Nvidia in talks to take equity stake in Perplexity AI

Nvidia is in discussions to participate in a funding round for AI search startup Perplexity AI, The Information reported Sunday (local time), citing sources. Should the multibillion-dollar round close, Perplexity's valuation would surpass $30 billion, rising more than 50 percent from its last funding round a year ago.

The proposal comes as the two companies have deepened their business ties. Sources said Perplexity's annualized sales have grown from under $250 million earlier this year to more than $750 million, driven in part by its AI agent, Perplexity Computer.

Notable is the shift in deal structure. Nvidia had previously explored paying billions of dollars to license Perplexity's technology and hire some of its staff — an approach it applied to semiconductor startup Groq and networking hardware firm Enfabrica — but has since pivoted to a conventional equity investment. The Groq and Enfabrica acquisitions are widely regarded as de facto acqui-hires.

Nvidia pursued a similar acqui-hire on Thursday, striking a deal with AI model developer Poolside under which it would pay $6 billion to license the company's technology and hire roughly 100 of its engineers.

Nvidia has for several years served as a de facto central bank for the AI startup ecosystem, pouring billions of dollars into dozens of companies that rent or buy its chips. It has already invested in Perplexity on multiple occasions.

Nvidia CEO Jensen Huang has cited Perplexity as his preferred chatbot in multiple interviews.

Perplexity built its AI search engine before ChatGPT gained web-search capabilities, making it a darling of early venture investors during the current AI boom. The company also joined Nvidia's Nemotron coalition — a consortium helping develop Nvidia's open-source models — in March.

Representatives from both companies meet multiple times a week to collaborate on hardware and software integration, The Information said.

Perplexity plans to use Nvidia's next-generation CPU, Vera, to handle agentic AI queries. The company has also signed a deal to use Nvidia chips through CoreWeave, a neo-cloud firm that Nvidia heavily backs.

Nvidia is developing its Nemotron open-source models as a Western counterweight to open AI models from Chinese developers, betting that open Western AI models will drive demand for its hardware. To that end, Nvidia has agreed to spend tens of billions of dollars over the coming years leasing back servers built on its own chips through partners including Lambda and Amazon, effectively becoming one of the largest users of its own hardware.

Perplexity CEO Aravind Srinivas has told CNBC the company is considering a 2028 IPO, and a successful deal would give it more runway to build out its business ahead of a listing.

Now in its fourth year, Perplexity does not train its own models from scratch; instead, it routes AI requests to whichever model best fits the task. Beyond its search engine and Perplexity Computer, the company sells access to both open-source and proprietary AI models developed by others, and is considered one of the top "AI-native" startups by revenue.

The company has raised more than $1.7 billion to date from investors including Nvidia, Accel and SoftBank.