You have almost certainly talked to Gupshup. Not the company - it does not really talk to people - but its pipes. The text from your bank confirming a transfer. The WhatsApp note that your food is three minutes away. The OTP you copy and paste without reading. A large share of those little messages, sent by tens of thousands of businesses across 130-odd countries, pass through one company most of its own users have never heard of.
That anonymity is the point, and it is also the strategy. Gupshup sells infrastructure - the unglamorous plumbing that lets a business hold a conversation with a customer on whatever app that customer already lives in. WhatsApp. SMS. RCS. Instagram. Voice. The company describes its own history as a move "from messaging to conversations," which sounds like marketing until you notice that the whole industry has been trying to make the same leap and most of it is still stuck on step one.
A near-death, then a business
Gupshup was not born as an enterprise platform. It started in 2004 in India as a consumer SMS network - a way to blast text messages to groups back when a phone screen was two inches of green and a text was a small luxury. It was co-founded by Beerud Sheth, an IIT Bombay and MIT engineer who had spent five years on Wall Street structuring mortgage bonds before deciding that building software was more interesting than pricing it. He had already co-founded Elance, the freelancing marketplace that eventually merged into Upwork. Gupshup was his second attempt at building infrastructure nobody notices and everybody uses.
The middle years were rough. Around 2014 the company chased the consumer-chat dream with an app called Teamchat, built around interactive "smart" messages. It did not become the next WhatsApp. What saved Gupshup was the decision to stop competing with messaging apps and start serving the businesses that needed to reach people inside them. By 2016 it had pivoted to an enterprise bot platform, plugging into Slack and Telegram. In 2018, when WhatsApp cracked open a business API that most of the industry did not yet understand, Gupshup was one of the first through the door.
That early WhatsApp bet is the kind of thing that looks obvious in hindsight and reckless at the time. Being first on the world's largest messaging app, in the markets where that app is the internet, gave Gupshup a distribution head start that money cannot easily buy. When you are the provider a bank in Mumbai or a retailer in Jakarta calls to get onto WhatsApp, you accumulate something more durable than a feature: you become the default.
What it actually sells
Strip away the category jargon - CPaaS, conversational AI, omnichannel - and Gupshup does three connected things. It carries messages across more than 30 channels through a single API, so a developer wires up once instead of thirty times. It gives non-developers tools to build bots and run campaigns without writing code. And, increasingly, it sells the conversation itself: AI agents, trained on a specific industry, that can handle a sales inquiry or a support ticket end to end.
The AI push is built on a model Gupshup calls ACE, and on agents it markets as "industry-trained" - the idea being that an agent handling a credit-card dispute needs to know different things than one recommending a pair of shoes. In 2023 the company shipped Auto Bot Builder, which reads a company's website and documentation and assembles a working chatbot from it, and it wired in Meta's Llama models. The pitch to a CFO is blunt: an agent that never sleeps, speaks thirty channels, and costs a fraction of a call center.
There is a reason the "industry-trained" label keeps coming up. A generic chatbot fails in a specific way: it is confidently wrong about the one thing a customer actually needs. Gupshup's argument is that the value is not in the model at all but in the context around it - the product catalog, the account rules, the regulatory guardrails, the tone a bank is allowed to use versus the tone a sneaker brand wants to. Building that context, per industry, is slow and unsexy. It is also the moat, because it is the part a rival cannot scrape off a homepage.
Who is actually paying
The customer list skews toward businesses that message at scale: banks, retailers, e-commerce apps, fintechs, airlines. Named users include HDFC Bank, Unilever, Swiggy and Ola - the sort of companies that send millions of notifications a day and cannot afford for the "your order shipped" text to fail. Depending on which count you read, somewhere between 50,000 and 100,000 businesses and developers use the platform. The volume number is the one that matters to the business model: more than 120 billion conversational messages a year.
That is the shape of the money. Gupshup runs a hybrid of usage fees - a markup on every message routed through carriers and WhatsApp - and software subscriptions for its bot-building, campaign and agent tools. It is a volume game with thin margins per message and enormous throughput, which is exactly why owning the AI-agent layer on top matters: agents are where the richer, stickier revenue could live.
Where it sits on the board
The obvious comparison is Twilio, the American company that won developers' hearts with clean documentation and a credit card. Sinch, Infobip, Bird and Vonage crowd the same space, along with regional players like Route Mobile and conversational-AI specialists such as Yellow.ai and Haptik. What separates Gupshup is less a single feature than a footprint. It went deep in emerging markets first - India, Southeast Asia, Latin America - where messaging is not a channel but the channel. Twilio won the developer; Gupshup won the geography.
Its answer to competition has been to buy the missing pieces. In a roughly twelve-month stretch it acquired Dotgo for RCS, Knowlarity for cloud telephony and voice AI (a deal reported around $100 million), and Active.ai for conversational banking. The work of welding three companies into one "conversation cloud" is not glamorous. It is also the kind of integration that is hard to copy from a standing start.
The founder's pattern
It helps to understand Beerud Sheth to understand the company, because Gupshup is his second version of the same idea. Elance, his first company, was not a consumer product people bragged about using - it was the marketplace and payment rails underneath a whole category of work, the freelancing economy that Upwork later carried forward. Gupshup is the same move in a different medium: not the app you open, but the layer that makes the app talk back. Both are bets that the durable money in technology sits one level below the surface, in the infrastructure everyone routes through and no one screenshots.
That instinct shows up in how Gupshup weathered its hard years. A founder chasing headlines would have leaned harder into the AI-agent hype and burned cash to look fast. What the company did instead was cut, consolidate its acquisitions, and keep its head down on the parts customers actually pay for - deliverability, compliance, uptime. It is a less exciting story than a demo video. It is also the reason a bank trusts you with the message that moves its money.
The reset years
The 2021 valuation was a peak, and peaks in software have a way of getting revisited. In late 2024 investor Fidelity marked its Gupshup stake down by roughly 65 percent, implying a valuation closer to $486 million. The company went through multiple rounds of layoffs across 2024 and 2025, trimming its headcount to somewhere around 770 as it chased profitability rather than growth-at-all-costs. It reported a 55 percent jump in revenue in FY23, which is healthy, though the base is modest and the private numbers stay private.
None of that is unique to Gupshup - the whole late-stage messaging and CPaaS category re-priced. What is telling is what the company did with the pressure. Instead of retreating, it narrowed. The bet now is that the same distribution it spent twenty years building - the channels, the carrier relationships, the compliance across 130 countries - is the hardest part of the AI-agent story, and that everyone racing to deploy agents will eventually need pipes that already work. Whether that bet pays out is the open question. The pipes, at least, are real.