so earlier), but it may be the one handpicked to showcase what Starlink can bring to unconnected regions of post-Jio India.

And Meghalaya needs it. According to a parliamentary reply by the Ministry of Communications in December 2022, 2,040 of the state's 5,623 villages, over 36 per cent had no mobile coverage at all. Under BharatNet, the government's flagship programme to connect every gram panchayat with broadband, only 6.5 per cent of Meghalaya's gram panchayats were connected via optical fibre as of 2023, against a national average of roughly 79 per cent. Another 571 GPs are classified as “service ready” on satellite, using older geostationary VSAT links with approximately 600 milliseconds of latency, which BharatNet’s own technical documentation acknowledges is unsuitable for low-latency applications. The remaining 1,162 GPs, 62.7 per cent of the total, are not service ready at all. And where connectivity does exist, it barely works. Digital payment transaction failure rates in the Northeast as a whole are 1.5 to 2 times the national average, driven by three times higher network downtime and 50 per cent slower speeds. Even in urban centres across the region, median wireless download speeds fall below the 20 Mbps threshold that the UN considers the minimum for usable broadband.

Yes, Starlink is great news. Perhaps the best connectivity news for underserved India in the last decade. But it’s also complicated.

Explore the data in full

Star Bucks

Starlink doesn't come cheap. Its India pricing is not yet known. A listing that briefly appeared on its website in December 2025 showed ₹8,600 a month and a ₹34,000 hardware kit, but the company said it was dummy test data. In Nigeria, a Starlink monthly subscription costs roughly $39 to $49, which works out to over 80 per cent of the minimum wage. In Kenya, annual Starlink costs range from 40 to 195 per cent of gross national income per capita, depending on the plan. We can expect promotional pricing at launch, possibly under $10 a month, to build a subscriber base fast. But even at the most aggressive pricing, the economics of satellite broadband do not make sense in regions that need it the most.

In Kenya, Karibu Connect, Starlink’s first authorized reseller in the country, uses a single Starlink terminal as backhaul for solar-powered community Wi-Fi hotspots, each covering roughly a square kilometre. Such a model, one that aggregates demand, shares costs, and distributes access through community infrastructure, is what India’s underserved regions need.

The Fault In Our Star(link)s

A constellation sees no national boundaries. It takes a State to point it at a map and teach it Borders 101.

In December 2024, unauthorised Starlink terminals were seized in India’s strategically sensitive Andaman and Nicobar Islands. That same month, a Starlink router and antenna, most likely smuggled in from Myanmar, were seized from the East Imphal district of Manipur. The north-eastern state has been embroiled in ethnic conflict since May 2023, and internet shutdowns have been imposed as a law and order measure. A Starlink terminal in the hands of a militia or an insurgent group renders an internet shutdown ineffective.

And India is not being paranoid for no reason.

Musk himself shipped thousands of Starlink kits to Ukraine within days of the Russian invasion, funded largely out of SpaceX's own pocket. Then, in September 2022, in the middle of a planned Ukrainian military operation against Russian forces in Crimea, Elon Musk unilaterally disabled Starlink access in the operational zone. As the journalist Ronan Farrow documented, it was one man’s risk assessment overriding a sovereign government’s military strategy.

In Brazil, Starlink terminals were found in the hands of illegal gold miners deep in the Amazon and among criminal networks in favelas. When a Brazilian Supreme Court judge ordered a ban on X, Starlink initially refused to comply, calling the order "illegal." Only after the court threatened to freeze Starlink’s bank accounts, did Starlink back down. In Iran, an estimated 50,000 smuggled terminals became one of the few channels for information during the 2025 internet shutdown. The state responded with drone surveillance, door-to-door seizures, and 25-year prison sentences.

Starlink increasingly shapes how states wage war, manage internal unrest, and fight crime. A government can ban Huawei or TikTok, and life goes on. But you cannot take away the internet from an entire region without consequences that ricochet through everything — schools, hospitals, banks, businesses, governance, and obviously, the lives people actually live. If Starlink fills the vacuum where no other connectivity exists, then acting against the company means disconnecting your own people. And that is a digital Strait of Hormuz.

Loading... (2021-2026)

India, to its credit, isn’t taking any chances either. It has adopted a full bomb squad approach, with all protective gear, making one careful snip at a time. Starlink opened pre-orders in February 2021, signed up over 5,000 customers, and set a target of 200,000 terminals by December 2022. Within months, the government issued a public warning asking citizens not to sign up, the country head resigned, and all pre-orders were refunded. The licence application with DoT, which would allow it to offer satellite internet services commercially in India, filed in November 2022, took over two and a half years to process. Along the way, Starlink’s progress required a Modi-Musk meeting in New York, a Modi-Trump-Musk meeting in Washington and a spectrum allocation battle that pitted Ambani and Mittal against Musk before both ended up signing distribution partnership deals with Starlink.

The 29-point compliance directive issued in May 2025 is arguably the most demanding laundry list of regulatory requirements any country has imposed on satellite internet providers. It mandates real-time location tracking of user terminals, data localisation on Indian servers, metadata sharing with law enforcement, website-blocking capability on demand, traffic routing exclusively through Indian ground stations, surveillance zones near international borders, compliance with internet shutdown orders, and 20 per cent indigenous manufacturing of equipment. As Irina Tsukerman, a national security lawyer, put it:

Starlink’s entire model is built on global interoperability, low latency, and centralized deployment. India wants the opposite: fragmented control, localized nodes, and bilateral oversight.

The IN-SPACe authorisation, granted in July 2025, caps the constellation at 4,408 satellites. The government has capped maximum connections at 20 million users. As of April 9, 2026, Starlink has cleared all security requirements set by law enforcement agencies. Digital Communication Commission (DCC) approval and Cabinet sign-off are potentially weeks away. The four-year obstacle course is nearly over, except for a disagreement between TRAI and DoT on spectrum pricing.

India has also licensed competitors — Eutelsat OneWeb, Jio-SES, AST SpaceMobile via Vodafone Idea, with Amazon's Kuiper in the queue. But satellite spectrum has not been allocated to any of them, and without it, no one can actually operate. This diversification strategy needs to be actively pursued. No single foreign provider should become the sole source of satellite connectivity for India’s unconnected regions. Even Taiwan, an American ally, wary of Musk’s business ties to China where Tesla operates several large factories, chose to partner with Eutelsat OneWeb instead of Starlink.

A Moonshot Beyond Starlink

So we allow Starlink in the short term. We regulate hard, ensure enough competition in the medium term. But what about the long term?

Countries are waking up to the risk of dependence on a single foreign satellite internet provider, especially one led by Elon Musk. And some of them are placing expensive bets to tackle it. China is building Guowang, a 13,000-satellite sovereign constellation. The European Union has committed over $11 billion to IRIS², a sovereign satellite system with both secure government and commercial broadband functions. Russia and Ukraine are building their own constellations for strategic autonomy.

India, conspicuously, is doing none of this. Apart from one stray remark by an ISRO official at a conference in August 2025, claiming 140 LEO satellites had been “worked out” on paper for urban broadband, the government has been silent. India’s official space policy since 2020 has moved decisively in the other direction: opening the sector to private players, transferring ISRO technology to industry, and licensing foreign operators through IN-SPACe.

So here is where I bring up Digital Public Infrastructure.

I know. DPI has become the hammer that makes everything look like a nail, especially in Indian policy spaces. But before you roll your eyes, hear me out.

India doesn’t need one company with 7,000 satellites to compete with Starlink. It needs an ecosystem where twenty startups with 30 satellites each can collectively function as one network. Let’s call it Open Rails for Broadband Internet Transmission (ORBIT). We standardise the user terminals so a single dish works across constellations. We standardise the inter-satellite links so satellites from different operators can handshake and route data between each other. We build and operate shared ground stations so every small operator doesn’t need to replicate SpaceX’s global gateway network. This can be done by a consortium of operators working with an ISRO subsidiary, like NPCI does in case of UPI. Individual startups then innovate over what remains: antenna design, coverage optimisation, uplink and downlink capacity. The shared infrastructure handles the interoperability, the spectrum coordination, the traffic orchestration. Spectrum allocation at preferential rates, government procurement and subsidized satellite launches by ISRO can be used as carrots to incentivize participation in the network. Such a DPI for satellite internet would drop the minimum viable constellation from thousands of satellites to dozens.

The security concerns are real and I have not tried to push them under a compassionate rug. But we cannot keep a large part of our population away from the internet. If national security requires keeping foreign satellite providers at arm’s length, then the State must fund the alternative itself. If the tower economics in the Northeast don’t make sense for a Jio or an Airtel, it is the State that has to make it work. But it hasn’t. You cannot invoke sovereignty to slow-walk Starlink and then offer nothing in its place. You cannot put the burden of sovereignty on the already underserved. Because, to repeat what I said at the start, the internet is a beautiful thing. Everyone should have it.

*The usual disclaimer: views here are entirely my own.

On January 10, 2020, the Reserve Bank of India (RBI) published its National Strategy on Financial Inclusion (NSFI) 2019-2024. The strategy notes:

Periodic evaluation of financial inclusion policies through monitoring of financial inclusion parameters provides policy makers and stakeholders with necessary insights to understand the achievements made in the country and to address issues and challenges through a coordinated approach.

To deliver on this, the RBI released its Financial Inclusion (FI) Index in August 2021. The FI Index was designed as a comprehensive, single-number measure incorporating data from banking, investments, insurance, postal services, and pensions, built in consultation with the government and sectoral regulators. It maps financial inclusion onto a scale of 0 to 100, where 0 represents complete exclusion and 100 indicates full inclusion. It reflects cumulative progress over time.

The index is composed of three sub-indices:

1. Access (weighted at 35%), which measures the availability of financial infrastructure;

2. Usage (45%), which captures how actively that infrastructure is used; and

3. Quality (20%), which accounts for financial literacy, consumer protection, and inequality in the distribution of services.

Here is what the index has shown since:

The number has indeed gone up, from 43.4 to 67.0 in eight years. And that’s a great thing.

In the five years for which the RBI published the breakdown up to the sub-index level, a few things clearly stand out. Access has always been far ahead of Usage. In March 2021, Access stood at 73.3 while Usage sat at 43.0. The composite index, by blending the two into a single number, obscures this gap entirely. Quality is the only sub-index that has ever declined, from 53.8 in March 2020 to 50.7 in March 2021. But the headline FI Index still went up, from 53.1 to 53.9. The composite absorbed a deterioration in the quality of financial inclusion and still reported progress.

But that is fine, as long as you also get the sub-index data.

And then, voila. Starting with the March 2022 release, the RBI stopped publishing the Access, Usage, and Quality scores. The annual press release now reports only the headline number, usually accompanied by a single sentence noting “growth across all sub-indices.” How much growth? Which sub-index? We have no way to know.

If the FI Index was built to give policymakers the insights they need to address challenges, why does it only seem to work as a vehicle for reporting progress? We got the detail only as long as the numbers went up. The only insight I get from the index is that the arc of history bends towards progress. And that is something I don’t need the Reserve Bank of India to tell me.

I got 97 parameters

The FI Index is built on 97 underlying parameters. Not only are their values never published but we do not even know precisely what all of them are. The only public indication of what goes into the index comes from a 2021 RBI Bulletin article, which describes the broad dimensions but not the specific parameter values or their individual scores. You cannot tell whether the improvement is coming from banking, insurance, pensions, or postal services. And inside banking, whether the improvement is from increasing banking correspondents or actual branches, both of which can give different quality of services to the end user. The index has no state or district breakdowns, no gender disaggregation either.

Then there is the Usage sub-index, which earns its label only partially. The NSFI 2019-2024 gives us some indication of what it may contain: formally banked accounts, adults with credit, adults with insurance, cashless transactions, frequency of account usage, remittances. These are not wrong as usage metrics. But they are the easier ones to count, drawn mostly from data that banks are already reporting to the RBI anyway. Whether there are richer demand-side indicators in the mix, actual surveys of whether people are actively and meaningfully using financial services, we will never know. Running a demand-side financial inclusion survey would not be trivial. But the RBI has shown it can commission complex surveys, like the ones that drive its monetary policy.

We have been here before.

In 1990-91, the gross enrolment ratios in India's primary schools were 83.8% and in elementary schools 78.6%. Against that, literacy rates for ages seven and above in 1991 were just 52.2%. It had become clear that enrollment was not the magic potion anyone thought it was.

In response to this crisis, Pratham Education Foundation started bringing out the Annual Status of Education Report (ASER) surveys. Every year, Pratham comes out with the ASER report, forcing all of us to, at least for a moment, face the utter pit that school education in India is. For example, the latest edition, ASER 2024, found that nearly 30% of school-going children in rural India still cannot read a Class II textbook. The Research on Improving Systems of Education (RISE) puts it well:

Established in a time when the primary assumption was that getting children into schools (inputs) would automatically result in children learning (outputs), ASER not only questioned this assumption but also highlighted the flaws in it.

ASER is still the only survey of learning in India that tells you what is happening at the district, state and national levels. It has brought learning to the forefront of education policy discourse and driven interventions to overcome the challenges.

Pratham recognised that in order to help children learn in school, we must first measure what children have already learned. The FI index falls short of that. It measures whether the pipes are being built. The water is someone else's problem.

The Proof of the Pudding

The FI Index is still stuck on enrollment. It tells you how many people have been brought into the formal financial system. It does not tell you whether any of them can manage a health shock, plan for retirement, or meet their monthly expenses without borrowing from a neighbour.

Dvara Research and XKDR Forum, writing in the Inclusive Finance India Report 2022, went and asked those harder questions. They built a framework that measured three distinct things: financial access, actual usage of financial products, and whether any of it was translating into household wellbeing. Then they applied it to India’s data in the World Bank’s Global Findex survey.

Inputs, which stand for access and ownership of financial products, scored 0.78. India has spent a decade opening accounts, enrolling people in insurance schemes, and pushing pension products through government programmes, sometimes without their knowledge. All those press releases are not lying.

Outputs, which show whether people are actually using what they own, scored 0.19. Outcomes, which show whether any of this is improving people’s financial lives, scored 0.30. Bank account ownership in India is widespread, the chapter notes, but most of these accounts are dormant.

In a separate study using CMIE’s Consumer Pyramids data covering 174,000 households, Dvara Research found that measuring financial inclusion through bank account ownership alone overestimates inclusion by 30 points for rural India. A bank account penetration score of 0.84 collapses to a composite score of 0.55 once you account for the full bouquet of what financial inclusion actually means: credit, insurance, pensions, investments.

To add to the problem, the rank correlation between supply-side state rankings and demand-side state rankings was 0.13 for rural India. States that look good on supply-side metrics do not necessarily look good when you ask households what they actually own, use, and need.

PDFs All the Way Down

The farmer in Dadenggre suffers from very different kinds of exclusion than a street vendor in Delhi. Excluded in different ways, for different reasons, needing different things. The unit at which financial inclusion problems actually get solved is not just the national or state level but also the district and block level. To do anything about financial exclusion, you need to see it at that level.

This data is not entirely absent. Every quarter, State Level Bankers Committees (SLBCs) produce detailed background papers with district-level financial inclusion statistics including branch coverage, credit-deposit ratios, account penetration, Business Correspondent deployment. The papers are produced on schedule, reviewed by bankers and government officials, and then infrequently uploaded on poorly maintained websites.

But the SLBC data is produced for the regulators and state government officials who sit around that table. A civil society organisation working on last-mile delivery, a researcher trying to understand why a particular district consistently underperforms on credit-deposit ratios, a journalist asking whether financial inclusion expenditure is actually reaching anyone: none of them have a clean, structured, comparable way to engage with this data. It exists, when it does, in PDFs, ZIPs, RAR files, excel sheets, sometimes scanned ones (I am talking to you, Bihar). It changes format across quarters. It changes shape across states.

I ran into this directly while working on Project FINER, which stands for Financial Inclusion in the North East Region (yes, I would be a great asset in any government scheme naming department). It is my attempt at making machine-readable financial inclusion data available down to the district level, and eventually the block, with tools to visualise and analyse it without any technical background. I want to expand it to cover all of India eventually. But most states do not make this data available in any usable form.

The Northeast is, oddly, better served than the rest of the country on this front. SBI has been publishing more structured SLBC data for the region on a separate site. For most of the rest of India, you are back to writing essays on why open data is the need of the hour.

Better Late

The Indicus Centre for Financial Inclusion (ICFI) newsletter, which you should be reading, flagged this sentence from the recently released National Strategy for Financial Inclusion (NSFI) 2025-2030:

To begin with, during the period of NSFI 2025-30, a new series of FI-Index with broader coverage and State/UT level disaggregation may be constructed and published.

This is, quietly, an acknowledgment that the current index has not been enough. Now we wait to see how this turns out. The previous NSFI has itself mentioned how "there is also a need to seek views and insights from the end users of the products themselves." But no such exercise was carried out, at least nothing that is publicly available.

ICFI, in a 2023 white paper, recommended that monitoring should move to a body independent of the Ministry of Finance, to avoid what they call “the inherent conflict between its roles as strategy formulator, implementor, and also monitor.” They suggest NITI Aayog, working with the Central Statistical Organisation. The people designing the policies should not also be the ones grading themselves. If past is any indicator, this may just be wishful thinking.

But I live on hope. The arc of history, and all that.

*The usual disclaimer: views here are entirely my own.

AI is now in everything, everywhere, all at once. And it is remaking our world with an impatience matched only by our indifference to the digital divides we never closed. Dario Amodei, CEO of Anthropic, wrote 'Machines of Loving Grace' in October 2024, perhaps the most serious attempt by an AI company leader to imagine what a world transformed by AI could actually look like. He talks of a world where AI compresses decades of biological progress into years, where he imagines the developing world growing at rates that could bring Sub-Saharan Africa to China's current per capita GDP within a decade. “Most people are underestimating just how radical the upside of AI could be”, he writes. But intelligence, he warns, 'isn't magic fairy dust'. AI ultimately lives in this world, with the slowness of our physical world and constraints from us, humans. The latter, I'd argue, includes our governance structures, political economy, and the technological inequalities we have decided to live with. That honesty matters. Because however powerful AI becomes, it cannot simply cut through the mess of the world it lands in.

Being the Class Topper Is Not Enough

Most people watching the AI race believe it is all about frontier labs: the Anthropics, the OpenAIs and the DeepSeeks of the world. That the first country to create what former US National Security Advisor Jake Sullivan called “God-like” Artificial General Intelligence will “win.” Jeffrey Ding, author of Technology and the Rise of Great Powers, thinks otherwise. He turns to past technological revolutions. The Soviet Union launched Sputnik, graduated twice as many scientists and engineers as the US, and spent a larger share of its GDP on R&D. It still collapsed. Conversely, the US was no class topper in the Second Industrial Revolution; Germany led in chemistry, Britain in basic research. Yet the US emerged as the clear superpower. It built a broad base of engineers, institutionalised chemical engineering as a discipline at MIT before its European peers, and created dense linkages between industry and academia that transferred knowledge from frontier labs to factories.

What matters, Ding argues, is not whether you create the best technology but how well you can diffuse it throughout the entire economy, raising productivity across the board. AI is a general-purpose technology (GPT), something which spreads throughout an economy and isn’t limited to any one industry. For GPTs, Ding argues, diffusion matters more than technological innovation for determining overall national power. It matters less whether you can build your own Silicon Valleys than if your Bengalurus ever reach your Byrnihats.

The Pragmatist Indian

India understands this, by choice or compulsion. Probably more compulsion than choice. When you cannot afford to compete with the frontier labs, you develop a sophisticated argument for why the frontier is the wrong destination anyway. As Srikanth Velamakanni, co-founder of Fractal Analytics, put it, “India’s Rs 10,000 crore AI mission is roughly what OpenAI spends in a weekend.” But the argument, it turns out, is largely correct.

Speaking at Davos, Ashwini Vaishnaw, India’s Electronics Minister, argued that nearly 95% of all AI use cases can be addressed with 20-50 billion parameter models and that ROI comes not from model size but from deployment. At the AI Impact Summit, he elaborated this point:

“Today models have already become commoditized. There are good models which are just maybe four or five months or maybe six months behind the topmost frontier models. They are now available in open-source.”

Nandan Nilekani, whom many have taken to calling India's de-facto CTO, is all on board with this argument. Nilekani talks of diffusion pathways, replicable playbooks built from lived deployment experience. Lessons from Maharashtra’s agricultural AI initiative, MahaVistar, were adapted for use in Ethiopia in three months. A parallel playbook reached dairy farmers through Amul in three weeks. Infosys has announced a partnership with Google, the Gates Foundation, and UNDP to develop 100 such pathways by 2030. “All of us who have a stake in AI being valuable to humanity have to accelerate diffusion,” he warned. “Or the consequences will be bad.”

Nilekani is also betting his money on “spoken AI in dialect of your choice”. Keyboard literacy was the bottleneck for PC-era services. Touch screens lowered it for smartphones. Voice in vernacular languages would remove it almost entirely, making AI accessible to populations that the loving grace of technology has never reached.

Perhaps. However, in government, the hand that writes the cheque rarely reads the keynote address. Ding argues that diffusion strategies are structurally hard to sustain. The benefits of diffusion — broader engineering talent, stronger industry-academia linkages, vernacular voice interfaces for farmers — accrue to dispersed interests. No single industry lobbies for them. They do not give you the reel material that a frontier model beating Claude at key benchmarks would. The result is a constant underinvestment in what actually matters for diffusion to work.

Everyone Else Gets a Hole-in-the-Wall

The good and the bad news is that none of this is new. Most of what AI needs to be useful, functional literacy, affordable devices, reliable internet, practical digital skills, was already pending work. Prerequisites that, had they existed, would have anyway lifted productivity and welfare on their own, without any AI in the picture. But AI promises to provide a much greater return on this long-deferred investment. And conversely, a much bigger chasm between the haves and have-nots if the to-do list of prerequisites remains unticked.

I like to think of the diffusion journey as a series of chains, each one a prerequisite for the next. And the breaks in those chains exist at every level: between states, between districts within those states, between households within those districts, and between individuals within those households. And cutting across all of this, by caste and gender, the gaps can be just as wide. Some have achieved basic literacy and numeracy for all. Others haven't. Some have a smartphone. Others don’t. Within households, the phone often belongs to one person, usually the adult male member, which means everyone else’s access is borrowed at best. Among those with a device, some have reliable internet. Others are on a connection that works only in that one corner of the house. Even where internet access exists, it may not be stable or fast enough. Then comes awareness of AI and what it can actually do. Then usability, which, for the vast majority who aren't comfortable in English, improves dramatically if the interface is in their own language, in voice, not text. And finally, the output must actually be useful, not marginally better but transformatively better, which AI can deliver but only if the models have been trained on the specific realities of Indian lives, Indian languages and Indian problems.

Jan van Dijk, a Dutch sociologist, mapped a similar logic in his work on the digital divide. He identifies four sequential barriers to technology access: motivational, material, skills, and usage. The key insight is that you cannot skip levels. A person without motivational access will not seek material access. A person with a device but no skills will not achieve meaningful usage. Each layer is a prerequisite for the next. Right now, several of those links are broken, and not the same links for everyone. Even if the frontier labs manage to 1000x the power of their AI models, it changes very little at the last mile.

Mark Warschauer, in Technology and Social Inclusion, opens with a case study close to home. This is the story of the Hole-in-the-Wall experiments conducted by Sugata Mitra, whom some people call a visionary and some call a charlatan. Mitra’s team installed computer kiosks in one of New Delhi’s poorest neighbourhoods. The idea was that children would teach themselves. But the internet rarely functioned, no content was available in Hindi, the only language the children knew, and they spent almost all their time on paint programs and games. Parents complained their children’s schoolwork was suffering. Warschauer’s verdict was blunt: minimally invasive education was, in practice, minimally effective education. The failure wasn’t the technology. It was the absence of everything else: content in the right language, human instruction, community involvement.

His second case study travels to Ennis, Ireland, which won a national competition to become an “Information Age Town” and received fifteen times more funding than the runners-up. Every family got a computer and an internet connection. The result was disastrous. The unemployed were told to sign in for welfare payments online and couldn’t figure out how the equipment worked. They returned to the office anyway. Some sold the computers on the black market. The three runner-up towns, with one-fifteenth of the money, actually had more to show for it, because they had spent on awareness, training, and existing community networks rather than on hardware.

Warschauer’s conclusion is that meaningful access requires four categories of resources working together — physical, digital, human, and social. Not necessarily in sequence, but in combination. Each reinforces or undermines the others. Invest heavily in one while ignoring the rest and you don’t just fail to help, you concentrate the gains among those who already had everything else in place: urban, educated, English-speaking. Everyone else gets a Hole-in-the-Wall.

The Sprint and the Marathon

India does need an AI Mission. Building some domestic capacity still matters, especially when training models tailored to India’s immense diversity does not always make great business sense. But the mission does not take the minister’s own argument seriously. The revised expenditure on the IndiaAI Mission for 2025-26 came in at Rs 800 crore, less than half the budgeted Rs 2,000 crore. Of what was spent, over 85% went to compute subsidies for model building: Sarvam, BharatGen, Soket AI, and others. Skilling, datasets, application development, the work that would actually mend the broken chains, received almost nothing. The mission alone cannot close the chains. There are long-standing non-AI problems — connectivity, literacy, electricity reliability, gender-equitable device access. They are governance problems that require funding and implementation across every tier of government.

Amodei ends his essay describing the world with AI’s loving grace as “a thing of transcendent beauty” and adds, “We have the opportunity to play some small role in making it real.” But the loving grace of AI will not arrive by itself. It will arrive, if it arrives, through the unglamorous work of fixing the broken chains, one node at a time. From Bengaluru, all the way to Byrnihat.

*The usual disclaimer: views here are entirely my own.

In 1997, as the Asian Financial Crisis rocked economies across the region, South Korea embarked on one of the most aggressive deregulation drives in history. Every regulation was to pass through a committee and justify its existence. Those with low compliance rates, poor cost-benefit ratios, and those misaligned with global standards were eliminated. Within a year, Korea had cut nearly half its regulatory bloat. In her Budget Speech last year, finance minister Nirmala Sitharaman highlighted efforts to reduce the “excessive regulatory burden” on the Indian economy. A Korean-style regulatory guillotine may be too bloody for India’s taste. But the scissors are finally out.

Finally, Some Self-Awareness

Last May, RBI brought in a framework that, for once, regulates its own conduct. And it’s not the only one going meta. Earlier in February, SEBI too issued a regulation to regulate… how it regulates. While this may seem like a case of regulators tying themselves up in their own red tape, these regulatory urges are not coming from nowhere. Last year’s Economic Survey makes a case for a “systematic” procedure for regulation making, one that is “fair, accessible and open”. And this shift has been a long time coming. The report of the Financial Sector Legislative Reforms Commission (FSLRC), a treasure trove of ideas on reforming India’s financial sector, had recommended these steps over a decade ago. RBI and SEBI are actually a little late to the party. IBBI had already put similar regulations in place as far back as 2018.

While India’s regulators are still taking baby steps, the US had already institutionalized this regulatory mindfulness in its Administrative Procedure Act in 1946. The EU’s ‘one-in, one-out’ rule forces it to cut an equal amount of red tape before making businesses jump through new hoops. The EU, as well as the UK and Australia, have independent bodies at the federal level that screen new regulations.

Heard, Not Heeded

The new rules codify a lot of what the regulators have already been doing. They require regulators to publish draft regulations for public consultation. That means no more regulatory ambushes. Regulators must cite exact legal provisions that empower them to make a regulation, ensuring they stick to their mandate. And they will have to clearly spell out the objectives behind regulations, so that the public can judge if it is genuinely designed to fill a gap, and if it’s summoning a hammer where a scalpel would have done. They also need to allow enough time, no less than 21 days, for comments on the proposed regulation. This signals a genuine push towards making consultation more participatory and not merely performative.

But is this enough? A regulator can nod politely at suggestions and then ignore them. RBI has promised a “general statement of its response”. But this gives it enough room to avoid publishing certain comments. SEBI, however, requires itself to give detailed rationale for rejected comments. IBBI publishes all comments it receives, and even when not required to, has provided rationale for rejected comments. Of course, there are times when an urgent regulation is required. This is allowed, but no regulator has yet specified that such measures should lapse after a fixed period unless replaced through the full procedure.

Don't Ask the Chef if the Food is Good

Now what happens if a regulator simply says, “We disagree,” and moves on? There’s really nothing stopping them. Without external review, the whole process risks turning into a mere ‘check-the-box’ exercise. RBI’s choice to bring in these requirements in the form of a softer “framework”, instead of hard “regulations” as other regulators have done, is also telling. This may give RBI flexibility to sidestep them, especially when the framework itself provides plenty of exceptions.

New regulations make headlines. Their reviews rarely do. There’s a perverse asymmetry at play here. Policymakers love to announce new regulations, but few check if they have actually worked. Regulations are policy experiments and even after going through a rigorous consultative process and analysis, some fail. Most regulators require a regular review of regulations now but only IBBI has mandated a fixed three year cycle.

RBI now requires itself to conduct an impact analysis for any proposed regulation. More elaborately, IBBI requires an economic analysis covering the cost and benefit to society, the economy, and other stakeholders. An obvious problem here is the confirmation bias intrinsic in marking your own homework. Even with the best intentions, it can start to resemble Michael Scott’s “World’s Best Boss” mug. So, such impact assessments must be validated by independent third-parties.

Good Faith Has an Expiry Date

Here again, IBBI has shown the way. In 2021, it commissioned an independent think tank to evaluate its performance. And evaluate it did. It noted that the regulator had sometimes missed its mandatory three-year review timeline. It also flagged the poor quality of its economic impact analyses. IBBI isn’t perfect, but by opening itself up to third-party scrutiny and publishing the findings, it demonstrated a rare willingness to be held accountable.

What IBBI has done voluntarily, every financial sector regulator should be required to do by law. Even IBBI, whose earlier leadership pioneered these efforts, has not repeated the exercise since. The model is straightforward: a mandatory, periodic third-party evaluation of each regulator’s rule-making conduct, covering consultation quality, impact assessment rigour, and review compliance, with findings made publicly available.

In India, CAG traditionally audits regulators on financial matters and Parliament’s Public Accounts Committee may conduct a wider exercise occasionally. But neither is a substitute for a regular, targeted structure that evaluates regulators on their core function — how they regulate. That gap needs filling. A single independent body, mandated by legislation, auditing all financial sector regulators on this one question alone is both achievable and overdue. After all, who better than a regulator to understand the value of oversight?

*The usual disclaimer: views here are entirely my own.

On November 5, 2020, NPCI issued a circular mandating that no single UPI app could process more than 30% of total transaction volume. Players already above that threshold were given until December 31, 2022 to comply.

Four months later, in March 2021, an SOP was published providing more detail. Once a payment app crossed 30%, it would be required to stop onboarding new customers. Existing users could keep transacting normally. The SOP repeated this across multiple sections, with some insistence. Just a pause on new registrations, and patience.

To get there, the SOP built a three-level alert architecture. At 25-27%, NPCI would send an email. The Third-Party App Provider (TPAP) and its PSP bank would acknowledge it. At 27.1-30%, a second email, this time requiring the TPAP to furnish “evidence of actions taken” would be sent. The only action available, I assume, would be to stop acquiring new users. A new customer trying to sign up for PhonePe or Google Pay would have to be told: We are too popular right now, please try some other app.

Cross 30%, and the app must stop new onboarding and submit an undertaking. At which point, NPCI may offer an exemption of up to six months. If the share is still not falling after the exemption, NPCI would “discuss the modalities.” If it continued beyond that, penalties under UPI Procedural Guidelines could be levied, or onboarding could be blocked from the central system entirely.

When the original deadline of December 31, 2022 finally arrived, it was extended by two years. When those two years elapsed, it was extended again. The deadline now stands at December 2026, six years after the first circular, and counting. In December 2026, one of two things will happen. NPCI will extend the deadline yet again, a familiar ritual of intent, or it will finally admit that enforcement was never genuinely on the table.

But Market Dominance in Itself Isn’t a Problem, Right?

In China, WeChat Pay and Alipay spent years as walled gardens: a WeChat Pay QR code could not be scanned by an Alipay user, and vice versa. The regulatory pressure to open these networks up took the better part of a decade and is still incomplete.

India built interoperability in from the start. A PhonePe user can pay a Google Pay user without either party noticing the difference. The network effect that makes a single messaging app or social network almost impossible to displace does not apply here in the same way. Better still, every user and merchant added to UPI, whether on Amazon Pay, Cred or Google Pay, strengthens the network for everyone. The rising tide lifts all boats.

And yet, the apps sitting on top of those rails have consolidated as sharply as any closed platform. In January 2026, PhonePe processed 9,914 million transactions — 46.6% of total UPI volume. Google Pay added another 34%. Together, 80.6% of all UPI transactions flowed through two apps. The remaining 83 apps on the platform collectively processed less than one-fifth of what PhonePe alone did.

UPI currently runs on a zero-MDR mandate: MDR, or merchant discount rate, is the small fee that payment networks typically charge merchants per transaction. For UPI, that fee is zero. Even if that ever changes, this is a space where MDR would remain tightly regulated. No dominant player is going to be allowed to ratchet up the price of a UPI transaction to extract monopoly rents. That is a genuine consumer protection, and it mostly works as intended.

At the same time, zero MDR has turned payments into a distribution channel rather than a product. For Google Pay, transaction data adds to an already vast consumer profile, sharpening a picture of what a customer wants, spends on, and worries about. For PhonePe, the payments app provides virtual real estate that hundreds of millions of people access dozens of times a day, and that real estate is then used to sell mutual funds, insurance, loans, flight tickets, and train bookings.

But where you have cross-selling at scale, you potentially have mis-selling at scale. PhonePe has no obligation to highlight the best mutual fund on the market. It will highlight the funds from its partner AMCs, or the ones paying it the highest trail commission. The UX can do the rest: dark patterns, nudges toward higher-margin products, pre-ticked consent boxes, friction applied asymmetrically to exits rather than entries. None of this is unique to the incumbents. But all of it becomes dangerous at 46% of a market.

In March 2020, a moratorium by the RBI on Yes Bank caused PhonePe to shut down completely for twenty-four hours. Yes Bank was PhonePe’s exclusive banking partner at the time. PhonePe has since diversified, but the episode pointed at something larger. A cyber incident, a regulatory action, a critical third-party failure — any of these could do the same today. What gets lost in that scenario is not just access to a payment rail. It is years of a user’s financial life: transaction history, saved billers, mandates, credit integrations, held inside an app they did not choose to make irreplaceable.

A payment app with 150 million daily active users also puts regulators in a bind. Any enforcement action against it risks disrupting payments for millions of ordinary users. So it gets delayed. Once and then again. That is the compounding logic of concentration. As a platform’s footprint grows, the cost of disciplining it rises. Eventually, it becomes too important to regulate on normal terms.

Switching Costs

There is a well-documented asymmetry in how people evaluate switching costs. Giving up a product you already use feels like a loss, and losses psychologically outweigh equivalent gains. The result is a persistent gap between how good a challenger’s product actually needs to be and how good it needs to feel to a user who is being asked to abandon something familiar. That gap is the moat for incumbents.

From roughly 915 million transactions in FY2018 to over 130 billion in FY2024, UPI’s growth curve was nearly vertical.

The vast majority of UPI’s current user base was acquired during the hypergrowth years, when PhonePe and Google Pay were the dominant, most-visible, most-aggressively-promoted options on the market. The habit loop was formed at the precise moment the market was being created, when there was no entrenched alternative to displace.

Explore the data in full

A user who started using UPI in 2017 or 2018 does not consciously think about their payment app today. They open it, scan the QR, get rewarded with that sweet sound of a successful transaction, and move on. The relationship with the app is, well, transactional.

Yet years of reliable experience have built something the user doesn’t notice until they’re asked to give it up. What happens if a transaction gets stuck on some app I’ve never used before? That fear, rational or not, is compounded by everything else that has quietly accumulated on the incumbent — transaction history going back years, saved billing numbers for electricity and gas connections, recurring autopay mandates for insurance premiums and loan EMIs, UPI Lite balances, credit line integrations. Switching is not just a habit change. It is a migration, and migrations have costs that most users have very few reasons to pay.

This raises a question about what a regulator should legitimately do at all. Behavioural lock-in earned through a genuinely good product is, in principle, legitimate competitive advantage. A regulator who tries to neutralise earned stickiness is punishing success and distorting the signal that markets are supposed to send. You would not mandate that a user switch from Google Search simply because familiarity makes them less likely to try Bing.

In most markets, an incumbent eventually charges for its dominance, and those charges create the opening a challenger needs: come in cheaper and get a foothold. Zero MDR closes that opening. The incumbent cannot monetise its position, but neither can a challenger undercut it on price. The only remaining axes of competition are product quality and distribution, and on both dimensions the incumbents, with their years of data, embedded habits, and Walmart-sized customer acquisition budgets, hold structural advantages that a new entrant cannot easily overcome.

The regulator’s legitimate target should be the structural barriers that prevent any challenger from reaching the scale at which competing habits could form at all. The question is not how to force users off PhonePe or Google Pay. It is how to lower the cost of forming a new habit somewhere else: through transaction history portability that lets users carry their financial data to a new app the way switching phones no longer means losing your contacts, photos, or even default settings.

The 30% cap addresses neither challenger incentives nor the data portability gap that makes switching difficult. It targets the symptom rather than the structure.

Data Portability

In October 2025, NPCI mandated that users be able to view and port autopay mandates across any UPI app of their choice. This matters for a reason beyond its immediate effect. It establishes a principle: that data generated by a user’s financial behaviour belongs to the user, not to the app that happened to record it. A right to data portability was included in the Personal Data Protection Bill of 2019 but did not survive to the final legislation. RBI and NPCI can rekindle it for the financial sector. That principle, if taken seriously and extended consistently, is the thread that unravels the switching cost problem from the inside.

The next logical step is transaction history. There is no reason a user switching apps should lose access to years of financial records. Standardised, exportable transaction history, in a format that any UPI app can ingest, would immediately reduce one of the most underappreciated switching costs. It would also solve a quieter problem: the fragmented transaction history of anyone who has ever used more than one app. The same logic extends to saved billers, splitting groups and credit line integrations.

The cleanest near-term path runs through NPCI itself, which has system-wide visibility on all UPI transactions and could develop a data portability utility that any app can plug into on customer consent. A more elegant architecture already exists in the Account Aggregator (AA) framework which was designed precisely around the idea that financial data belongs to the customer and should travel on consent. Extending AA’s scope to UPI-generated data would require onboarding TPAPs as Financial Information Providers and more difficult work of data standardisation. The dominant players will be least willing to cooperate. Making them the first to comply, through asymmetric obligations tied to market share, should be NPCI’s job.

True data portability will ensure that the cost of leaving is determined by the user’s preferences rather than by accumulated friction. That is what regulation is supposed to do.

Giving Challengers A Fight To Fight

Zero MDR is not one policy. It is two policies that have been bundled together and treated as inseparable.

The first policy is a public good commitment. Small-value, everyday payments should cost nothing. A vegetable vendor in rural Maharashtra should not pay to accept a ₹40 transaction. A migrant in Bengaluru sending remittances to his home in Nagaland should not have to pay an extra fee. That commitment is worth defending.

The second policy is a market structure choice. Nobody competes on price. Ever. For any transaction. That choice, reasonable as a temporary subsidy to drive adoption, has calcified into a permanent feature.

The costs of running this system have never been zero. Between FY22 and FY25, the government paid ₹8,276 crore in subsidies to banks and ecosystem participants to keep UPI transactions free. The government paid out ₹2,196 crore in FY26 and has budgeted ₹2,000 crore for FY27, against an industry ask of ₹10,000 crore. With 300 million transactions processed every day at zero MDR and ₹2,000 crore to cover it, the ecosystem is being slowly starved of the capital it needs to reach the next 300 million users.

The costs that are not recovered do not disappear. They get deferred. And deferred infrastructure investment has a way of making itself visible. NPCI operates the central infrastructure but derives no direct revenue from UPI transactions. Its capacity investment is funded by membership fees, government subsidies, and cross-subsidies from other products. As transaction volumes grow and subsidies are cut, the pressure on that funding model grows. On April 12, 2025, UPI’s payment success rate fell to between 50% and 80% for roughly five hours, the fourth disruption in three weeks. The root cause was a flooding of NPCI’s systems with automated status-check requests, a load the infrastructure was not built to absorb.

NPCI has already quietly acknowledged that zero MDR cannot hold everywhere. By mid-2024, NPCI was in advanced discussions to introduce an interchange of approximately 1.2% for credit line transactions on UPI. The circular was expected within weeks. It never came. But the acknowledgment had been made.

The Payments Council of India has already shown where we should go from here on. Zero MDR preserved for all merchants with annual turnover below ₹20 lakh, roughly 90% of the merchant base by count. A modest 30 basis points on transactions above that threshold, covering approximately 5 million large organised merchants who are already paying 2% for credit card acceptance and are unlikely to abandon UPI over a 30 bps charge. With that, preserve zero MDR for all transactions below a value threshold, say ₹2,000, across all merchant sizes. Above that threshold, introduce regulated interchange tiered by merchant size. A graded MDR structure, designed carefully, would give challengers a price axis to compete on. It would give NPCI a revenue argument for building portability infrastructure.

The 30% cap asked dominant players to shrink. True data portability removes artificial stickiness and empowers the customer. A graded MDR asks the ecosystem to grow in a way that makes room for someone else.

The first approach has had six years. Time for the alternatives?

*The usual disclaimer: views here are entirely my own.