NVIDIA is in talks to make a follow-on investment in AI search company Perplexity, according to a report by The Information published on August 24, 2026. The envisioned valuation exceeds $30 billion — more than 50% higher than the company's valuation from a funding round just a year earlier. Neither company has announced an agreement, and the investment amount or resulting stake remains unknown. Still, the negotiations coincide with a period in which Perplexity has expanded beyond search-answer screens into an agentic platform that runs multiple models and external tools. For NVIDIA, more of that workload translates into greater demand for GPUs, CPUs, and inference software.
The "Computer" That Pushed the Valuation Past $30 Billion
Perplexity's previous valuation was reportedly $20 billion, set during a September 2025 funding round. The new figure of over $30 billion represents an increase of at least $10 billion in roughly a year. The Information cited surging revenue as the factor driving the higher valuation.
The company's annualized revenue reportedly grew from under $250 million at the start of 2026 to over $750 million — more than tripling in less than eight months. However, this is a run rate reported for a private company, not audited annual revenue, nor is it ARR reflecting annually renewed contracts. Customer retention, discounting, and gross margins have not been disclosed.
Even so, the growth pattern reflects a shift in the product. On February 26, Perplexity launched "Perplexity Computer," a cloud-based agent that at launch drew on 19 different AI models to handle everything from research to coding to deployment within a single conversation. The service expanded from Max subscribers to Pro and enterprise users, and now consumes credits based on usage. As of August 6, 2026, the pricing guidance set 100 credits equal to $1, with Pro users required to purchase additional credits once their initial allotment is used up.
Traditional answer generation ends once a response is delivered to the user. Computer, by contrast, calls external services, runs code, verifies results, and then proceeds to the next action. Because a single request can trigger many model calls and extended execution time, usage-based billing fits naturally into this model. Dividing the reported $30 billion valuation by the reported $750 million in revenue yields roughly 40x, though both figures were described as "more than," so this is not a precise multiple. Given that Computer is said to have driven part of the revenue increase, whether that usage solidifies into recurring revenue will be a key test of the $30 billion valuation.
From Investor to Vera Testing Partner
The relationship between NVIDIA and Perplexity did not begin with this round. NVIDIA joined as a new investor in Perplexity's $73.6 million Series B in January 2024. In an April of the same year $62.7 million round, Perplexity listed NVIDIA among existing investors who reinvested.
Beyond the equity relationship, the companies' ties have deepened. Perplexity's developer API has run on NVIDIA A100 and H100 GPUs on AWS, with inference optimized using TensorRT-LLM. NVIDIA's customer case studies explicitly cite rising inference costs from user growth as a key challenge for Perplexity. In other words, NVIDIA has served as both an investor and a supplier of the computing infrastructure underpinning Perplexity's services.
In March 2026, Perplexity became a founding member of the NVIDIA Nemotron Coalition. Member companies contribute data, evaluation, and development expertise to open models trained on NVIDIA DGX Cloud. Perplexity is no longer just a customer buying finished models — it has also become a contributor helping to build NVIDIA's model infrastructure.
Further, in July, Perplexity tested NVIDIA's new "Vera" CPU for agentic coding workloads. According to NVIDIA, tasks involving cloning repositories and running tests were about 1.5 times faster than on x86 CPUs, while spinning up multiple sandboxes was up to 1.9 times faster. These are vendor-published measurements, not independently verified results. Still, the fact that Perplexity is considering production deployment of Vera shows NVIDIA extending its footprint into Perplexity's computing stack beyond GPUs.
$750 Million in Revenue, Matched by a Cloud Contract of the Same Size
While agents grow revenue, they also increase the underlying compute costs. Computer calls multiple models for a single request, repeatedly connecting to search and external services and executing code. Credit-based billing can convert usage into revenue, but if payments to model providers and cloud operators rise faster, profit margins can erode.
Bloomberg reported that in January 2026, Perplexity signed a three-year, $750 million contract to use Microsoft Azure. Averaged evenly, that comes to $250 million per year — about a third of the currently reported annualized revenue. However, contract costs are not necessarily expensed evenly each year, so this comparison alone cannot be used to calculate profit margins. Perplexity has also stated that it is not shifting its existing AWS spending, meaning total compute costs cannot be captured by the Azure contract alone.
This arrangement carries two advantages for NVIDIA. Even when Perplexity purchases compute from Microsoft or AWS, NVIDIA still earns revenue as long as NVIDIA accelerators power those clouds. In addition, if Vera moves into production, NVIDIA gains a foothold not only in GPUs but also in CPUs used for agent execution. NVIDIA's motivation for backing Perplexity appears less about locking the company into a specific cloud and more about capturing as much of its resulting compute workload as possible across NVIDIA's own product lines.
However, there is no reported contract stipulating that investment funds must flow back into NVIDIA product purchases. The fact that investment and product adoption are occurring in parallel does not, by itself, establish a circular transaction. What matters is not where the money flows, but whether independent customer demand is driving Perplexity's revenue growth, and whether that revenue leaves sufficient profit after paying for compute costs.
NVIDIA's Toolkit: Equity, Licensing, and Talent Acquisition
According to The Information, before discussing an equity investment, NVIDIA and Perplexity also considered broader forms of cooperation, including technology licensing and talent hiring. It is unclear which of these options remain under negotiation. Still, NVIDIA has already used contracts other than outright acquisitions to obtain the technology and talent it needs.
A notable example is its deal with Groq. According to NVIDIA's annual report, in December 2025 the company obtained a non-exclusive license to Groq's LPU technology and hired some of its employees. NVIDIA did not acquire equity, existing products, or customer contracts, but the total consideration reached $17 billion. This shows NVIDIA can selectively acquire intellectual property, talent, and equity through separate contracts rather than bundling them into a single acquisition.
NVIDIA's investment balance has also expanded rapidly. The book value of NVIDIA's non-marketable equity securities grew from $22.251 billion as of January 25, 2026, to $42.336 billion by April 26 — a net addition of $17.899 billion over three months, with $27 billion in conditional investment commitments still outstanding as of the latter date. While the size of any Perplexity investment remains unknown, the current negotiations fit within this broader pattern of sharply increasing investment activity by NVIDIA.
Perplexity, too, has options. Raising equity capital allows it to fund compute resources and product development without ceding full control of the company. Combining licensing arrangements with talent transfers would give NVIDIA faster access to needed technology and personnel. In exchange, Perplexity's retained development capacity and future revenue sources would depend heavily on the specific contract terms. Whether the deal under negotiation settles into a straightforward equity investment or comes bundled with other arrangements remains to be seen.
Three Conditions Worth Watching Once the Deal Is Finalized
The first condition is whether the $30 billion figure represents a pre-money or post-money valuation, and how much NVIDIA is investing. Even at the same headline valuation, differences in total round size and preferred terms would change both the dilution facing existing shareholders and the degree of influence NVIDIA gains. At this point, neither can be determined.
Second is the substance behind the reported annualized revenue of over $750 million. The value implications differ greatly depending on whether Computer's usage is being sustained by one-off credit purchases or by ongoing enterprise contracts. Disclosure of gross margins, cloud spending, and customer retention rates would allow, for the first time, a real test of whether the roughly 40x valuation multiple is justified.
Finally, there is the question of whether NVIDIA's relationship with Perplexity stands as an independent commercial one. If the timing of Vera's production rollout, the terms under which NVIDIA products are purchased, and any contractual links to the investment are disclosed, it would become possible to separate strategic investment from demand generation. If Perplexity can grow its agentic business into a highly profitable operation while freely choosing the optimal compute infrastructure across multiple clouds, then the $30 billion figure would represent not merely a valuation of a search company, but a valuation of a new software execution layer.