Nvidia is negotiating a stake in Perplexity that would value the AI search startup at more than $30 billion, a bet that follows Perplexity crossing $750 million in annualized revenue. According to Techmeme, the talks mark one of the largest potential valuations yet for a company built primarily on retrieval-augmented answers rather than a foundation model of its own.

The timing matters. Perplexity has spent the past two years positioned as the scrappy challenger to Google Search, ChatGPT's search mode, and Microsoft's Copilot — a product category that skeptics dismissed as a feature, not a business. Crossing $750 million in ARR changes that calculus: it puts Perplexity in the same revenue tier as several well-funded foundation model labs, without the enormous training-compute bill those labs carry.

For Nvidia, a Perplexity stake would extend a now-familiar playbook: take an equity position in the companies most likely to become long-term buyers of its GPUs, rather than simply selling them hardware at arm's length.

Why Nvidia keeps writing these checks

Nvidia has spent 2025 and 2026 building a portfolio of strategic stakes across the AI stack — including OpenAI, xAI, and CoreWeave — that goes well beyond its core business of selling chips. The pattern is consistent: identify companies burning heavily on inference and training, and turn part of that spend into equity upside instead of pure revenue.

• It locks in demand visibility for future GPU generations.
• It gives Nvidia a seat at the table on infrastructure decisions before contracts are signed.
• It hedges against customers eventually building or buying custom silicon instead.

A Perplexity deal would fit that template almost exactly. Perplexity's product is inference-heavy by design — every query triggers multiple retrieval and generation calls — which makes it a steady, scaling consumer of the exact compute Nvidia sells.

What a $750M ARR AI-search business actually looks like

Hitting $750 million in ARR is a meaningful milestone for any startup, but it is particularly notable for one whose core product is a search interface layered on third-party and in-house models. It suggests three things are true simultaneously: enterprise and consumer subscriptions are both scaling, the API/business offering is generating real revenue beyond the free consumer app, and unit economics are healthy enough to justify a valuation multiple north of 40x revenue.

That multiple is steep even by AI-era standards, and it only holds up if growth continues at a similar pace — a dependency any AI builder evaluating a similar model should treat as the central risk, not an afterthought.

What this means for AI builders

The practical signal here isn't the valuation number itself — it's confirmation that AI search has become investable as a standalone category, with its own capital structure separate from the foundation-model labs.

• Teams building on top of foundation-model APIs for search or retrieval products now have a comparable to point to when raising: AI search is not a feature, it's a line item investors will fund directly.
• Infrastructure providers are increasingly willing to trade compute credits or investment for guaranteed future usage — worth exploring for any startup with heavy, predictable inference load.
• Competitive pressure on Google, OpenAI, and Microsoft to keep expanding their own answer-engine products is likely to intensify, in our estimation, now that a pure-play competitor is attracting this level of capital.

AiiN's takeaway

Nvidia doesn't need to own Perplexity to sell it GPUs — it already does. What a $30 billion-plus stake would buy is alignment: a direct financial interest in Perplexity's growth, and a harder floor under the assumption that AI search remains a distinct, well-capitalized market rather than a feature that Google or OpenAI eventually absorbs for free. For builders in the space, the lesson is less about the specific number and more about the mechanism — compute providers are increasingly willing to become investors in their heaviest customers, and that changes who you pitch, and how, when you're raising on the back of inference-heavy growth.