Perplexity AI

IWhat it does

Perplexity AI is a private company. Its shares do not trade on any exchange, and I found no sign of an IPO filing⁵⁷. Anything said here about its value is an estimate built on press reports, not on audited accounts.

The company runs an answer engine. You type a question, and it searches the web, reads the pages it finds, and writes a short answer with links to the sources. That is different from a classic search page that shows ten blue links. The company was founded in August 2022¹.

Since then the product line has grown. There is a browser called Comet, which launched in July 2025 for subscribers to the top-tier Max plan²⁶. In February 2026 the company launched Perplexity Computer, a cloud service that runs multi-step tasks using a mix of AI models. TechCrunch counted 19 models and said it was limited to the $200 a month Max plan at launch¹². In April 2026 a version that runs on the user’s own device was introduced, and by May 2026 the Mac app was open to any Mac user, with Personal Computer itself needing a Pro or Max plan¹⁸.

There is also a business side. Perplexity sells an Enterprise plan, and on 28 October 2026 it announced FedRAMP Moderate certification for it. FedRAMP is the US government’s security approval scheme for cloud software. The company says this lets federal agencies buy through the General Services Administration (GSA) schedule, and it describes a promotional offer of $0.25 per agency through 31 March 2027⁴. Developers can also buy access to its search system through an API².

One point on how it works matters for the risks later. Perplexity does not build every model it uses. The Microsoft deal reported in January 2026 gives it access to models from OpenAI, Anthropic and xAI¹³. Its own assets are the web index, the product layer on top and the user relationship. The company says its index tracks more than 200 billion unique URLs⁴. That is a company claim I could not check. Its blog index shows steady output in late 2026, mostly security research and explainers on retrieval-augmented generation (RAG, where a model reads fetched documents before answering), which also work as marketing for the API and enterprise lines³.

IIHow it makes money

The core business is subscriptions. The free tier brings people in. Paid tiers add more searches, more powerful models, and the agent products. The company page lists Pro at $20 a month, and a higher Max tier². TechCrunch put Max at $200 a month¹². A third-party profile adds an Enterprise Pro plan at $40 a seat and a layer of usage-based credits for agent tasks, but I could not confirm either from the company, so treat both as unverified¹⁵.

The company dropped advertising. In 2024 it was among the first AI services to show sponsored follow-up questions. By February 2026 executives told the Financial Times, as relayed by MacRumors, that ads could erode trust in answers, and said they did not plan to revisit the idea¹¹. This is a real choice with a cost: it gives up a revenue line that Google has used for two decades, and it bets that people pay for answers directly. Meanwhile OpenAI began showing ads to free and low-cost ChatGPT users in February 2026, so a rival is trying the opposite path¹¹.

Distribution deals are a second lever, but they have been uneven. In November 2025 Perplexity agreed to pay Snap $400M in cash and equity over a year to put its search inside Snapchat²⁴. Snap said in May 2026 that the relationship had ended amicably in the first quarter and that its guidance assumes no contribution from Perplexity⁸. The Airtel offer in India gave 12 months of Pro free to Airtel’s roughly 360 million customers. According to Sensor Tower and Appfigures data cited by TechCrunch, India monthly active users peaked near 22 million in October 2025 and fell to nearly 14 million by July 2026, a drop of 37%⁹. Appfigures estimated India net mobile revenue at about $156,000 a month in July 2026, which is small next to the user count⁹. Free offers bring users. They do not prove those users pay.

Costs matter as much as revenue here, and I cannot see them. Every answer needs web retrieval and model inference, and Perplexity pays other firms for models and cloud. The three-year, $750M Microsoft Azure agreement reported by Bloomberg in January 2026 is a sign of scale of spending, though PYMNTS notes that AWS remains its preferred cloud provider¹³. No gross margin has been published, so I cannot say how much of each subscription dollar is left after compute.

Revenue lines and what is known (reported = company statement; press-reported = anonymous sources; estimated = third party)

IIIThe market and the competition

Perplexity sits between two large markets. One is web search, dominated by Google, whose parent Alphabet reports about $471B of trailing twelve-month revenue²⁰. The other is general AI assistants. OpenAI claims about 800 million weekly users for ChatGPT, against Perplexity’s ’tens of millions’, according to TechCrunch¹². That gap in scale is the central fact of the competitive picture.

The direct rivals are the same companies that supply Perplexity’s models. OpenAI and Anthropic both offer web search inside their assistants. Google has put AI answers into its own search and offers an AI Pro plan, which it gave free for 18 months to eligible Reliance Jio users in India⁹. OpenAI made a lower-priced ChatGPT Go plan free for a year in India in August 2025⁹. Microsoft and Amazon have their own assistants too; Amazon runs a shopping assistant called Rufus²³. A third-party profile calls Google and Microsoft the largest long-term threats¹⁵. I agree with the direction of that view but it is a judgment, not a measurement.

Perplexity’s response has been to move up the stack. Executives now say they track recent revenue rather than daily query counts, and are focused on a more selective set of users and on enterprise subscriptions¹². The agent products and the Comet browser aim to make Perplexity the place where work gets done, not only where questions get answered. Agents are also where the fights are, because they act on other companies’ sites. Amazon sent a cease-and-desist letter on 31 October 2025 saying Comet breached its terms by not identifying itself as an agent, and Perplexity called the letter bullying²³. PYMNTS later noted that Amazon sued in November¹³.

A fair summary: the market is large and growing, but Perplexity is a small player in it, and its larger competitors can copy the core feature. A second point is price. Google, OpenAI and others use free or cheap plans to win share, and Perplexity has had to match that with promotions of its own, as the Airtel offer shows⁹. Competing on price against firms with far deeper pockets is a hard way to protect margins. A third point is the shift from search to agents: whoever owns the browser and the operating system has an edge in letting software act for the user, and Perplexity owns neither, which is part of why it built Comet and a Mac app¹⁸.

IVThe moat and how it could be disrupted

I rate the moat as contestable, one step above thin. The honest list of advantages is short. The brand is known among people who follow AI. The product experience for cited answers is polished. There is a self-built web index, which the company says is large⁴. And there are distribution relationships, such as Motorola preinstalls and the government channel, which a third-party profile lists¹⁵.

None of these is hard to copy. Cited AI answers now exist inside ChatGPT, Gemini and Claude. Perplexity depends on rivals for its best models¹³, so the supplier can also be the competitor. Switching costs for a search user are close to zero: you can ask the same question elsewhere in seconds. Subscriptions may be stickier for heavy users of Computer and Comet, because their files, memory and workflows live there, but I have no retention data to measure that.

The index and crawling are a double-edged advantage. In August 2025 Cloudflare reported that Perplexity’s crawler changed its identity and ignored robots.txt blocks, estimating 3 to 6 million requests a day from the undeclared crawler on top of 20 to 25 million from the declared one, and removed Perplexity from its verified bot list⁶. Perplexity rejected the report as mistaken and called it a publicity stunt³⁰. If more sites block AI crawlers, answer quality falls, and Cloudflare itself saw less specific answers when the stealth crawler was blocked⁶. Licensing the content instead costs money; Perplexity has a deal with Getty Images and a publisher revenue share⁷¹⁰.

How it could be disrupted: a big platform bundles the same answers for free; model suppliers raise prices or restrict access; publishers win court orders that limit what Perplexity can read or quote; or the browser and agent bets fail because sites block agents. Any one of these would hurt. A moat that can be breached from four directions is not much of a moat yet.

VThe numbers we can see

Perplexity publishes no accounts. All figures below are press-reported or estimated. In March 2025 Bloomberg reported, via TechCrunch, that annual recurring revenue (ARR, meaning the yearly value of subscriptions running today) had reached $100M²⁵. In August 2025 the company’s head of communications said ARR was above $150M, and a source told TechCrunch in September it was approaching $200M¹⁶.

A newer figure comes from Sacra, a research site. It estimates $232M at the end of 2025 and $750M annualised in August 2026, and credits the Computer launch and usage pricing¹⁵. This is a single third-party estimate, not a company statement, and it implies more than a threefold rise in about eight months. I use it as the top of my range, not as my base. I could not find a second source.

Valuations climbed fast. Reported figures are about $520M in January 2024¹, $9B in December 2024¹⁵, $14B in mid-2025¹⁶, $18B after a $100M extension¹⁷, and $20B in September 2025 on a $200M raise¹⁶. I found no confirmed round after September 2025.

One filing is worth stating carefully. SEC EDGAR lists a Form D from December 2024 under the name ’Perplexity AI Inc Nov 2024 a Series of CGF2021 LLC’, with a Delaware address⁵. The name suggests a special purpose vehicle that pools outside investors into Perplexity shares, not the company itself. I could not open the form and cannot tell you the amount, so I give it no weight.

I did not find a reliable current employee count. On users, TechCrunch says tens of millions, while Sacra says over 100 million monthly¹²¹⁵. These disagree by a wide margin, and I show both.

Figures with the type of evidence

VIWho runs it and who owns it

Aravind Srinivas is co-founder and chief executive¹⁵. The other co-founders listed are Denis Yarats, Johnny Ho and Andy Konwinski¹. I could not verify from a primary source who currently holds which role beyond the CEO.

Investors named by TechCrunch in the January 2024 round include Jeff Bezos, Tobias Lutke, Nat Friedman, Nvidia and Databricks¹. A third-party profile adds Accel, IVP and SoftBank Vision Fund 2¹⁵. TechCrunch reported Accel led the $500M round at $14B¹⁶. I do not have the cap table, the share classes or the preference terms, so I cannot say what any one holder is owed first in a sale.

Public statements show a CEO who is willing to be provocative. In March 2026 he called AI-related layoffs a glorious future on the All-In podcast, as Fortune reported¹⁴. In August 2025 the company made an unsolicited $34.5B cash offer for Google’s Chrome browser, which Perplexity confirmed to TechCrunch¹⁷. In January 2025 it was reported to have bid to merge with TikTok US²⁷. These show ambition and also show a company that seeks attention. Whether that is good for owners depends on execution, which I cannot judge from outside.

Governance is opaque. There is no public board list I could confirm, no audited statements, and no information on whether founders hold special voting shares. For an outside reader, that means the usual checks, such as related-party deals, option pool size and stock-based pay, cannot be done. The Snap agreement, which included equity as part of the $400M, is one example of a deal whose full terms are not public²⁴. Key-person risk is real too: from outside, the company’s public voice and big moves appear to run through the CEO¹⁴¹⁷.

VIIValuation: is the last price supported?

Equity value in US$ billions. Not a share price: a private company has none. Latest mark: primary round, press-reported (The Information via TechCrunch); post-money not specified. Staleness: about 14 months old. I found no confirmed newer round.

What the last valuation assumes

At $20B, using my base discount rate, dilution and a 7 times sales exit, the price needs about $8 billion of annual revenue by the end of 2030. At a 12 times exit it needs about $4.8 billion. Starting from my $0.45B run-rate, $8B means growth near 100% a year for four years. At the $0.75B estimate it is closer to 81%. That also needs the business to turn a profit at scale, which no public data shows.

Bear, base and bull

• Bear. Run-rate revenue today of $0.30B, close to the $0.23B end-2025 estimate, treating the $0.75B figure as unproven. Revenue reaches $1.0B by end 2030, about 33% a year. Exit at 4 times sales, below Alphabet’s 7.1x, because growth stalls and margins are thin²⁰. Discount rate 25% for venture-stage risk. Dilution 35% from further raises. Result about $1.0B.
• Base. Run-rate revenue today of $0.45B, between the end-2025 estimate and the August 2026 estimate¹⁵. Revenue reaches $3.0B by end 2030, about 57% a year. Exit at 7 times sales, equal to Alphabet’s multiple²⁰. Discount rate 20%, my own judgment for a private AI company with litigation exposure. Dilution 25%. Result about $7.3B.
• Bull. Run-rate revenue today of $0.75B, taking the Sacra estimate at face value¹⁵. Revenue reaches $8.0B by end 2030, about 76% a year. Exit at 12 times sales, between Alphabet and Snowflake or Datadog at about 18.6x¹⁹²². Discount rate 18%. Dilution 20%. Result about $38.3B.

Comparables used

Where I come down

The latest confirmed valuation of $20B sits well above my base of $7.3B, and also above the safety-adjusted $5.5B. It is below my bull case of $38.3B. To believe the price you would have to accept the Sacra revenue estimate, then expect growth to continue at a pace almost no software company has kept up for four years, and expect the legal and platform fights to be resolved in Perplexity’s favour. The market may still be right. Investors in private rounds see data I cannot: gross margin, retention, and the pipeline of enterprise deals. Fast growth in AI has surprised on the upside before, and strategic buyers might pay a premium for users and a search index. A buyer such as a large platform company would be valuing it differently from a stand-alone discounted cash flow, and my terminal multiples are tied to listed companies, which may be too low for a leader in a new category. Treat my figures as a wide range, not a point estimate. This is research, not investment advice.

What I cannot judge

I cannot judge gross margin, cash burn, retention, the true size of the paying base, the terms of the preferred stock, or the outcome of any lawsuit. Revenue itself rests on press reports and one third-party estimate. I did not read court orders or the Form D contents. I could not run a web search for news in the last few weeks because the search tool was unavailable, so a newer funding round or IPO report may exist that I missed.

VIIIThe IPO

No IPO plan reported. I found no report of a public or confidential IPO filing in the sources I could open. SEC EDGAR shows only a Form D from December 2024 under an investor vehicle name, which is not an IPO document⁵. No exchange, ticker, banks or size have been reported in what I read. Because web search was unavailable, I could not check the very latest news, so this status reflects absence in the pages I opened, not a confirmed denial.

What to watch

• A Form S-1 on SEC EDGAR, public or confidential-filing news
• Any new priced round above $20B
• Company statements of ARR or paid subscribers
• Rulings in the Britannica, Reddit, New York Times and CNN cases¹⁰²⁸²⁹
• The Amazon agent case and any appeal outcome¹³¹⁵
• Whether federal sales after FedRAMP certification show up in revenue⁴

IXWhat could go wrong

Legal risk is the most visible. Perplexity faces copyright and scraping suits from many publishers. TechCrunch lists The New York Times (filed 5 December 2025), the Chicago Tribune, News Corp, Britannica, Merriam-Webster, Nikkei, Asahi Shimbun and Reddit¹⁰. CNN sued in May 2026²⁸. In a ruling reported on 4 August 2026, a federal judge let Reddit’s core claims against Perplexity go forward and dismissed only some state-law claims²⁹. I have not read the court rulings. Damages and any injunction are unknown. A ruling that limits what an answer engine may quote would hit the core product.

Platform risk comes second. Amazon’s dispute over agentic shopping is a test case for whether AI agents may act on third-party sites²³. A third-party profile says a preliminary injunction was issued and then paused by the Ninth Circuit pending appeal; I could not verify that from a court source¹⁵. Dependence on rival model suppliers is a third¹³.

Now the owner’s questions. What most likely destroys this business? I think it is not a single event. It is slow margin squeeze: free offers train users to expect answers at no cost, while model and cloud bills continue, and large rivals bundle the same feature. What can an owner actually take out? Nothing today. There is no dividend, no public market, and the company has raised about $1.5B, which suggests cash is being consumed¹⁷. An owner gets paid only through a sale or listing, and the price then depends on the preferences held by earlier investors.

What has to go right at once? To reach the revenue my math says the $20B price needs, roughly $8B a year by 2030 at a mature-software multiple, several things must hold together. As an illustration only, I put rough odds on four: the consumer product stays relevant against Google and OpenAI (about 60%), the agent and enterprise lines reach multi-billion scale (about 50%), litigation and blocking stay manageable (about 70%), and funding needs do not force dilution beyond about 25% (about 70%). Multiplied, that is about 15%. These odds are my own judgment, not data, and they are not independent in practice. The point is that a chain of reasonable bets multiplies down fast.

The next best alternative for the same dollar is a listed company with disclosed revenue and cash flow. In the comparables list, Alphabet trades at about 7 times sales and carries the search franchise Perplexity is trying to take share from²⁰. This is not advice, and I cannot judge what suits any reader. It is a reminder that private-company paper comes with less information and less liquidity.