From IIT Hostel Room to $20 Billion: The Perplexity AI Story Nobody Told You

The Boy in the Hostel Room Who Decided to Think Differently

Picture a small, dimly lit hostel room at IIT Madras in 2015.

A young man sits hunched over a second-hand laptop, writing lines of code while everyone else on his floor is laughing, socializing, and enjoying the freedom of college life.

He is not antisocial.

He is not strange.

He is simply consumed by a question that most people his age would never think to ask: what if machines could learn to think the way humans do?

That young man was Arvind Srinivas, and the question he was chasing in that tiny room would eventually become the engine behind Perplexity AI, one of the fastest-growing artificial intelligence startups the world has ever seen.

Nobody handed Arvind a trust fund, a famous last name, or a Silicon Valley address.

What he had was a middle-class upbringing in Chennai, a sharp mind sharpened further by relentless curiosity, and an instinct to spot gaps that billion-dollar companies were too comfortable to notice.

By 2026, Perplexity AI has grown into a $20 billion company that handles over 30 million searches every single day, with major investors including Nvidia, Amazon founder Jeff Bezos, and SoftBank backing its meteoric rise.

This is the story of how that happened — and why it matters for every young person who has ever been told their background is a ceiling rather than a launchpad.

We strongly recommend that you check out our guide on how to take advantage of AI in today’s passive income economy.

From Electrical Engineering to AI: When a Competition Changed Everything

Arvind Srinivas did not walk into IIT Madras with a master plan to build an artificial intelligence empire.

He was assigned a course in electrical engineering, a solid and respectable path that most students followed without questioning.

His friends were living the full college experience — cultural festivals, sports, late-night conversations, and the kind of social freedom that comes with finally living away from home.

Arvind was doing all of that too, but there was always another layer running underneath.

While still in his early undergraduate years, before he had even fully grasped what machine learning meant as a discipline, he entered a machine learning competition hosted at his college.

He did almost no preparation.

He showed up, worked the problem with raw instinct and whatever baseline knowledge he had picked up from reading on his own, and he won.

That moment cracked something open.

It was not just a trophy or a certificate — it was a signal from his own brain that this field, this strange and powerful intersection of mathematics and human cognition, was where he was meant to work.

After completing his undergraduate studies, Arvind went to the University of California, Berkeley in 2017, where he enrolled in a PhD program in Computer Science.

Berkeley is not just a prestigious institution — it is one of the key hubs where artificial intelligence research and real-world application meet, and for Arvind, being there was like being handed a map of the future.

During his time as a doctoral student, he secured internships at three of the most influential AI research labs in the world: OpenAI, Google Brain, and DeepMind, in 2018 and 2019 respectively.

These experiences gave him something that no classroom could fully teach: a ground-level understanding of where the technology was strong, where it was weak, and where the most important unsolved problems were hiding in plain sight.

Perplexity AI, in many ways, was born in those lab corridors.

The Gap That Google and ChatGPT Left Wide Open

In November 2022, OpenAI released ChatGPT to the public, and the internet changed overnight.

For the first time in the history of consumer technology, millions of ordinary people experienced a computer that did not just retrieve information — it understood what they were asking, processed the meaning behind the words, and delivered answers in natural, flowing human language.

It felt like science fiction had arrived quietly at the front door.

But as the excitement settled, users started noticing something important.

ChatGPT, as powerful as it was in those early months, did not have access to real-time information.

It could not browse the internet.

It could not tell you what happened in the news this morning, what a product currently costs, or what the latest research on a medical topic actually says.

Its knowledge had a cutoff date, and beyond that date, it was essentially blind.

Google, on the other hand, had the entire internet indexed and updated constantly, but it responded to every search with a list of blue links that sent users somewhere else to read more.

You searched, you clicked, you read, you came back, you clicked again.

The process worked, but it was fundamentally built around navigation, not understanding.

Arvind Srinivas saw the gap between these two worlds with perfect clarity.

One tool had real-time data but buried it inside endless links.

The other had conversational intelligence but was cut off from the living web.

What if there was a single product that combined both — something that pulled fresh, verified, real-time data directly from the internet and presented it in clean, readable, sourced summaries?

That question became the founding idea of Perplexity AI.

Building Perplexity AI Without a Billion-Dollar Budget

One of the most fascinating parts of the Perplexity AI story is not just the product itself — it is how Arvind built it without the kind of resources that most people assume are required to compete at this level.

Training a large language model from scratch costs hundreds of millions of dollars.

Infrastructure, compute, data licensing, and research talent all add up at a scale that shuts most founders out of the conversation before it even begins.

Arvind made a strategic decision that was both humble and brilliant.

Instead of trying to build his own foundational AI model, he built a platform that integrates the best models that already existed.

Perplexity AI works by pulling together models like GPT from OpenAI, Gemini from Google DeepMind, and Meta’s open-source Llama through application programming interfaces, commonly known as APIs.

The platform sits on top of these models and uses them as engines, while its own technology handles the search layer — crawling the web in real time, pulling verified sources, synthesizing answers, and displaying citations alongside every response so users know exactly where the information came from.

This approach was smart for three reasons.

First, it eliminated the need for massive infrastructure investment at the early stage.

Second, it meant Perplexity AI was not dependent on any single model — if one model became unavailable for legal, commercial, or technical reasons, the platform could switch to another without disruption.

Third, it positioned Perplexity not as a competitor to the large model labs, but as a layer above them — a user-facing product that delivered something none of them were offering directly to consumers.

The result was a lean, adaptable, and genuinely useful product built by a small team with a tight budget and a very clear idea of what problem they were solving.

December 7, 2022: The Launch That Surprised Even Its Own Founder

On December 7, 2022, Arvind Srinivas launched Perplexity AI alongside three co-founders.

The team was small.

The company was new.

The space was already occupied by some of the most powerful technology companies in human history.

Arvind had every reason to expect a slow start — a few hundred curious users, some positive feedback, maybe a mention in a tech newsletter.

What he did not expect was 3,000 queries on the very first day.

Within weeks, that number was climbing.

Within months, it was in the millions.

Within a year, the platform had achieved what most startups spend a decade chasing: a 1,000x growth in usage, measured from its earliest days to the point where it became a recognizable name in the artificial intelligence conversation.

The growth was organic at first — users telling other users, developers writing about it, journalists covering it as a genuine alternative to traditional search.

Then the investors arrived.

Nvidia, the company that makes the graphics processing units that power most of the world’s AI infrastructure and is currently the world’s most valuable semiconductor company, took notice and invested.

Jeff Bezos, the founder of Amazon and one of the most strategically minded investors alive, also put money in.

SoftBank, the Japanese technology investment giant that has backed companies like Uber, WeWork, and Arm Holdings, joined as well.

By 2025, Perplexity AI’s valuation had reached $20 billion.

By 2026, it handles more than 30 million searches every day, making it one of the fastest-growing AI startups in the world by any measure.

Why Google and OpenAI Did Not Build This First

This is the question people ask most often when they hear the Perplexity AI story.

Google has more search data than anyone on the planet.

OpenAI has some of the most advanced language models ever built.

Microsoft has Bing.

Meta has billions of users.

Why did none of them build what Arvind built?

The answer is not about capability.

It is about business models and the incentives that shape decisions at the top of large organizations.

Google earns the vast majority of its revenue from advertising.

Its entire search product is designed to show ads alongside results, and those ads only generate money when users click on links and visit websites.

If Google built a Perplexity AI-style answer engine that gave users a single clean answer without links, users would stop clicking, websites would stop receiving traffic, advertisers would stop paying for placement, and Google’s core revenue stream would collapse.

The product would be great for users and catastrophic for the business.

OpenAI and Microsoft, on the other hand, generate enormous revenue by selling access to their large language model APIs to other companies — the same companies that are building products on top of those models.

If OpenAI launched a clean, neutral AI search interface, it would be directly competing with the businesses it is simultaneously trying to serve as an infrastructure provider.

The conflict of interest is structural, not accidental.

Perplexity AI had none of these constraints.

It had no ad revenue to protect, no enterprise API customers to keep happy, and no legacy product built on a business model that a better product would destroy.

It could simply build the most useful thing for the user, without compromise.

What Perplexity AI Actually Feels Like to Use

To understand why Perplexity AI has grown so fast, it helps to understand what the experience actually looks like for someone using it for the first time.

Imagine you want to know what the best affordable electric cars available in 2026 are, including their range, charging time, and real-world user feedback.

On a traditional search engine, you type the query and receive a list of links — automotive review sites, manufacturer pages, sponsored content, forum threads.

You open five tabs, read five different articles, try to cross-reference the information, and spend twenty minutes piecing together an answer.

On ChatGPT without browsing enabled, you get a confident, well-written response that may be based on data that is two years old.

On Perplexity AI, you type the same question and within seconds receive a structured, sourced answer that pulls from current automotive publications, consumer feedback forums, and manufacturer specifications — all synthesized into a single readable response with numbered citations at the bottom that you can verify with one click.

It is the difference between being handed a pile of ingredients and being handed a finished meal.

The product does not just answer the question — it shows its work, which is the thing that makes it trustworthy in a way that other AI tools have struggled to achieve.

The Vision That Drives Arvind Srinivas in 2026

Arvind Srinivas has never framed what he is building as a search engine.

He frames it as an answer engine — a fundamentally different category of product built on a fundamentally different assumption about what users want when they open a browser and type a question.

The assumption is this: people do not go online to find links.

They go online to understand things.

They want to compare options, evaluate claims, learn new concepts, make informed decisions, and leave the session knowing more than they knew when they arrived.

Search, as it has existed for the last two decades, was built around retrieval — find the thing, hand it to the user, let them figure out the rest.

Perplexity AI is built around comprehension — find the thing, process it, summarize it, source it, and present it in a form that the user can immediately act on.

This vision, if it plays out at the scale Arvind believes it will, does not just challenge Google’s market share.

It challenges the fundamental assumption that has shaped how the internet works since the late 1990s.

In 2026, with the global AI market growing at a pace that is reshaping every industry it touches, that is not a small bet — it is one of the boldest ideas being pursued by any technology company anywhere in the world.

And it is being pursued by a man who grew up in a middle-class family in Chennai, coded through the night in a hostel room nobody thought twice about, and decided quietly that the way things were was not the way things had to be.

What Every Founder Can Take From the Perplexity AI Story

The Perplexity AI story is not just a profile of one exceptional person.

It is a blueprint for how to build something meaningful when the odds are structurally against you.

Arvind did not have generational wealth.

He did not have famous founders as mentors from day one.

He did not start with a hundred-person team or a sprawling office in San Francisco.

He started with a clear problem, a lean strategy, a willingness to use what already existed rather than reinventing everything from scratch, and an almost stubborn refusal to be intimidated by the size of the companies standing in his way.

The decision to build Perplexity AI as an aggregator platform — integrating existing models rather than training from zero — is the kind of strategic thinking that textbooks teach but very few founders actually apply under pressure.

It is the thinking of someone who understands that being right about the problem matters more than being the one who built every tool used to solve it.

The growth — from 3,000 daily queries at launch to 30 million daily searches by 2026 — did not happen because Perplexity AI had more money than its competitors.

It happened because the product was genuinely better for the user in a specific, measurable, everyday way.

That is the lesson.

Not funding.

Not pedigree.

Not timing, exactly — though timing matters.

The lesson is clarity: knowing precisely what problem you are solving, who you are solving it for, and why every other solution currently available falls short in one specific way that you can fix.

Conclusion: A $20 Billion Proof That Background Is Not Destiny

There is a version of the Perplexity AI story that focuses on the billions, the valuations, the famous investors, and the headline numbers.

That version is impressive, but it misses the point.

The real story starts in that IIT Madras hostel room, with a boy who had no roadmap, no rich uncle in Silicon Valley, and no guarantee that any of it would work.

It continues through late nights at UC Berkeley, internships at OpenAI, Google Brain, and DeepMind, and the slow accumulation of a worldview that the future of information was not about finding pages — it was about understanding answers.

It reaches its current chapter with a $20 billion company, 30 million daily users, backing from Nvidia, Jeff Bezos, and SoftBank, and a product that has forced the entire technology industry to reconsider what search actually means.

Perplexity AI did not win because Arvind Srinivas got lucky.

It won because he saw something real, built something honest, and refused to believe that the size of his competitors was a reason to slow down.

In 2026, that story is still being written.

And the most interesting chapters, by every indication, are still ahead.

We strongly recommend that you check out our guide on how to take advantage of AI in today’s passive income economy.