Why the Company That Sells the Chips Wants a Piece of the Search Startup

Here's the deal: The Information dropped a story on Sunday, August 23, saying Nvidia is in discussions to join a new equity round for Perplexity that would value the AI search startup at more than $30 billion. Reuters picked it up the next day, and the Reuters version keeps the hedge intact — every figure is attributed to The Information. Both Nvidia and Perplexity either declined to comment or didn't respond. So nothing here is signed. This is a conversation, not a deal.

The conversation still matters, though. Perplexity's last round valued it at roughly $20 billion — that was the $200 million raise The Information reported on September 10, 2025. Going above $30 billion means a jump of more than 50% in twelve months. And the party allegedly putting the last stamp on that jump is the company that manufactures the GPUs Perplexity runs on. That's the actual story here.

Nvidia is already a Perplexity shareholder, by the way. This isn't a new relationship, it's a bigger one. The September 2025 investor list included Accel, IVP, SoftBank Vision Fund 2, Jeff Bezos, NEA, Databricks — and Nvidia. What's new isn't the check, it's the structure they considered first. According to The Information, before landing on a straight equity stake, Nvidia weighed paying billions to license Perplexity's technology and hire specific people out of it. Sound familiar?

It should, because it happened three days earlier. On August 20, Newcomer reported that Nvidia agreed to license Poolside's "model factory" on a non-exclusive basis for $6 billion, plus a separate $1 billion equity investment at a $12 billion pre-money valuation, with all three founders staying put while 109 employees moved to Nvidia. It isn't an acquisition, but it functions like one. The fact that Nvidia reportedly kicked the same tires on Perplexity and then backed off toward plain equity is the most interesting detail in the whole report.

Three Characters — The Chip Seller, The Interface Seller, and The Silence

Start with Nvidia. It's a chip company that has also become the most aggressive venture investor in AI. Per CNBC's May 9 tally, Nvidia committed more than $40 billion to equity investments in AI companies in the early months of 2026 alone — $30 billion of that into OpenAI. The balance sheet tells the same story from another angle: non-marketable equity securities sat at $22.25 billion at the end of January, versus $3.39 billion a year earlier. Nvidia's annual SEC filing says it put $17.5 billion into private companies and infrastructure funds over the fiscal year, "primarily to support early-stage startups."

Then Perplexity, run by Aravind Srinivas, founded in 2022. It started life as "the chatbot that cites its sources." That identity has shifted a lot. The Comet browser it shipped in July 2025 is now the center of gravity. Comet isn't a browser with an AI button bolted on — the browser itself is supposed to be the agent. Beyond reading and summarizing pages, Comet Assistant executes multi-step tasks on your behalf: booking flights, triaging email, filling out forms. Through 2026 Comet went free and global across Mac, Windows, Android and iOS, and its agentic browsing capabilities got folded into Samsung Internet.

The third character is unusual, because it's the absence of confirmation. Reuters ran the story without independent verification language. Neither company commented. There's no signal the round has closed. And AI has a track record of "in talks" reports that didn't land as advertised — Nvidia's own $100 billion OpenAI commitment being the loudest example. That was announced as a letter of intent on September 22, 2025, and by December CFO Colette Kress was saying publicly that they still hadn't completed a definitive agreement. Nvidia itself demonstrated how far apart an announcement and a signature can be.

Perplexity's cap table explains its position in the stack, too. Bezos, SoftBank, Nvidia. AWS is the primary cloud, and on January 29, 2026 it added a three-year, $750 million Azure agreement with Microsoft, letting it deploy models through Microsoft Foundry including OpenAI, Anthropic and xAI models. The short version: Perplexity rents its infrastructure and borrows most of its models, then sells the interface. Which means the entire valuation rides on one question — who owns the user's first screen.

What Was Reported, and What Nobody Has Confirmed

Three claims sit at the center. One: the new round would value Perplexity above $30 billion. Two: that's more than 50% above the roughly $20 billion round from a year ago. Three: annualized revenue has climbed past $750 million, from under $250 million at the start of the year. Roughly a 3x year.

One honest caveat on that third number. The $750 million ARR figure happens to be identical to the size of the Azure deal Perplexity signed with Microsoft in January. Some analyses have flagged that the two numbers get conflated in coverage. Perplexity has never published an official ARR figure, and some market trackers put the actual run rate closer to the $450–500 million range as of spring 2026. So treat "$750 million ARR" as a reported number, not an audited one.

Item What was reported Confirmation status
New round valuation Above $30 billion Unconfirmed, no comment from either side
Prior round valuation ~$20 billion (closed Sept 2025) Also press-reported at the time
Valuation increase More than 50% Derived from the two figures above
Annualized revenue $750M+ (from under $250M in January) No official company disclosure
Licensing alternative Multi-billion tech license plus targeted hiring Considered, execution unknown
Nvidia's existing stake Shareholder since at least Sept 2025 round Confirmed via investor lists
Azure agreement Three years, $750 million (2026-01-29) Reported by Bloomberg, actually signed

Run the math and $30 billion on $750 million of ARR is about 40x sales. That's uncomfortable in public markets and fairly ordinary in private AI right now. The same argument played out when Anthropic took investment from Microsoft and Nvidia in November 2025 at a valuation in the $350 billion range. The multiple isn't really the question. The question is whether the growth rate that justifies the multiple holds. Perplexity tripled over the last twelve months. Whether it can do that again is basically the whole $30 billion thesis.

It's also worth asking why the licensing route got dropped. For a company like Poolside, which sells model-training infrastructure, carving out the technology and licensing it makes sense. Perplexity's assets aren't really the stack — they're users, brand, and distribution deals. Samsung placement, Comet's installed base, publisher partnerships. You can't license those out of a company. If Nvidia really did pivot to equity, that's the most plausible reason.

Who Gets What

Nvidia's logic is easy to read. It's running a compute-landlord playbook: take stakes in the companies that consume the most Nvidia silicon, and collect on both the hardware revenue and the equity appreciation. OpenAI, Anthropic, xAI, SSI, CoreWeave, Nebius, Mistral, Poolside — the roster keeps growing. Perplexity sits on a different rung, though. Most of that list is either model builders or infrastructure sellers. Perplexity is a consumer-facing app. This is one of the first times Nvidia has reached all the way to the top of the stack.

The top of the stack matters because that's where inference demand originates. How many GPU cycles a search query burns is far less interesting than how many an agent burns completing a task. One Comet flow that researches flights, compares them, and books one is not remotely comparable to a single question-and-answer. Nvidia putting money into the agent app layer is a direct bet on the shape of its own future demand curve.

Perplexity gets more than cash. Being on Nvidia's cap table means better positioning in GPU allocation, which is arguably scarcer than money for AI startups right now. A $30 billion mark is also a recruiting weapon, since option value scales with it. Reports have pointed to a 2028 IPO target, and stepping the private mark up one more rung before that is defensible preparation.

Existing shareholders smile too. SoftBank and Bezos get a 50% paper markup in a year. Accel led a $500 million round at a $14 billion valuation in June 2025 and would be sitting on a little more than a double fourteen months later. All of that is paper, though. Private shares only pay when someone buys them, and the secondary market for AI startups isn't always liquid.

And there's one more stakeholder people skip: Nvidia's own shareholders. When Nvidia buys equity in its customers and those customers spend the money on Nvidia chips, the circular-financing question keeps coming back. The Anthropic deal is the cleanest illustration — Nvidia committed up to $10 billion and Microsoft up to $5 billion on November 18, 2025, and Anthropic committed to purchase $30 billion of Azure compute. The money makes a loop. It's accounting-legal and strategically coherent, but the quality-of-revenue debate isn't going away.

Two Precedents — One Still Unsigned, One Already Dead

Start with the success case: Nvidia and OpenAI. The official announcement went up on Nvidia's newsroom on September 22, 2025 — at least 10 gigawatts of Nvidia systems, with Nvidia intending to invest up to $100 billion progressively as each gigawatt deploys. Jensen Huang said Nvidia and OpenAI "have pushed each other for a decade, from the first DGX supercomputer to the breakthrough of ChatGPT." Sam Altman said "everything starts with compute." Markets loved it and the stock ran.

But that case is a warning as much as a win, because the announcement was a letter of intent. In December 2025, CFO Colette Kress said publicly the companies still hadn't completed a definitive agreement — more than two months after the headline. That's the exact temperature to read today's Perplexity story at. Nvidia's investment discussions are real events. The gap between a discussion and a wire transfer can be several quarters wide.

The failure case comes from the browser side: OpenAI's ChatGPT Atlas. It launched in October 2025 aimed squarely at Comet. On July 9, 2026, OpenAI announced it was sunsetting Atlas, and the product stopped working on August 9, 2026. The reasoning is the part that stings. OpenAI's internal conclusion was that "the browser is a feature, not the destination." Instead of maintaining a standalone browser it folded browsing into ChatGPT and Codex, shipped a Chrome extension, and built a remote cloud browser for agent tasks. Nine months, then out.

That cuts both ways for Perplexity. The good news is the most dangerous competing product is gone. The bad news is that the best-resourced company in the market concluded that a standalone AI browser is not a sustainable category — and Perplexity's $30 billion mark rests on exactly the opposite proposition, that Comet becomes the default interface. For scale: 2026 estimates put Atlas at roughly 10–15 million monthly actives at its peak and Comet at 3–5 million. The category is still small in absolute terms.

Keep Poolside as a third reference point. In the August 20 deal, Nvidia didn't buy the company. It structured $6 billion of technology licensing plus $1 billion of equity plus 109 employees walking across the street — widely read as a design that captures the substance while sidestepping merger review. That the same shape was reportedly considered for Perplexity tells you Nvidia is now operating comfortably in the gray zone between investment and acquisition.

How the Competition Punches Back

Google is the quietest and scariest player here. Chrome still holds roughly 70%-plus of global browser share. Google integrated Gemini into Chrome in September 2025, and in January 2026 it added a Gemini sidebar plus "Auto Browse," an agentic feature that autonomously handles tasks like grocery ordering. Google's strategy is simple: don't ask anyone to install a new browser, put the agent inside the browser already installed. Which is precisely the lesson Atlas taught.

OpenAI retreated from browsers but not from browsing. It absorbed the capability into ChatGPT itself. With ChatGPT still around the mid-50s percent of web visits across the largest chatbots, Gemini in the high 20s and Claude in the high single digits, OpenAI chose to put the agent inside the window users already open daily. Perplexity has to convince people to open a new one.

Anthropic is running a third route — a Chrome extension for Claude plus a heavy enterprise agent focus. And here's the awkward part: Comet's agent runs on Claude Sonnet by default for Pro users and Claude Opus for Max users. A meaningful chunk of Perplexity's product edge sits on a competitor's model. That's the structural weakness under this valuation. Without owning the model, defending margin and differentiation at the same time is hard.

There's a legal front too. On August 4, 2026, the Ninth Circuit vacated the district court's preliminary injunction in Amazon v. Perplexity. The panel held that when a user directs Comet Assistant to reach Amazon, it is the user — not Perplexity — who "accesses" Amazon's computers under the CFAA, because communications route through the user's own machine rather than server-to-server. It's the first federal appellate treatment of whether AI agents acting for users may legally access online platforms. The ruling is narrow, though: it covers the CFAA and California's CDAFA. Breach-of-terms theories survive, and agents with more autonomy or direct server-to-server calls remain exposed.

Amazon, eBay, airlines and banks aren't going to sit still. Post-ruling, responses tend to split two ways — block agents technically, or open paid APIs and admit them on controlled terms. If the second path wins, Perplexity's cost structure changes, because the web it currently traverses for free becomes a toll road.

So What Actually Changes

If you're a developer, nothing in your codebase changes this week. Two things are worth tracking. First, Comet is inside Samsung Internet — as agent-driven traffic grows, your accessibility, structured data and login flows start getting graded by agents rather than humans, and the Ninth Circuit just gave that traffic legal room to grow. Second, your bot policy needs new logic: distinguishing a user-directed agent from a scraper is now a real product decision, not a theoretical one.

If you're an investor, remember two numbers. 40x — $30 billion divided by $750 million of reported ARR. And 3x — the trailing twelve-month revenue growth. The multiple only works if the growth holds. Separately, the fact that Nvidia has planted more than $40 billion of equity across its own customer base means some portion of Nvidia's revenue comes back funded by Nvidia's own capital. However you read that, a balance sheet where non-marketable equity went from $3.39 billion to $22.25 billion in a year is worth pulling up yourself.

If you're an enterprise operator, Comet deployment is going to reach your agenda. Comet Enterprise supports silent MDM rollout across macOS and Windows with hundreds of policies governing what the agent may do. The real question isn't convenience, it's audit trail: when an agent logs into your internal SaaS and takes an action, whose account carries the log, and who owns the incident. Deploying before you can answer that is how you make your security team's next quarter miserable.

If you're a regular user, nothing changes today. Comet is already free, and your browser behaves identically whether the company is worth $20 billion or $30 billion. But keep one thing in mind for the medium term. Search results were free because ads paid for them. In a world where an agent shops and books on your behalf, some other revenue model takes that slot — commissions, subscriptions, paid placement. The capital moving right now is really about who gets to collect the toll on that new model.

🥄 Three Things You're Probably Wondering

— So what does this mean for me? Directly, not much. If you don't use Comet you won't feel it at all. But if you run a web service or hold AI-exposed stocks, the underlying trend — agents traversing the web on users' behalf — is worth watching. This story is a signal that serious capital is positioning around it.

— Is this actually confirmed? No. It's a single-outlet scoop, and neither Nvidia nor Perplexity confirmed it. Reuters ran it as an attribution to The Information. And Nvidia's own $100 billion OpenAI commitment sat unsigned for more than two months after the letter of intent went public. Until the round closes and numbers are disclosed, "these conversations happened" is as far as this goes.

— Can Comet actually beat Chrome? Too early to call, and honestly the indicators lean the other way. Chrome is still around 70% share, and OpenAI killed its own AI browser after nine months on the conclusion that "the browser is a feature, not the destination." Comet's monthly actives are still estimated in the low millions. A $30 billion valuation isn't evidence that Comet wins — it's someone buying the option on that outcome early.

References

Numbers and criteria are as of announcement and may change. Investment calls are yours to make!