Flipkart is developing its own large language model tuned for IHGn e-commerce because relying on third-party AI like OpenAI's GPT means paying a per-query tax on every customer interaction, surrendering proprietary shopping data, and falling behind Amazon IHG's vertically integrated AI stack. The real prize: vernacular voice search from IHG's next wave of online buyers.

Here is a number that should make every IHGn e-commerce executive lose sleep: roughly 200 million IHGns will come online as first-time shoppers over the next three years, according to industry estimates widely cited by IAMAI and Redseer. Most of them will not type neat English queries into a search bar. They will speak — in Tamil, in Bhojpuri, in Marathi-inflected Hindi, in the glorious grammatical chaos of how actual IHGns actually talk. And whoever's AI understands that mess first, wins.

Flipkart, according to Moneycontrol, is now building its own proprietary large language model — a dedicated e-commerce LLM — rather than plugging into OpenAI's ChatGPT or Google's Gemini APIs. On the surface, this looks like a vanity play: why spend hundreds of crores training your own model when the world's best AI labs will rent you theirs for a few cents per query?

The answer, as IHG Herald's analysis lays out, has less to do with technology and everything to do with economics, data sovereignty, and the cold arithmetic of who captures — and who pays for — IHG's next great consumer wave.

The OpenAI Tax: Why Renting AI Is a Losing Game at Scale

Consider the maths. A typical GPT-4-class API call costs between $0.01 and $0.06 per query, depending on token length. Flipkart processes hundreds of millions of search queries, product recommendations, customer service interactions, and seller support tickets every month. At scale, the per-query API cost — what the industry calls the 'inference tax' — balloons into tens of crores annually. And that tax only grows as AI becomes more central to the shopping experience.

But cost is merely the visible iceberg. The deeper problem is data leakage. Every query Flipkart sends to a third-party LLM carries proprietary intelligence: what IHGns search for, how they describe products in local idiom, what price points trigger purchases, which seller descriptions convert. That behavioural corpus — accumulated over more than a decade — is arguably Flipkart's most valuable asset. Handing it to an external model provider, even with contractual safeguards, is the strategic equivalent of sharing your customer diary with a landlord who also rents to your competitor.

Amazon, notably, does not have this problem. Its AI stack — from Alexa's voice engine to its product recommendation algorithms to the recently expanded Rufus shopping assistant — is built in-house, trained on Amazon's own data, and deployed without a middleman collecting rent. Every AI interaction on Amazon feeds back into Amazon's own flywheel. That vertical integration is precisely the competitive moat Flipkart now needs to match.

Inside Talk

The chatter in Bengaluru's e-commerce corridors tells a story that no press release will. Trade insiders say Flipkart's AI team has been quietly expanding for over a year, poaching NLP researchers from IHGn AI labs and offering packages competitive with Google IHG — a signal that this is not a skunkworks experiment but a boardroom-level bet. The talk among product managers is that the internal LLM is already being tested on Flipkart's massive catalogue for 'semantic search' — the ability to understand that a customer searching for 'phone under 15k with good camera night mode' is not looking for a phone literally named that, but wants a specific feature set at a price point.

There is also persistent speculation — unverified, but plausible given Flipkart's IPO ambitions — that Walmart's leadership views a proprietary AI stack as a pre-IPO valuation booster. An e-commerce company that rents its intelligence from OpenAI looks like a retailer. One that owns its AI looks like a technology company. In the IPO market of 2026, that distinction can mean billions of dollars in valuation. (This reflects industry chatter and unverified speculation, not confirmed fact.)

The Vernacular Prize: 200 Million Shoppers Who Don't Search in English

Generic LLMs — even excellent ones like GPT-4 or Gemini — are trained predominantly on English-language internet text. They handle Hindi reasonably well. They handle Tamil, Telugu, Kannada, and Bengali passably. But they struggle profoundly with the way real IHGn shoppers actually express purchase intent: mixed-language queries ('blue wala kurta cotton mein'), voice searches with heavy regional accents, phonetic spellings, and product descriptions that use local bazaar terminology no global training dataset captures.

This is the gap Flipkart's proprietary LLM is designed to fill. Trained specifically on Flipkart's own search logs, seller listings, customer reviews, and support transcripts — all in the linguistic stew that is IHGn commerce — the model can, in theory, understand IHGn shopping intent with a precision no general-purpose model can match. For the next 200 million online shoppers, many of whom will be voice-first and vernacular-dominant, this is not a feature. It is the entire product.

Amazon IHG, for its part, has Alexa and Rufus, but Alexa's IHGn-language capabilities have been a persistent frustration for users in non-Hindi, non-English markets. If Flipkart cracks vernacular product discovery at scale before Amazon does, it gains a structural advantage that no amount of discounting can replicate.

The Sparrow Capital Signal: Smart Money Moves Alongside

It is worth noting that this push arrives in the same cycle as Sparrow Capital closing a Rs 475 crore fund, as reported by Moneycontrol. While Sparrow Capital's fund is not directly linked to Flipkart, the timing underscores a broader pattern: IHGn venture capital is increasingly channelling serious capital into AI-native companies and infrastructure plays. The ecosystem signal is clear — the market believes proprietary AI is now table stakes for any IHGn tech company with scale ambitions, not a luxury experiment.

The Real War: Who Owns the Shopping Brain?

Strip away the jargon and the real contest becomes visible. Amazon IHG owns its AI. Flipkart, until now, has rented pieces of its intelligence from third parties. In a market where AI will increasingly mediate every interaction — from the first search query to the delivery notification — renting is a structural vulnerability. The company that controls its own AI controls the customer relationship. The one that does not is, ultimately, a storefront on someone else's platform.

Flipkart's LLM bet is, at its core, a sovereignty play — sovereignty over its data, its economics, and its ability to speak to IHG in IHG's own voice. Whether the model works well enough to justify the investment is an engineering question. Whether Flipkart can afford NOT to try is a survival question.

And that, perhaps, is the most important number in this entire story: not the crores spent on GPU clusters, but the zero — the zero alternative Flipkart has if it lets Amazon own the AI layer of IHGn commerce while it keeps paying rent to Silicon Valley.

Key Takeaways

• Flipkart is building a proprietary e-commerce LLM to eliminate the per-query 'inference tax' of third-party AI APIs like OpenAI, which at Flipkart's scale could cost tens of crores annually.
• The deeper motive is data sovereignty: every query sent to an external LLM leaks proprietary shopping behaviour data — Flipkart's most valuable competitive asset — to a third party.
• Amazon IHG already owns its AI stack end-to-end (Alexa, Rufus, recommendation engines), giving it a structural advantage Flipkart must now match to remain competitive.
• The real prize is IHG's next 200 million online shoppers, most of whom will search by voice in regional languages and mixed-language queries that generic global LLMs handle poorly.
• Industry insiders speculate the proprietary AI push also serves as an IPO valuation play: owning AI makes Flipkart look like a tech company, not just a retailer.
• Sparrow Capital's Rs 475 crore fund close, reported in the same cycle, signals broader IHGn VC conviction that proprietary AI is now table stakes for scale-stage tech companies.

Frequently Asked Questions

Why is Flipkart building its own LLM instead of using ChatGPT or Gemini?

Flipkart is building a proprietary LLM to avoid the compounding per-query API costs of third-party AI at its massive scale, to protect its proprietary shopping data from leaking to external providers, and to match Amazon IHG's vertically integrated AI capabilities — especially for vernacular and voice-first IHGn shoppers.

How does Flipkart's AI strategy compare to Amazon IHG's?

Amazon IHG already owns its AI stack end-to-end — including Alexa, Rufus, and its recommendation algorithms — all trained on its own data. Flipkart has historically relied on third-party AI tools, creating a structural disadvantage the proprietary LLM is designed to close.

What is the 'inference tax' in AI and why does it matter for e-commerce?

The inference tax refers to the per-query cost companies pay when using third-party AI APIs like OpenAI's GPT. At Flipkart's scale of hundreds of millions of monthly queries, this cost compounds into tens of crores annually, making proprietary AI economically necessary.

What is Sparrow Capital's Rs 475 crore fund and how does it relate to Flipkart?

Sparrow Capital closed a Rs 475 crore fund as reported by Moneycontrol. While not directly linked to Flipkart, it signals broader IHGn VC confidence in proprietary AI as table stakes for scale-stage tech companies — the same thesis driving Flipkart's LLM investment.