Money & Machines: (UPI) Zero for Whom?; The Sovereign AI Dilemma

In this edition, we revisit the longest-running payments debate i.e. UPI MDR; and we explore what happens when sovereign AI, venture capital and national interest collide.

By Joseph Rai, Arti Singh • Jul 18, 2026 • 12:28 PM

11 min read

Money & Machines: (UPI) Zero for Whom?; The Sovereign AI Dilemma

11 min read

In this edition of Money and Machines, we revisit the longest-running payments debate i.e. UPI MDR; and On the AI front, we explore what happens when sovereign AI, venture capital and national interest collide. As the government asks Sarvam AI and BharatGen to develop Mythos-like capabilities, we ask a larger question: where should governments draw the line between enabling innovation and steering it?

(UPI) Zero for Whom?

This week, The Economic Times reported that the government is examining a proposal to restore MDR on UPI payments for large businesses, with an annual turnover above Rs 1-1.5 crore. The fee may be set below 0.5% and apply to transactions above Rs 2,000.

I have been writing about UPI's zero-MDR policy ever since the government abolished merchant discount rates in 2020. Every few months, the same cycle repeats with industry bodies submitting representations and then someone saying the government is seriously considering restoring MDR.

Then…nothing.

Six years later, even the industry's optimism has turned into sarcasm.

When I asked one of the top payments executives whether he finally sees positive signs. "Yes," he said. "For the last six years."

"I can't comment until I actually see a circular," he further noted.

The bigger problem is that the entire debate starts with an assumption that has never really been true – that the UPI is free.

It isn’t.

And, time and again I have written about this great ‘zero-MDR’ illusion that on paper, accepting UPI became free; however, in practice, an entirely different market evolved. Payment aggregators started charging "platform fees", "technical fees", "experience charges", or a flat bundled payment gateway pricing.

Large merchants understood regulations, had the scale and bargaining power to negotiate commercial terms. Whereas, small merchants didn't – which actually means this segment ended up paying for something that was supposed to be free ‘by law’.

Now this is something that you will hardly see open discussions around – but the fact is the payments industry has always been divided into two very different merchant economies.

One set is millions of kirana stores and small online businesses, while the other are the likes of Reliance Retail, Tata, Amazon, Flipkart, major fuel retailers, telecom companies, utility providers and large organised retail chains.

Both accept UPI. But small merchants typically accept whatever pricing is offered, and large ones negotiate everything from acquiring fees to settlement terms.

Industry estimates suggest that while nearly 80% of merchants are small businesses, they contribute only about 20% of UPI payment value. The remaining 20% – large merchants – account for roughly 80% of transaction value.

Let's also look at it through international experience. This isn't the first time policymakers have tried to make digital payments cheaper for merchants. More than a decade ago, the United States capped debit card interchange fees through the Durbin Amendment after large merchant lobbies argued that payment acceptance costs were too high.

The argument was lower payment costs for merchants would eventually translate into lower prices for consumers. But that is not how it played out.

One of the most cited post-Durbin studies by the Federal Reserve Bank of Richmond found that the benefits were distributed unevenly across merchants. Businesses selling high-value goods reported some of the biggest reductions in payment acceptance costs; whereas sectors dominated by small-ticket transactions reported little benefit, and in some cases even higher costs after the pricing structure changed.

Once again, negotiation power determined who benefitted the most.

The Richmond Fed survey also found that nearly two-thirds of merchants reported little or no reduction in their debit acceptance costs. And despite one of the central promises of the regulation, very few merchants actually reduced prices for consumers.

In other words, regulating payment fees did not eliminate costs but rather resulted in an uneven playing field.

Now, coming back to India.

One point worth noting is that the MDR debate has evolved since 2020. Back then, Reliance Retail, which was India's largest retailer Reliance – and still is – benefitted from zero MDR as a merchant.

But today, it is also building a payments story.

Through Jio Payment Solutions, the group has spent the past year aggressively building its merchant acquiring business, leveraging pricing and banking relationships, as exclusively reported by The Head and Tale last year.

That makes Reliance an interesting case study. One may argue that payments is just the entry point into a much larger merchant financial services play for Jio.

But what makes the MDR debate more interesting today – more from the country's largest merchant point of view – is that, unlike six years ago, Reliance is no longer participating from just one side. If MDR is eventually restored for large merchants, the Mukesh Ambani-owned entity will incur UPI payment costs; and at the same time, Jio Payment Solutions, as the group's payment aggregator arm, can earn from MDR. And if Jio PA powers Reliance Retail's payments, the economics stay within the Reliance ecosystem. At a group level, it almost evens out.

Similarly, Amazon – through its position as India's largest online marketplace and its payment aggregator business – also finds itself on both sides of the MDR equation.

No matter what happens, one thing is clear that running a payments network as massive as UPI costs money – a constant investment in tech, infra by banks and payments players.

So, the bigger question is whether an ecosystem of this scale can continue relying on shrinking government incentives and commercial workarounds indefinitely? Growth has already started to moderate as the market matures, while the curbs on high-volume segments such as real-money gaming have also impacted transaction volumes for parts of the ecosystem.

MDR may return this time, or may not.

But even if it doesn't, the zero-MDR era is already over. The cost of running UPI did not go anywhere when the government made it free, but moved to whoever had the least power to say no – small merchants paying fees they could not negotiate, customers paying "convenience fees" on transactions that were supposed to be free, taxpayers funding incentives (that continue to shrink), and to some payment companies absorbing a part of the burden.

The real story here is not whether a circular gets issued, but whether the government is finally ready to acknowledge the reality that already exists. Because when we say UPI has zero MDR, the question worth asking is – Zero for whom? And, the next question perhaps is whether it is finally willing to create a level playing field for everyone participating in the UPI ecosystem.

-- by Arti Singh

_____________________

Will the government's AI ambitions test the VC playbook?

Even before Narendra Modi became the Prime Minister of the country, he was widely cheered by corporate India for his business-friendly policies in his home state of Gujarat as chief minister.

After becoming Prime Minister as well, he has been lauded for his policies by corporate honchos and has frequently maintained that the "government has no business to do business."

But in the age of artificial intelligence, that could potentially change.

According to The Economic Times, citing top officials, the government has asked homegrown AI startups Sarvam AI and BharatGen to tweak their foundational models and develop Mythos-like capabilities.

On one hand, the government's proactiveness in asking startups to tweak their models seems understandable at a time when calls for sovereign AI grew after the US government heavily restricted and temporarily shut down Anthropic's Claude Fable 5 and Mythos 5 models last month before easing the restrictions. Extreme times call for extreme measures is probably what the Indian government is thinking, and it appears that it wants to actively participate in developing models that protect our national security.

However, what is quite surprising is that one of the AI startups that the government is reportedly asking to tweak its foundational models is Sarvam AI. The startup, founded by Vivek Raghavan and Pratyush Kumar, has raised millions of dollars from venture capital firms, including Lightspeed, Peak XV Partners and Khosla Ventures, since it was founded in 2023. And just last month, the startup stormed into the famed unicorn club after raising a whopping $234 million led by Indian IT major HCLTech in an ongoing funding round. (Reports have it that global chipmaker Nvidia is also expected to participate in the funding round.)

In the light of the government's intervention, what would this mean for the venture capital firms that are constantly thinking about exiting with handsome returns from their investments? Will it be easy for them to accept changes in strategy such as the ones that the government is reportedly making on Sarvam AI?

To be sure, the Indian government has supported Sarvam AI under its flagship IndiaAI Mission as one of the startups selected to develop an indigenous, sovereign Large Language Model (LLM). As part of the mission, the government allocated a massive infrastructure subsidy to Sarvam AI, giving it six months of access to 4,096 Nvidia H100 GPUs. The Economic Times also reported that the government is set to pick up a 1-2% equity stake in Sarvam AI.

Again, government backing could serve as a booster for venture capital firms as the possibility of the startup running aground becomes lower. But venture capital firms would not be ecstatic to learn that the government is asking one of their portfolio companies to change its strategy unless they are willing to play the role of passive investors who are interested only in the financial outcome, no matter how it gets generated.

But venture capital firms have hardly operated that way. a16z, the US-based marquee venture capital firm founded in 2009 by Marc Andreessen and Ben Horowitz, in its blog post underscores the importance of governments regulating the harmful effects of AI rather than AI model development itself. "For decades, the United States' approach to regulating technology has been based on how that technology is used-not how it's made. For instance, there's no law dictating how to build a computer," it wrote, highlighting that venture capital firms would rather prefer governments not advise companies on how to build technology.

It will be interesting to see the extent of the Indian government's intervention, or advice, in shaping a startup's strategy based on national security and national interest. Notably, in March, Sarvam AI had already launched Chanakya, a specialised enterprise AI framework that works entirely offline inside the government's own buildings. Because it is completely disconnected from the internet, secret government data and military maps stay locked away, making it impossible for hackers to steal them from the outside.

There will definitely be people in favour of the government's intervention, stressing the need for sovereign AI. While that is a valid argument, what also needs to be understood is that frontier AI models and strategic capabilities cannot be replicated overnight. Just because a government asks a startup to build a Mythos-like model does not mean it can be done in a jiffy. Even China, despite its deep technological capabilities, has not been able to create a singular alternative to the US AI frontier models of OpenAI and Anthropic in either the global or Indian markets. Building sovereign AI is a long and arduous process, one that will require patient capital, technological breakthroughs and time. And perhaps the government should know where and when to draw the line between enabling innovation and steering it.

-- by Joseph Rai

_____________________

This Week in Review

Report

Inside the fintech bets of Blume Ventures

The Monthly Crypto Map

The Monthly Crypto Map: June 2026

News

Neo Group raises Rs 350 crore in funding round led by Peak XV Partners

Kissht gets AMFI registration for mutual fund distribution biz

Finova Capital FY26 revenue up 21% to Rs 921 crore, net profit falls 11%

Groww Q1 revenue up 66% to Rs 1,501 crore, net profit surges 94%