E-commerce platform Meesho’s proprietary AI engine already drives 74.57% of all orders on its platform to power personalised recommendations, dynamic pricing, and automated cataloging. Now, the company is betting Rs 1,870 crore — which forms 44% of its Rs 5,421 crore IPO proceeds — towards owning the infrastructure to run this AI, which it thinks will be cheaper than renting it from cloud giants like Amazon Web Services, Azure, or Google Cloud.
The investment reflects a critical inflection point: Meesho’s AI has grown so large that the economics of relying on external cloud computing no longer works. The platform’s BharatMLStack AI engine processes 5.92 billion data points daily, and makes 6.4 trillion real-time predictions at peak, as of September 30, 2025, up 40% in just four months since filing its draft red herring prospectus (DRHP). Meanwhile, data storage jumped 34% to 55 petabytes during the same period. Server and software costs surged 60% to Rs 437.2 crore in the six months ended September 2025, as per the company’s red herring prospectus. At this trajectory, continuing to rent infrastructure would mean spiralling costs.
What do analysts say?
“When deploying AI-based solutions at consumer scale, cloud infrastructure costs from providers like Amazon Web Services, Microsoft’s Azure, and Google Cloud can quickly become unsustainable,” said a Bangalore-based analyst tracking the sector. “When you’re processing billions of data points daily and serving 200 million-plus users, the difference between AWS/Azure pricing and owned infrastructure can be hundreds of crores annually.”
According to the company’s RHP, Meesho will deploy Rs 1,390 crore for cloud infrastructure and Rs 480 crore for salaries of AI and machine learning teams at subsidiary Meesho Technologies.
Such optimisations become even more critical for Meesho as it operates at thinner margins as compared to other e-commerce players. The platform operates with an average order value of Rs 274, roughly one-third of traditional e-commerce players at Rs 830-1,000. At such levels, even minor inefficiencies in recommendations or logistics routing can destroy unit economics, analysts suggest.
The payoff for moving to in-house infrastructure is visible with Valmo, Meesho’s logistics aggregation platform. Valmo has scaled from handling just 1.83% of orders in FY23 to 61.06% by April-June 2025, processing nearly two-thirds of the platform’s 1.26 billion orders. By routing orders across 13,683 small logistics partners, Valmo helps sellers save up to 12% on shipping costs per parcel. As more orders get processed through this routing mechanism, load on the AI engine will also increase, validating the need for internal infrastructure.
Benefits of the shift for Meesho
The strategic shift also follows Meesho’s contentious experience with third-party providers. AWS initiated arbitration proceedings against Meesho over Rs 127 crore in alleged unpaid bills under a February 2022 contract, with Meesho filing a counterclaim of Rs 86 crore citing service deficiencies. The company has since moved to Google Cloud, but the dispute underscores dependency risks.
“The upfront capex of Rs 1,390 crore might seem high, but compared to ongoing cloud bills at this transaction volume, the payback period could be 3-4 years,” another analyst said.
Beyond powering its marketplace, Meesho is piloting commercialisation of its AI capabilities—offering GenAI-based customer support to external businesses. Early clients include Spinny, which is using the service to reduce customer support costs in tier-2+ markets.
“If Meesho can successfully commercialise its AI stack, it creates optionality beyond e-commerce,” another analyst said. “But it’s very early days. Amazon took years to build out AWS. For now, this should be viewed as a hedge, not a core revenue driver.”
The company’s financial results show the strategy’s long term promise and near term risks. Losses narrowed 72% to Rs 701 crore in H1 FY26 from Rs 2,513 crore year-ago, while revenue grew 29% to Rs 5,578 crore. Orders jumped 53% to 1.261 billion and annual transacting users grew 34% to 234.2 million. However, adjusted Ebitda loss widened to Rs 552 crore from Rs 54.4 crore, driven by technology and marketing investments.
Major investors Prosus (12.34% stake) and SoftBank (9.3% stake) are not selling shares in the IPO, signalling confidence in the long term full-stack approach, analysts suggest. The IPO opens December 3 at Rs 105-111 per share, with listing scheduled for December 10.