Nvidia is in talks to invest in Perplexity at a valuation of more than $30 billion, The Information reported on Sunday, citing people with knowledge of the discussions. The round would be worth billions of dollars.

Start with the status. This is one outlet reporting unnamed sources describing talks that have not closed. Neither Nvidia nor Perplexity has confirmed anything, no other investors in the round have been identified, and the size of Nvidia's own cheque is not specified. Rounds at this stage frequently change in size, price and participants, and some do not happen. Everything below is conditional on a deal that may not exist in this form.

The number that is not a valuation

The revenue figure is the more informative one, because it is a measurement rather than a negotiation.

Perplexity's annualized revenue has grown to more than $750 million from less than $250 million at the start of this year, tripling in under eight months. The reported driver is Perplexity Computer, an agent that professionals use to automate tasks on a computer.

That distinction matters for how the business should be read. A search product monetizes attention and competes with an incumbent that gives its product away. An agent that performs work competes with the cost of the time it replaces, which is a different and generally more forgiving price ceiling. Whether the growth is durable depends on retention rather than on signups, and nothing in this report speaks to retention.

At more than $30 billion against $750 million of annualized revenue, the implied multiple is around forty times. The previous round, about a year ago, was at roughly $20 billion. So the valuation has risen by half while revenue tripled, which means the multiple has compressed rather than expanded. That is an unusual direction for an AI round and is the most interesting thing in the report.

Why Nvidia in particular

Nvidia investing in AI companies is not new, and it is the pattern rather than the instance that is worth watching.

We have reported on Nvidia buying into the electricity supply for the customers who buy its chips, which is the same move one step further back in the chain. A chipmaker that invests in the companies that buy its chips is doing something that can be read two ways, and both readings are legitimate.

The benign reading is straightforward industrial logic. Nvidia's constraint is not demand for accelerators, it is whether the applications that consume them mature fast enough to justify the buildout. Capital directed at promising applications is capital spent widening its own market, and it is what a supplier with a strong balance sheet and a long view would do.

The less comfortable reading is circularity. When a supplier funds its customers, some portion of the revenue it books is money it provided. That does not make the revenue fake, but it does mean the demand signal is partly endogenous, and an outside observer cannot easily tell how much. The concern is not fraud, it is measurement: if you want to know how much independent demand exists for accelerators, vendor-funded purchases are the part of the number you have to strip out first, and the disclosure needed to do that is generally not available.

We flagged this in our coverage of the BIS warning that the AI buildout is a financial stability risk, and it recurs every time one of these investments is reported.

The relevant caveat about scale

Perspective helps here. A few billion dollars is a rounding item against Nvidia's cash generation and against the roughly $3 trillion of AI commitments across nine companies that we have written about. A single investment in a single application company does not move the aggregate picture.

What it does is add one more data point to a pattern, and patterns in this industry have been assembling quickly.

What to watch

Two things, and neither is the valuation.

Whether the deal is announced at all, and on what terms, will show whether the reported talks were advanced or exploratory. And whether any disclosure accompanies it about commercial arrangements between the two companies, because an investment paired with a compute commitment is a materially different transaction from an investment alone.

This story reports unconfirmed talks described by a single publication. It is not investment advice.