Perplexity's revenue increased by 50% in one month, accelerating the shift from search to AI Agents

AIMPACT message, April 8 (UTC+8) — Data obtained by the UK’s Financial Times shows that AI search company Perplexity’s estimated ARR (annual recurring revenue) exceeded $450 million in March, up about 50% from the previous month. The growth is mainly driven by its AI Agent product Computer launched in February and the subsequent introduction of a usage-based billing model: advanced subscription users (monthly fees up to $200) pay extra based on actual usage after they exhaust the included credits. However, usage-based revenue naturally fluctuates more, so it isn’t directly comparable to the earlier pure subscription model. Before the new billing was introduced, Perplexity used two years to grow its ARR from $16 million to $305 million.

This transition marks an acceleration of Perplexity’s shift in focus from a chat-style search engine (once seen as Google’s biggest challenge in two decades) to AI Agents that can carry out tasks for users. The whole industry is exploring how to price Agents that are more compute-intensive. Perplexity currently has more than 100 million monthly active users and tens of thousands of enterprise customers, but it is still not profitable. Its main costs are the inference fees it pays to model providers such as OpenAI and Anthropic. A person familiar with the matter said its advantage is that it can route requests to the most suitable models—for programming using Codex and Claude Code, for writing using GPT-5, and for reasoning using Opus.

Even though the 50% month-over-month growth rate looks striking, the gap remains large in a horizontal comparison: Cursor’s ARR for programming tools has grown from under $100 million in 2024 to $2 billion; Anthropic’s ARR reached $19 billion at the end of February; and OpenAI’s was $20 billion last year. Perplexity’s valuation was estimated at $20 billion in September last year. Investors include Nvidia, SoftBank Vision Fund 2, Amazon founder Jeff Bezos, and Meta’s former Chief AI Scientist Yann LeCun. (Source: Financial Times)