Issue #16

What actually happened

On August 23, 2026, The Information reported, on anonymous sourcing, that Nvidia is in talks to invest in Perplexity as part of an equity round that would value the startup at more than $30 billion, with Perplexity's annualized revenue said to have risen to more than $750 million from under $250 million at the start of the year. Reuters syndicated it, and within a day the story had been re-reported worldwide. The proposed valuation is up 50 percent from the $20 billion Perplexity reached in its September 2025 round, when it raised $200 million.

Why we rate this needs context

I checked what stands behind each number. The talks themselves: one anonymous source, everywhere, at once. Benzinga's own write-up notes "Nvidia and Perplexity did not immediately respond to Benzinga's request for comment." Neither company has confirmed anything.

The revenue figure: $750 million is an annualized run-rate, meaning a recent period extrapolated to a full year, never audited, and Perplexity has never published a real annual figure. Divide the reported valuation by the reported run-rate and you get the number that explains the coverage: 40x, on revenue nobody outside Perplexity has shown a receipt for.

The structure is the part with independent sourcing, and even there the sourcing covers the pattern, not this deal. The Bank for International Settlements described the general shape back in June, months before this story: "Chip makers and hyperscalers take equity stakes in AI labs or neocloud providers, who in turn commit to multi-year purchases of chips or computing power." Nvidia first put money into Perplexity back in late 2023, well before generative search became a crowded category. And a critic quoted the day the story broke put the sharp version on record: "NVIDIA is effectively bailing out anyone in the AI industry as a means of inflating their valuations and keeping them buying compute." That is one commentator's read, not a finding.

The timing is context, not a receipt. The report landed with Nvidia's stock down 4.8 percent over six sessions heading into its August 26 earnings print, three days before Nvidia tells Wall Street how the quarter went. The record does not establish why the story surfaced when it did.

The steelman, and why it still falls short

The good-faith case: Nvidia has been a Perplexity investor since late 2023, so a follow-on is ordinary portfolio behavior, and a run-rate tripling from under $250 million to over $750 million in eight months, if true, can justify a step-up from $20 billion to $30 billion. Nothing about the report requires bad faith. But the case rests entirely on numbers nobody can inspect: unconfirmed talks, an unaudited annualized figure, a 40x multiple. And the reported deal has the shape of the loop the BIS describes: the company reportedly buying the equity is the same company selling the GPUs that revenue gets spent on. If the talks are real and the vendor's money can prop up the customer's valuation, the valuation stops being independent evidence of anything.

The mechanism

This is how the pattern works when it operates, per the BIS description, not a finding that it is operating here. Circular financing manufactures validation: an equity stake from the dominant chip vendor reads as a credibility signal, the higher valuation reads as traction, and the traction justifies more compute purchasing, which flows back to the vendor. Central bankers have a name for that pattern, and it is not a compliment. Anonymous single-source deal stories amplify the effect: syndication makes one unverified account look like independent confirmation from a dozen outlets. Whether this deal, if it closes, actually runs that loop depends on terms nobody outside the two companies has seen.

What to do with this

• Distinguish annualized run-rate from audited annual revenue. The working question here: show me the audited annual, not the annualized run-rate, and tell me who is writing the check.
• When a vendor invests in its own customer, discount the resulting valuation as a traction signal. The BIS has flagged the equity-for-compute-commitment loop as a pattern, not a one-off.
• If a vendor's own stock is propping up its customer's valuation, that "traction" is not a signal worth copying into your own pitch deck.
• Hold our call to account: no signed deal by the August 26 earnings call, and if one lands within 90 days, the number comes in closer to $25 billion than $30 billion, with a compute-purchase commitment attached.