Inference is like electricity: Paytm and Groq on the new utility shaping fintech

India’s fintech story has always been a story of scale. But at TechSparks 2025, the conversation shifted to something more elemental: speed. Not the speed of payments, which India has long mastered, but the speed of intelligence.

In a fireside chat moderated by YourStory COO Sangeeta Bavi, Paytm’s CBO Narendra Yadav and Groq’s APAC GM Scott Albin outlined how real-time AI is quietly becoming the next competitive layer in financial services.

AI steps out of the back office

For Paytm, AI’s role inside the company has evolved quickly. “Last year AI started at the back end… for cost optimization, operations, product development,” Yadav said. “Now it is moving to the front layer.”

The clearest example is Paytm’s new AI Soundbox, an upgrade to the near-ubiquitous payment alert device used by millions of merchants. Beyond just announcing transactions, the latest version interprets them, analyses patterns, and even enables languages to converge inside a shop.

“A tourist may be speaking in Spanish and the merchant in Hindi, but both are able to interact,” Yadav said, noting how early pilot users have stretched the device far beyond its intended purpose.

The company expects it to become the default form factor. “AI Soundbox will become the sound box,” Yadav said. “The feature prefix will go to the earlier generation.”

Billions of transactions, each touched by AI

Paytm’s scale provides a sense of why inference speed matters. “We settle over $20 billion of merchant payments every month,” Yadav said. “Practically, AI is touching each of these transactions multiple times.”

Fraud detection, risk scoring, search, and customer-side interfaces all depend on inference running at near-instantaneous speed. According to Yadav, Paytm’s internal forecasts on AI usage were quickly rendered obsolete; one reason the company has now formalized a partnership with Groq.

Groq’s worldview: inference as a utility

Albin distilled inference in plain terms: it’s “the work to take what you type into the model and turn it into an output.” Every query is a compute request, and around the world, the queues are getting longer.

“If you don’t get a response instantly, that’s the system queuing. It doesn’t have enough capacity,” he said. The result is a global scramble to build data centers, cooling infrastructure, and affordable compute.

“We think of inference like electricity,” Albin said. “It’s becoming a critical utility…That's going to be the future that we're building for.”

But Asia, he pointed out, is still compute-short across markets. Many cutting-edge models don’t run locally because the infrastructure isn’t available. That gap, he argued, is an opening for newer architectures built specifically for speed and cost efficiency; Groq’s pitch in one line: fast inference at a fraction of traditional costs.

Why Paytm chose a relatively young player

Bavi asked the obvious: Why did Paytm choose Groq over more established multinational vendors?

“The technology is great… it suited all our requirements. It’s fast inference at the most economical cost,” Yadav said. He added that Paytm has tracked Groq from the early days, including Founder and CEO Jonathan Ross’ work designing and building the Groq LPU after his time building Google’s TPU. “They became the natural choice for us.”

The partnership is focused on inference. Paytm will run its models and future AI workloads on Groq’s infrastructure, especially as AI becomes more embedded in customer-facing surfaces.

Albin also highlights India as a strategic market for Groq, stating that they have, “At Groq, we have 2.5 million developers globally and half of those are based in Asia, with nearly 500,000 developers in India alone.” where he also notes that sees the country with a massive strategic advantage for its compute infrastructure to meet the surging demand for AI.

What’s next inside Paytm’s AI pipeline

Yadav also offered a glimpse into what’s currently brewing inside Paytm’s AI pipeline. The AI Soundbox remains the most visible initiative, with the company preparing to take it to millions of merchants as a full-fledged intelligence device rather than just a payment notifier.

On the consumer side, Paytm Travel is being rebuilt to shift from an experience that is “efficient” to one that feels intuitive, anticipatory, and tailored to each user’s history. Real-time fraud detection is another area undergoing a significant overhaul, with AI expected to sharply reduce false positives even as transaction volumes continue to climb.

Paytm is also leaning on alternate data to make credit assessments more inclusive and bring previously ineligible users into the fold. Taken together, Yadav said, “practically all aspects are getting updated” from risk to product development to the merchant-facing experience.

Governance, sovereignty, and the widening gap

On regulation, Albin was frank: “I don’t think anyone’s doing this well.” Countries are experimenting with frameworks, often at the risk of slowing local innovation. But one trend that is clear is that privacy norms will harden, and data will increasingly be expected to remain within borders.

He noted that this shift toward “the security of everything” – food, energy, compute – will create inefficiencies but will also pave the way for national AI champions.

For financial services especially, there’s a widening gap in AI innovation. Success will hinge on which firms move fastest to engage customers while ensuring robust guardrails and internal governance for real-world deployment.