At Happy Llama 2026, India’s premier AI startup conference, a panel of founders, investors, and operators tried to answer a deceptively simple question: Where does revenue actually come from for AI startups today?
The panel included Himanshu Upreti, Co-founder of Ai Palette; Vardhan Dharnidharka, Principal at Stellaris Venture Partners; Vishal Virani, Co-founder and CEO of AI startup Rocket; and Vishnu Acharya, Strategy and Corporate Development Head at Razorpay and Partner at Razorpay Ventures.
Their discussion revealed a simple reality behind the AI boom: even though building AI is easier than ever, building a real business around it is still hard.
Confusion in the AI Market
AI has created a strange market dynamic. Buyers, founders, and investors are all navigating a space where capabilities are evolving faster than expectations.
Dharnidharka observed that even customers are struggling to understand what to buy. “I think with AI, there is a lot of confusion in the market. Not just from the people who are building or investing, but also from people who are buying,” he said.
Enterprises are flooded with tools that appear similar on the surface. That makes differentiation harder for startups trying to sell. As a result, many AI deals today involve more collaboration than traditional SaaS, with startups building solutions alongside customers rather than simply selling finished products.
“There is a lot of co-building and co-solutioning that is required in this ecosystem to build trust on both sides,” he said.
That trust is often what turns a prototype into revenue.
The Problem of Curiosity Revenue
One of the most striking concepts discussed on the panel was what Virani called “curiosity revenue”. In the AI era, the barrier to trying new software has become extremely low. A user can sign up, experiment, and pay a small amount within minutes.
But that does not mean they are a serious customer. “People were buying tools out of curiosity. That is what we call curiosity revenue,” Virani explained.
This behaviour creates misleading signals for startups. A surge of signups or small payments can look like traction even when the product has no long-term retention.
“The biggest problem is serious revenue. How many users actually stay on your platform?” he added.
For AI companies, the real challenge is not acquiring users but identifying the ones who will stay.
Virani described three typical user groups. The first group experiments briefly and leaves. The second group consists of builders or professionals exploring ideas, and who might pay for a few months. The third group represents real businesses with operational needs. Only the third category generates sustainable revenue.
Also, many AI products are built on top of large model providers such as OpenAI or Anthropic. That creates a structural margin challenge. “You have to ask yourself what you are building,” Virani remarked. “Are you building a startup, or are you just doing GTM (go-to-market) for Anthropic and OpenAI?”
If most of the revenue ends up paying model providers, the startup’s own margins shrink dramatically. “Whatever you get in your bank account, 80% of the money will go to Anthropic and OpenAI,” he exclaimed.
For founders, that forces an important strategic decision. They must either create proprietary value on top of models or risk becoming distribution layers for someone else’s infrastructure.
The Enterprise Pilot Trap
If consumer AI faces retention challenges, enterprise AI faces another problem: too many pilots and not enough contracts. Upreti explained that many companies are happy to experiment with AI but reluctant to scale deployments.
“There is a lot of money for experimentation but those pilots are not converting into six- or seven-figure deals,” he said. This happens because AI often sells a vision rather than a well-defined workflow. “When you talk about AI, it is more about hope.”
In traditional SaaS, customers understand exactly what the product will deliver. AI products are different. They promise possibilities rather than guaranteed outcomes. That creates a dangerous temptation for founders. Founders make the mistake of selling something very ambitious, but the product does not deliver on the promise.
The solution is tricky. AI pilots must have measurable outcomes tied to business metrics. “If you are not anchoring your AI solution to a business outcome that has a budget holder, you might end up just doing a pilot,” Upreti said.
Multi-Channel GTM Strategy
Should startups focus on enterprise sales, partnerships, or product-led growth (PLG)? Channels matter, but founder intuition matters more, believes Dharnidharka.
Different products require different GTM motions. Some are bottom-up products adopted by individual users. Others are top-down enterprise tools.
AI startups are also experimenting with a new type of enterprise engagement. Some companies embed engineers directly with customers during pilots, building solutions alongside them.
“Building things has become a lot more commoditised,” Dharnidharka said. “Why not build it right there with the customer and take instant feedback?” That iterative process is increasingly replacing traditional sales cycles.
Virani believes the AI era has fundamentally changed how startups approach the market. In SaaS, companies often focus on one channel early on. AI startups cannot afford that luxury.
“If you are building a hyper-growth generative AI company, it is not possible without a horizontal play,” he said. That means running multiple channels simultaneously. Partnerships, product-led growth, influencer marketing, and enterprise sales all operate at once.
Virani shared an example from Rocket. “We are working with Fortune 100 companies and some of the largest businesses in the world. Many of them came through a $25 card swipe in a PLG motion,” he said.
In other words, even large enterprise relationships can start as self-serve experiments.
What Payment Data Reveals About AI Startups
Acharya offered a unique perspective from Razorpay’s vantage point. The company processes payments for many startups and also invests in them. That gives it a real-time view of what is working.
Acharya said most AI revenue today falls into two categories.
The first category improves existing workflows. Customer support and inbound sales are common examples. Here, the value proposition is easy to measure. “You can compare an AI agent with a human agent and measure the ROI,” he said.
The second category creates entirely new workflows that did not exist before. These are harder to monetise because customers are still figuring out their value.
Acharya also noted a clear pattern in revenue quality. “We see a large movement of low retention revenues,” he observed. Many startups generate early payments that quickly disappear. But a few categories are already producing meaningful recurring revenue.
“We are seeing strong revenue in chat and voice agents,” he said. Other promising areas include coding tools and outcome-driven AI services built for specific industries.
Biggest GTM Mistake Founders Make
The panellists also highlighted the mistakes they observe.
One stood out above all others. Giving the product away for free.
“Free is the best mistake you can make,” Acharya said. Charging early forces startups to prove value and filters out unserious customers.
Upreti echoed the same warning. “Please don’t take any pilot just for the sake of it,” he lamented. Pilots must be time-bound and tied to measurable results.
Virani framed the problem more bluntly. “Use vibe coding, but don’t build a vibe business,” he suggested. AI hype may create rapid experimentation and funding. But companies that ignore unit economics will struggle once the cycle cools.
“Keep your unit economics in control,” he said. “When the hype settles, you should still be there running your business."
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.