Three post-90s entrepreneurs have been running their startup for four years, and the company is valued at 200 billion.

The valuation of leading AI companies in Silicon Valley is still soaring.

According to The Information, Perplexity AI is wrapping up a new round of equity financing, with NVIDIA in talks to participate. The round is expected to reach billions of dollars, and upon completion, Perplexity's valuation will exceed 30 billion U.S. dollars (about 202 billion yuan), representing a growth of over 50% from the 20 billion U.S. dollars valuation after its previous financing round.

Perplexity AI is a U.S. AI startup founded in 2022, whose founding team includes three post-90s generation members. In December 2025, football star Cristiano Ronaldo also announced his investment in Perplexity and became its global brand ambassador.

This company initially focused on AI conversational search services, and has now launched AI agent products that can automatically perform complex computer tasks. If this round of financing is closed, the four-year-old company will become one of the highest-valued independent players on the AI search track.

Interestingly, before discussing equity investment, NVIDIA initially considered another plan: spending billions of dollars to purchase partial technology licenses from Perplexity and poach some of its employees. But the negotiations eventually shifted to more traditional equity investment.

Post-90s Generation Pushes Google Out of Its Comfort Zone

Many people's impression of Perplexity still stays at "a smarter AI search that can search and summarize". When you ask a question, it directly provides a comprehensive answer with source links. If you only regard it as the search version of ChatGPT, you will seriously underestimate this company.

The core founding team of Perplexity all have backgrounds marked by top AI laboratories.

The soul figure is CEO Aravind Srinivas, born in 1994, from India. After completing his bachelor's and master's degrees at IIT Madras, he pursued a doctorate in computer science at the University of California, Berkeley. He successively interned at DeepMind and Google Brain, and returned to OpenAI in 2021 to work as a research scientist. CTO Denis Yarats comes from Meta AI, and Johnny Ho, Chief Strategy Officer, is in charge of ranking and back-end systems. Both of them are post-90s generation members.

In fact, the company has another co-founder named Andy Konwinski, a member of the post-80s generation. As he is also the co-founder of Databricks, he is reported not to participate in the work on a full-time basis. The four people gathered together with a very simple initial idea: traditional search engines return a bunch of links, and users need to click in to filter information by themselves. Why not directly provide refined answers with sources attached?

This idea was not favored when ChatGPT first became popular. OpenAI was developing general large models, and Google had a search entrance with millions of users. Why could a startup break into this field?

Perplexity chose a differentiated path: instead of training its own foundational models, it invokes and routes models from multiple vendors according to different tasks, and then performs RAG (Retrieval-Augmented Generation) through real-time online retrieval. It can use other people's models and focus on retrieval and orchestration itself, featuring light assets and fast iteration.

In December 2022, only one month after ChatGPT ignited the global generative AI wave, Perplexity, which was founded four months earlier, rolled out its product form. It can directly retrieve information across the entire network, refine answers, and attach source links. This model was once regarded as a "Google killer", pushing Google out of its comfort zone.

With this path proven feasible, the financing pace has also accelerated. From 2023 to 2025, its valuation reached 20 billion U.S. dollars. But the real catalyst that pushed the valuation from 20 billion to 30 billion U.S. dollars is not search, but a product called Perplexity Computer.

On February 25, 2026, Perplexity launched Perplexity Computer, a cloud-based AI agent product for professional users, which can automatically perform various complex work tasks on the computer end. It is reported that Perplexity Computer has formed direct competition with similar agent products from OpenAI and Anthropic.

This product has completely changed the company's growth curve. In March 2026, Perplexity's annualized recurring revenue exceeded 450 million U.S. dollars, and by August, the annualized revenue had further climbed to more than 750 million U.S. dollars, while the figure was less than 250 million U.S. dollars at the beginning of 2026. In less than eight months, the revenue nearly tripled.

The user scale is also considerable. Perplexity executives said that the monthly active users of its search and agent tools have exceeded 100 million, including tens of thousands of enterprise customers. Other data shows that the monthly active users have steadily stayed between 30 million and 45 million in 2026.

From search to Agent, Perplexity has completed the most difficult leap on the AI track.

NVIDIA's Calculation

The key investor behind this round of 30 billion U.S. dollars valuation is NVIDIA.

Jensen Huang once said in an interview that he uses Perplexity in his daily work. But the CEO's personal preference cannot explain an investment of billions of dollars. The underlying logic is that the more tasks Perplexity Computer performs and the longer it runs, the more computing power it consumes. As a result, Perplexity has transformed from a search company into a major computing power consumer, and NVIDIA can sell more chips.

The cooperation between the two sides has already been launched. Perplexity has stated that it will use NVIDIA Vera central processing units to run agent-based AI queries, and has also signed a chip usage agreement with CoreWeave, a cloud service provider heavily supported by NVIDIA. According to The Information, representatives of the two sides now meet multiple times a week to communicate on hardware and software development.

Selling servers on one hand and investing in application entrances on the other, NVIDIA has bet in advance on the next round of computing power demand creators. This is NVIDIA's underlying logic. This strategy is not new.

In 2024, NVIDIA began to negotiate investment with OpenAI. In September 2025, the two sides officially announced strategic cooperation: NVIDIA promised to invest up to 100 billion U.S. dollars in the construction of OpenAI's data centers, while OpenAI promised to purchase a large number of NVIDIA chips.

In March 2025, Microsoft announced that it would pay about 650 million U.S. dollars to AI startup Inflection AI to obtain the license of its AI model and hire most of its employees. Subsequently, NVIDIA and Microsoft jointly participated in a new round of financing of Anthropic, in which NVIDIA promised to invest up to 10 billion U.S. dollars to accelerate the computing infrastructure, and Microsoft made an additional investment of up to 5 billion U.S. dollars, while Anthropic promised to purchase tens of billions of U.S. dollars worth of computing power resources from Microsoft Azure. The common pattern of these transactions is the deep integration of chip supply and equity investment.

According to statistics from PitchBook, NVIDIA participated in nearly 67 venture capital transactions in 2025, higher than the 54 transactions in 2024. Many of these invested companies immediately used the money they received to buy NVIDIA GPUs.

In March this year, NVIDIA took the lead in establishing the Nemotron Alliance, and a host of star teams including Perplexity, Mistral, Cursor, and Thinking Machines were brought into the same camp to share computing power and data.

In addition to negotiating investment in Perplexity, it was recently reported that NVIDIA has agreed to pay 6 billion U.S. dollars to obtain the model usage license of AI programming startup Poolside, recruit about 100 engineers, and make an additional 1 billion U.S. dollars equity investment in it at a pre-money valuation of 12 billion U.S. dollars. This is essentially a disguised talent acquisition, aiming to accelerate the self-development of the open-source Nemotron series models. For Poolside, what NVIDIA wants to buy is technology and talents; for Perplexity, NVIDIA initially wanted to do the same, but finally turned to equity investment.

Objectively speaking, this set of strategies also has hidden worries. How much of the money NVIDIA invested will eventually be converted into real chip purchases, and how much of it is just valuation bubble? Do these companies really have so many users requiring computing power, or is it because NVIDIA keeps investing that makes the demand look particularly strong?

Series A Investors See 200x Return

Perplexity's financing speed is a typical capital myth.

In September 2022, before the product was even launched, Perplexity secured 3.1 million U.S. dollars in seed round financing. In March 2023, the company completed 25.6 million U.S. dollars in Series A financing with a valuation of 150 million U.S. dollars.

Entering 2024: in January, the company completed Series B financing led by IVP, with a financing amount of 73.6 million U.S. dollars and a valuation of 520 million U.S. dollars; in April, Perplexity announced the completion of about 63 million U.S. dollars in financing, with a valuation exceeding 1 billion U.S. dollars, officially becoming a unicorn; a few months later, Perplexity completed a new round of 250 million U.S. dollars financing with participation from SoftBank, reaching a valuation of 3 billion U.S. dollars.

In 2025, the company completed 3 rounds of financing, pushing its valuation to 20 billion U.S. dollars, and football star Cristiano Ronaldo also joined as an investor at this time.

Perplexity's investor list is extremely luxurious: industrial capitals including NVIDIA, SoftBank and Databricks are all gathered here; tech giants including Jeff Bezos, former Twitter vice president Elad Gil, and former GitHub CEO Nat Friedman have made personal investments; even Jeff Dean, head of Google AI, and former YouTube CEO Susan Wojcicki are also on the list. This investment from executives of competing companies is sufficient to prove its high quality.

By the beginning of 2026, another round pushed the valuation to about 21.2 billion U.S. dollars, with undisclosed investors. This new round of financing further raises the valuation to 30 billion U.S. dollars, representing a 200x surge in less than four years.

In terms of returns, based on the 150 million U.S. dollars valuation when Series A investors entered, the return reached 200x in three and a half years. Even for investors who entered in the last round (at a 21.2 billion U.S. dollars valuation in early 2026), they have a 40% paper profit at the current 30 billion U.S. dollars valuation.

Perplexity has an annualized revenue of 750 million U.S. dollars, and its 30 billion U.S. dollars valuation implies a P/S ratio of about 40 times. For horizontal comparison: OpenAI's latest valuation is about 500 billion U.S. dollars, with annualized revenue of about 20 billion U.S. dollars and a P/S ratio of about 34 times; Anthropic has a valuation of about 85 billion U.S. dollars, annualized revenue of about 5 billion U.S. dollars, and a P/S ratio of about 20.5 times. Perplexity's revenue growth rate is indeed more aggressive, tripling in 8 months. But a 40x P/S ratio can only be sustained if the revenue continues to grow exponentially.

But in fact, Perplexity does not develop its own foundational large models. Its Sonar model is fine-tuned based on Meta's Llama, and its core capabilities are built on multi-model routing and retrieval augmentation. Its highest-performing answers rely on third-party models from OpenAI, Anthropic, Google and xAI. Once these models raise prices or restrict access, Perplexity's cost structure will face pressure. Between these "suppliers", it is both a customer and a competitor.

Fortunately, the emergence of Jensen Huang has just completed this puzzle. NVIDIA is building its own open-source model ecosystem through the Nemotron Alliance, and Perplexity is one of the core nodes of this alliance. If Perplexity can rely more on NVIDIA-series models rather than competing companies in the future, the 30 billion U.S. dollars valuation will have a more solid foundation. This investment is both a financial bet and an ecological binding.

Perplexity CEO Srinivas previously stated publicly that the company plans to push for an IPO in 2028. That means this round of financing at a 30 billion U.S. dollars valuation is most likely the penultimate or even the last round before listing.

This article is from the WeChat Official Account "ChinaVenture", written by Li Man, edited by Wang Qingwu, and published with authorization from 36Kr.