The number came from the stage, not a filing. Srinivas disclosed the milestone in conversation with Semafor editor-in-chief Ben Smith at the outlet's Silicon Valley & The World summit on Monday, and Semafor reported it the same day. A run rate annualizes recent revenue rather than reporting audited results, so the figure is the company's own claim — but the trajectory behind it is independently documented: The Information put Perplexity under $250 million at the start of the year and above $750 million by August.

Four times bigger in eleven months#

The shape of the year is a near-quadrupling in eleven months, and the agent product gets most of the credit. The Information's reporting and Latent Space's newsletter both point to Perplexity Computer, the agent launched in February, as the driver of the second half of the year's growth.

WhenAnnualized revenueSource
Start of 2026Under $250 millionThe Information
August 2026Over $750 millionThe Information
November 2026$1 billionSrinivas, at Semafor's event

For scale, the milestone puts Perplexity in a small club. Cognition, the maker of the Devin coding agent, has also crossed $1 billion in annualized revenue this year, and Cursor is reported at $4.6 billion. The difference is what Perplexity sells: not a flagship model of its own, but an answer engine and an agent layer on top of other labs' models.

The agent did the growing#

Computer breaks a complex request into steps and invokes models such as Claude, GPT and Gemini to do the research, the coding and the paperwork. September brought Automations, which pick up recurring tasks where they left off; November brought Teammates, which take on ongoing jobs rather than one-off prompts. The agent also runs fully locally on supported Macs — a pitch aimed squarely at users who want an agent working on their machine, not just in a browser tab.

Illustration of a small robot assistant checking off tasks on a laptop screen.
Perplexity Computer, the agent product behind the growth, checks off tasks on a laptop. Graphic: AI Frontier Post.

The company has been shipping around the agent, too. A Monitor API watches a topic or a page and fires a webhook when something relevant changes, and a new research lab, Frontier Research, comes with access to more than 10,000 GPUs and plans to publish papers, data, models and code. India, meanwhile, has become the company's largest market — a reminder that the growth is not only an American enterprise story.

One-fifth the burn#

The other half of Srinivas's announcement was a spending claim: Perplexity reached the $1 billion mark having burned about a fifth as much cash as Anthropic. The comparison is his, not an audited figure, and comparing burn across companies with very different businesses is inherently rough — but the logic behind it is straightforward. Anthropic trains frontier models, which means enormous, capital-intensive training runs. Perplexity routes its users to models built by OpenAI, Anthropic, Google and others, so it pays for inference and distribution, not for the training runs themselves.

That is also the vulnerability in the story. Perplexity's moat is product and distribution, not weights: its growth depends on models it does not own, sold by labs that are building their own search and agent products. The aggregator model is cheap to run and fast to ship — and it leaves the company one pricing or access decision away from a much harder quarter.

The IPO clock moves up#

In a CNBC interview the same day, Srinivas moved the listing timetable forward. “The rate at which our business is growing, I wouldn’t rule out filing something in the first quarter of 2027,” he said — a year ahead of the 2028 plan he had set out in 2025. Nothing has been filed, and “wouldn’t rule out” is not a commitment, but the direction is clear: Perplexity wants to be ready to join the public markets soon.

Illustration of a golden stock-exchange bell beside a rising bar chart.
An IPO filing could come as early as the first quarter of 2027. Graphic: AI Frontier Post.

The timing is not accidental. Anthropic prices its own long-awaited IPO this week, targeting a valuation of roughly $1.4 trillion, and the public-market window for AI companies is opening wider than it has been in years. Perplexity's last confirmed valuation is $20 billion, set by its September 2025 round. Latent Space estimates $30–50 billion at the multiples other agent labs command — an estimate that only holds if the unit economics behind the run rate hold up, which is exactly what a public filing would let outsiders check.

What to watch#

First, whether the run rate turns into durable revenue: how much of the $1 billion is subscriptions, how much is enterprise deals, and whether Computer keeps compounding or plateaus once the early adopters are in. Second, the filing itself: a first-quarter 2027 S-1 would force the first audited look at the numbers, including the burn claim. Third, the dependency question: Perplexity's growth runs on other labs' models, and those labs are racing to build the same agent products. And fourth, the valuation: if the market prices Perplexity anywhere near the $30–50 billion range, it will be a verdict on the aggregator model itself — proof that you can build a billion-dollar AI business without training a frontier model at all.

Sources: Latent Space (November 16, 2026); Semafor, via Nakoda News (November 17, 2026); Inc42, reporting Srinivas's CNBC interview (November 17, 2026); TokenPost (November 17, 2026). The Information's earlier reporting on the revenue trajectory is cited in the Latent Space piece.