Nvidia’s $30B Perplexity Bet on AI Agents
Perplexity has reportedly committed $750M to Microsoft Azure over three years while running at more than $750M in annualised revenue. Nvidia may value it above $30B.
Perplexity has reportedly committed $750 million to Microsoft Azure over three years while running at more than $750 million in annualised revenue — and Nvidia may value it above $30 billion. That is not a search valuation. It is a wager that Perplexity can become the employee who actually gets work done.
Nvidia is reportedly discussing an investment in Perplexity as part of a funding round that would value the company above $30 billion. The valuation would be more than 50% higher than Perplexity’s reported $20 billion mark a year earlier. Neither company confirmed the talks, so keep the champagne in the fridge. But the direction of travel matters more than whether this specific cheque gets signed.
The comfortable story is that Perplexity is an AI search engine taking a swing at Google. That story is now too small.
What Nvidia appears to be buying into is a more valuable proposition: an AI layer that researches, reasons across tools, produces an output and starts behaving like a junior operator. Search is merely the front door.
The reported deal is bigger than a funding round
According to Reuters, citing The Information, Perplexity’s annualised revenue has climbed from less than $250 million at the start of 2026 to more than $750 million. That is blistering growth, even by AI standards.
But the most interesting part is not the revenue figure. It is the shape of the relationship reportedly under discussion.
The Information reported that Nvidia had previously considered paying Perplexity billions to license some of its technology and hire staff, before talks shifted towards a conventional equity investment. If that is accurate, it means Nvidia did not just see another customer for chips. It saw software and talent worth getting close to.
That distinction matters. Nvidia has already won the first phase of AI: selling the picks and shovels. The next phase is messier. Value will pile up in the companies that decide which model to use, connect AI to real business systems, manage cost and turn an answer into an action.
Perplexity sits right in that traffic jam. It began by making internet search feel less like digging through a rubbish tip. Now it is pushing “Computer”, an agent product intended to carry out multi-step work. Perplexity says Computer is available in Microsoft 365 applications including Excel, Word, PowerPoint and Outlook, and can use connectors such as Snowflake, Salesforce and HubSpot for workflows.
That is a long way from “summarise this article”. It is a direct attack on the low-grade knowledge work that fills calendars, inboxes and payrolls.
Why the $750 million Microsoft commitment changes the maths
In January, Bloomberg reported that Perplexity agreed to spend $750 million on Microsoft Azure cloud services. DatacenterDynamics reported the commitment was for three years.
On one hand, that is a serious signal. You do not sign a three-year cloud commitment of that size because you plan to remain a cute chatbot with a few million curious users.
On the other hand, it is precisely why a $30 billion valuation deserves some hard questions.
AI companies are not normal software companies. Old-school software could create a product, sell another licence and enjoy fat gross margins. AI often carries a meaningful cost each time a customer uses the product: inference, retrieval, model access, data movement, storage, security and the humans needed to keep the thing from doing something stupid.
A $750 million cloud commitment is not revenue. It is a very large future cost obligation. It may be entirely sensible if usage and gross profit scale behind it. It may also become a millstone if customer growth slows, model costs fall unevenly, or competitors turn the same capabilities into a feature.
This is the bit too many founders skip because it ruins the mood: revenue growth is not the same as economic quality.
A company can grow rapidly while becoming more dependent on capital. It can have fantastic demand and terrible unit economics at the same time. I have watched plenty of entrepreneurs confuse the two. It is an expensive mistake because the market cheers during the first act and asks where the cash went in the second.
Perplexity’s reported revenue trajectory is impressive. Its cloud commitment shows ambition. Neither fact, by itself, proves durable profitability.
The real product is not search — it is delegated judgement
Google should not be worried merely because Perplexity can provide an answer with citations. Google has survived plenty of better search interfaces.
The strategic threat is the point at which users stop searching, comparing and assembling information themselves. When they instead say: “Review these accounts, pull the relevant data, identify the outliers, draft the recommendation and put it in the board pack.”
That is not search. That is delegated judgement, with a human still responsible for the final call.
The software that owns that workflow gets a far better chance of being paid than software that simply sends someone to a webpage. It also gets embedded deeper into the customer’s systems and habits. Once an AI tool has access to your documents, CRM, spreadsheets, policies and approval chains, replacing it is more painful than switching a search box.
That is why the Microsoft link is strategically revealing. Perplexity’s reported Azure commitment and its Microsoft 365 integrations place it closer to the actual work than the old search battlefield. It is trying to live where people make forecasts, write sales proposals and prepare investment memos — not just where they look up facts.
For Nvidia, that is attractive for another reason. More agents doing real work means more persistent demand for compute. Training frontier models makes headlines; running millions of useful, repeated workflows can make infrastructure demand stickier.
The overlooked risk: everyone wants to own the control panel
Here is the contrarian bit: Perplexity’s biggest problem may not be Google. It may be the companies helping it scale.
Microsoft owns Azure and has Copilot. Nvidia provides crucial computing infrastructure and is building its own software stack. The model companies want direct enterprise relationships. Salesforce, ServiceNow, Atlassian, Adobe and a dozen others want agents inside their own workflow products.
Everybody wants to own the control panel between the user and the model.
That makes Perplexity’s position both valuable and precarious. If it becomes the neutral operator that can route work across models and tools, it can be enormously powerful. If Microsoft, Nvidia or the major model labs absorb the same functionality into their platforms, it risks becoming an expensive feature with a nice brand.
This is why the reported licensing-and-hiring discussion is worth watching. Strategic investments are rarely just financial. They create optionality. A large platform gets a closer look at the technology, the people and the commercial plumbing. The startup gets money, credibility and access — but may become less independent in practice.
Founders love calling this “strategic alignment”. Sometimes it is. Sometimes it is the corporate version of being hugged tightly enough that you cannot move.
What this means for you
If you are a founder, stop pitching “AI-powered” anything. That phrase is already dead on arrival. Identify one workflow where your customer currently wastes money through delay, repetition or poor judgement. Then measure a real result: time saved, revenue recovered, errors reduced, conversion lifted or headcount avoided. If you cannot name the number, you have a demo, not a business.
If you are an operator, do not give an AI tool the keys to the kingdom because it writes decent emails. Start with a contained, high-frequency workflow: weekly sales reporting, supplier comparisons, customer-support triage or first-pass financial analysis. Set a human owner, create an audit trail and track cost per completed task. The aim is not to use AI. The aim is to make a process materially better.
If you are an investor, look past the valuation theatre. Ask four boring questions: What does each completed task cost to serve? How much of revenue is recurring? What happens if a platform partner ships the same feature? And does usage create genuine switching costs or just novelty?
Perplexity’s reported $30 billion valuation is not proof that AI search has won. It is proof that the market believes the next valuable AI company will not just find information. It will turn information into work.
That is the prize. And it is a much bigger one than search.