The rapid rise of AI coding tools has triggered a surge of startups offering so-called “vibe coding” platforms in the US and India. But after an early funding rush and lofty valuations, many of these companies are grappling with a more difficult challenge: sustaining margins.
The platforms rely heavily on large language models (LLMs) from providers such as OpenAI, Anthropic and Google. Access to these advanced models typically comes through paid APIs, making model usage one of the largest cost components for AI coding startups.
Margin Squeeze
“Gross margins across the category are under severe pressure,” Deepak Dhanak, co-founder and chief operating officer of Rocket, told Fe. “Most platforms are operating below 50% gross margins, with many in the 20–30% range,” he added.
Combined with low user retention, the economics of entry-level subscription plans are often weak. “Entry pricing tiers become structurally loss-making,” Dhanak said, noting that most platforms depend on LLM tokens and still subsidise compute costs. Rocket, he added, operates at gross margins above 60%. “We treat margins as a product requirement, not just an accounting output.”
Dhanak said that several platforms have benefited primarily from the novelty around AI coding tools rather than sustained usage. If the use-cases remain limited to experimentation rather than production-grade outcomes, user engagement tends to fall sharply.
“That shows up in retention,” he said. “Month-three retention is under 50% across the category, and for many players it drops into the twenties. When the initial excitement fades, users don’t return unless the product delivers repeatable value.”
Signs of this pattern have begun to appear even among prominent AI coding tools in the US. Anysphere’s Cursor, which reached a valuation of about $29 billion in November last year after multiple funding rounds, has seen some enterprise users reassess their subscriptions.
Bridging the Gap
Even so, startup founders say demand remains strong in certain segments, particularly among small and medium-sized businesses seeking cheaper ways to build custom software.
Mukund Jha, founder of Emergent AI, said several traditional manufacturing firms were using the platform to create their own customer relationship management (CRM) and enterprise resource planning (ERP) systems.
“The gap between what users want — functioning software — and what an AI model produces is still wide,” Jha said. “Our focus is to bridge that gap and give users something deployable and maintainable.”
He added that building software is only part of the process. “It has to be deployed, run and maintained over time. We provide a unified platform for that.”
Emergent AI said it reached annual recurring revenue of more than $100 million within eight months, driven largely by adoption among non-technical users and smaller businesses. Jha said margins remained healthy and expected token prices for AI models to fall over time.
Rocket, meanwhile, said its pricing strategy avoids heavy free-tier subsidies that some rivals have used to attract users.