Razorpay SVP says firms must plan for failures beyond their stack
At Inc42’s inaugural CTO Summit 2026, Razorpay SVP Prabhu Ram said firms must assume disruptions can arise in their own systems and in external technology layers, noting no single firm owns all seven layers. He stressed that consumers expect digital services to “just work,” requiring response times within milliseconds or a few seconds.
Ram contrasted this with companies like Apple, Microsoft, and Amazon that have brought critical components in-house, giving them control over the entire seven‑layer supply chain. He noted that Oracle’s Palanivel Saravanan highlighted how firms are moving AI models closer to their data to improve user responses. Ram cautioned that AI cannot fix weak engineering foundations and can worsen problems if deployed without adequate observability—the ability to monitor internal system states and identify failures.
“But in the case of any of us here, we don't own all seven layers,” Prabhu Ram, SVP of engineering, Razorpay said.
“Everybody is looking at how they can get the models much closer to their data, so that the model can learn better and respond to the user better,” Palanivel Saravanan, Oracle said.
The remarks were made during a panel titled ‘The Architecture Of 10X: Engineering For India’s Next Scale Curve’ at Inc42’s inaugural CTO Summit 2026, moderated by Jesper Ludolph, general partner at Exfinity Venture Partners. Other panelists included Kuku cofounder and CTO Vikas Goyal, BookMyShow’s Noel Curtis, and Oracle’s Palanivel Saravanan.
What we know
“But in the case of any of us here, we don't own all seven layers,” Ram said
Speaking at Inc42’s inaugural ‘The CTO Summit 2026’, Ram said companies must assume that disruptions can originate both within their own systems and in the underlying technology operated by other players
Contrasting this dependence on external infrastructure with greater control over the technology stack, Ram pointed to Apple, Microsoft, and Amazon as examples of companies that have brought critical components in-house
“Everybody is looking at how they can get the models much closer to their data, so that the model can learn better and respond to the user better.”
Ram, meanwhile, cautioned that AI cannot compensate for weak engineering foundations and can compound problems when deployed without adequate observability — the ability to monitor what is happening inside a system and identify what has gone wrong