NVIDIA in Talks to Invest in Perplexity, Valuation Exceeds $30 Billion, Shovel Sellers Will Ultimately Buy the Mine
According to The Information, NVIDIA is discussing participating in a new round of equity financing for Perplexity, with a valuation exceeding $30 billion and a financing scale of several billion dollars.
Compared to a valuation of about $20 billion in September 2025, this round has increased by at least 50%. Perplexity's annualized revenue has risen from less than $250 million at the beginning of the year to over $750 million, with around $500 million in April, significantly driven by Perplexity Computer (a cloud AI agent for professionals that automates computer tasks).
NVIDIA had previously considered another path: paying several billion dollars to acquire some technology licenses from Perplexity and poaching a group of employees. Negotiations then shifted to ordinary equity investment, with specific reasons not yet disclosed.
Last week, NVIDIA adopted a similar "license + poaching" structure with Poolside: paying $6 billion for a non-exclusive license for model development software, planning to recruit 109 employees, while also investing an additional $1 billion (pre-investment valuation of $12 billion). Perplexity had once discussed a similar arrangement but has now shifted to becoming a direct shareholder.
Market mechanisms indicate a clear event-driven nature. Funds are flowing into high-growth AI search and agent sectors, with NVIDIA, as an existing shareholder, increasing its stake to secure ecological binding and potential computing power demand; beneficiaries include existing shareholders of Perplexity and high-valuation growth stocks, while those under pressure are AI application companies with lower valuations or slowing revenue growth.
NVIDIA is already one of the early investors in Perplexity, having previously participated in multiple rounds of financing alongside SoftBank, Accel, IVP, Jeff Bezos, and others.
Source: Public Information
ABAB AI Insight
NVIDIA has been continuously penetrating the AI software layer in recent years through a combination of "licenses + talent + equity". Previously, it has used similar non-acquisition structures to acquire technology and engineers from Groq and Enfabrica, and the Poolside case has expanded the scale to a $6 billion license. The shift from licensing and poaching to pure equity in Perplexity shows its greater emphasis on cash flow at the application layer and long-term binding in agent scenarios, rather than merely internalizing model capabilities.
In terms of capital pathways, NVIDIA reinvests chip profits into downstream software and applications, forming a closed loop of "selling shovels + holding mines". Perplexity's revenue has jumped from less than $250 million to $750 million, mainly relying on the Computer product landing in professional workflows, proving that agents are moving from demonstrations to paid scenarios. NVIDIA's increased stake can secure its future GPU/CPU procurement while sharing in valuation expansion.
Similar cases can be seen with Microsoft's talent + licensing arrangements with Inflection and Google's similar operations with Character.AI. The current AI application layer is transitioning from model competition to scenario implementation and revenue validation, with Perplexity entering through search + agent, shifting from a traffic tool to an enterprise automation tool.
Essentially, this represents a restructuring of the industry chain. Hardware giants are no longer satisfied with just selling chips; instead, they are embedding directly into the application layer through capital, upgrading customer relationships from "procurement" to "equity binding". The mechanism lies in the mutual reinforcement of computing power demand and software value: applications drive chip sales, chip profits feed back into application valuations, forming a closed loop.
ABAB News · Law of Cognition
- Shovel sellers will ultimately buy the mine
- Licenses are alternatives to acquisitions, equity is the ultimate binding
- Revenue curves are more honest than valuation curves.