Aravind Srinivas and his billion-dollar AI startup Perplexity have grown up — but that doesn't mean he's done taking big swings

Welcome to this special edition of LinkedIn News Tech Stack, which brings you news, insights and trends involving the founders, investors and companies on the cutting edge of technology, by Tech Editor Tanya Dua. This week, we have a profile of Aravind Srinivas , the founder of the AI unicorn startup Perplexity , which ranked #1 on LinkedIn’s Top Startups 2024 list, and top headlines from the world of AI. We’ll be back with our other sections next week. You can check out our previous editions here.

A deep dive into one big theme or news story every week.

When Aravind Srinivas was in middle school, each time he drove past the sprawling campus of the Indian Institute of Technology (IIT) with his parents in Chennai, India, his mother would turn to him and say: “This is where you’re going to study one day.”

Her enthusiastic assertion isn’t unusual. Having grown up in the country myself, I can attest that a majority of middle-class Indian parents have traditionally nudged their children toward one of two professional paths: becoming a doctor or an engineer. And attending one of the prestigious IITs has perhaps been regarded as the most reliable path to the pinnacle of success.

So, when Srinivas cleared the competitive entrance exams and was admitted to IIT in 2012, his mother was, naturally, over the moon. But even she couldn’t have predicted that just over a decade later, her prodigal son would be at the forefront of one of the most disruptive technologies — as the co-founder and CEO of the AI startup Perplexity .

Since its launch in 2022, the buzzy startup has raised over $160 million from a roster of high-profile names, including 亚马逊 founder Jeff Bezos, Meta chief AI scientist Yann LeCun and notable angel investor Elad Gil , reaching the increasingly unattainable “unicorn” status with its valuation surpassing $1 billion. It has grown to over 100 employees and climbed the popularity charts, even edging out ChatGPT in terms of user visit duration (at over seven minutes), according to recent Similarweb data.

The San Francisco-based company ranks No. 1 on LinkedIn’s Top Startups 2024 list, a data-driven ranking of the young companies that are growing fast and gaining attention from job seekers and investors alike.

Perplexity is reimagining online search in the AI era

Perplexity’s ascent can be attributed, in part, to its radical rethinking of how search should evolve and adapt for today’s increasingly AI-driven world.?

The company has sought to differentiate itself with a chatbot-style UI that offers succinct, AI-powered summaries to user queries — replete with links to sources — standing out from traditional search engines like Google, which have churned out pages after pages of endless links and ads for decades.

Srinivas has proudly embraced the disruptor role, and pitched his company as the antithesis of Google. He hasn’t shied away from taking regular potshots at the search behemoth either.?

When 谷歌 ’s AI model Gemini came under fire in March for producing historically and racially inaccurate images, for example, he slammed the company in an interview with The Verge, calling it “poor execution.”?

"I admire what they’ve built at Google — it’s massive, and it works,” Srinivas told me during the first of two hour-long interviews. “But there’s no reason that they should have the last say on search."

The irony is that Srinivas’s admiration for Google co-founders Sergey Brin and Larry Page, as well as CEO Sundar Pichai , is no secret. And, his first-ever research internship was at Google DeepMind in London. But that didn’t stop him from believing that there was a better way to evolve search for the AI era.

During the same internship, he would spend all of his free time voraciously reading up on Google – including “How Google Works,” by ex-CEO Eric Schmidt and former SVP of products Jonathan Rosenberg, and “In the Plex,” by Wired editor-at-large Steven Levy . This is when the entrepreneurship bug bit him.

One key takeaway that stayed with him and would ultimately serve as the impetus behind Perplexity was the first book’s foreword, in which Larry Page reflected that he couldn’t see himself being anything but a professor or an entrepreneur, because those were the only careers where he could pursue his own vision.?

Srinivas realized that as a PhD student at the University of California, Berkeley, he was already on one of those paths. And without a clear startup idea, he decided to devote himself to deep learning models with Ashish Vaswani , one of the authors of Google’s seminal paper on transformer models, “Attention Is All You Need.”

Then came the turning point. By the summer of 2022, even before ChatGPT burst onto the scene, industry insiders were buzzing about early generative AI startups like Jasper, which were starting to make waves among mainstream users and reel in real revenue.?

That’s when Srinivas realized it was the right time to start a company, and teamed up with co-founders Denis Yarats , Johnny Ho and Andy Konwinski to get cracking. While he had crossed paths with Meta researcher Yarats and Databricks co-founder Konwinski before, he barely knew former Quora engineer Ho when he decided to bring him onboard as the company’s co-founder and chief strategy officer. What convinced him? His attitude.

“I don’t think you need to test smart people on how they solve problems — what you need to test is, are they willing to do the thousands of boring things that startups involve?” Srinivas said of Ho. “I asked him to add a simple button to log data — something really basic — and he did it without hesitation. That showed me he’s someone who’s ready to do the grunt work."

In retrospect, Srinivas’s instinct was spot-on, and the timing couldn’t have been more fortuitous. He had rightly assessed that the world was just at the start of the AI boom — the second trend that Perplexity has certainly benefited from — with AI set to reinvent industries and business functions across the board. Like other top AI startups, Perplexity has managed to buck the overall trend and raise money, in back-to-back rounds, in what is otherwise a tough environment for most startups.?

In the beginning, Perplexity wasn’t necessarily looking to come after Google’s lunch. One of the team’s first prototypes, for example, applied OpenAI ’s code generation tool Codex to scrape through databases like Twitter using natural language prompts, tapping AI to quickly surface information like the “top posts about ChatGPT” on Twitter.

At another point, the co-founders set out to develop an enterprise algorithm that could translate natural language into the database language SQL, geared toward data analysts. But they soon realized that most businesses were either locked into other data platforms or didn’t have enough data to require an advanced tool like theirs, so the market opportunity was limited. That’s when they pivoted to search.

Srinivas had long admired PageRank — Google’s proprietary algorithm that drives how the company ranks web pages in its search results — and how it had propelled the company to the annals of tech history. But he wanted to take things up a notch, infusing traditional search indexes with AI, something that was now possible thanks to the rapid advancements in large language models (LLMs) in the span of a few months.?

He envisioned a product that would use AI to synthesize information from various sources and provide users with accurate and straightforward responses, without them having to sift through multiple web pages. The goal was to enhance the way people access information, making it faster, simpler and more reliable. In his view, Google’s ad-driven revenue model began prioritizing money over the user experience — making it ripe for disruption.

"Spiritually, I’m just carrying forward their mission of organizing the world’s information and making it universally accessible and useful,” Srinivas said.

The company has a lot going for it — but isn’t without its challenges?

One of Perplexity’s defining features is its use of citations. Every answer it generates is backed by sources, organized like footnotes on a research paper. This approach not only gives credibility to the results but also addresses the issue of “hallucinations” — a common problem in AI-generated content where the model invents information.

“Citations are like currency in academia — if every sentence you write is backed by one, then you cannot make arbitrary statements that you think are true,” Srinivas told me. “But without one, it becomes a potential hallucination.”

Ironically, citations are what led to Perplexity’s first major public controversy, when publishers including Forbes, The Verge and Conde Nast raised concerns about its content usage and attribution practices. The use of publisher data without adequate sourcing for its “Perplexity Pages” feature generated widespread backlash, including accusations of plagiarism.

The issue came to a head just as Srinivas was turning 30 in May, he recalled, calling the period "pretty difficult."

Reflecting on the episode, Srinivas said that he felt incredibly misunderstood, especially since Perplexity had always prided itself on citing sources. Still, the team used the opportunity as a learning experience, moving quickly to improve the transparency of how sources were credited. Ultimately, the startup announced a partnership program with several media companies, sharing a percentage of advertising revenue with them if their content is referenced in its AI-powered search engine.

One of the guiding principles that Srinivas said he relied on during that time was the persistence he had honed during his academic career, when he would spend half his day in class and the other half in a coffee shop doing his research until late at night.?

“I didn’t have a proper desk back then, so I just made do. That’s also the mindset I brought to Perplexity — this willingness to work through anything,” he said. "I didn’t feel great about how it went down, but it showed me that we were doing something right to be put under the magnifying glass. In the end, I’m happy with how we handled it.”

IVP partner Cack Wilhelm , who broke from convention as an infrastructure investor to lead Perplexity’s consumer technology Series B round of $73.6 million earlier this year, said that Srinivas ‘s never-give-up attitude is what makes him so unique.

“In the most positive of ways, he almost doesn't adhere to social norms and is willing to go to the end of the earth for Perplexity,” she said. “That’s how he got Yann LeCun to invest — by camping outside his office for hours. Most other people would give up.”

The publisher controversy also prompted Srinivas to rethink the new-age tech industry maxim of “move fast and break things,” popularized by Meta CEO Mark Zuckerberg to promote a culture that encouraged moving at breakneck speeds and taking high risks for companies to realize their full potential. Instead, for Srinivas, it reinforced the importance of balancing speed with careful consideration of outcomes.?

"You can’t always just move fast and break things," he told me. "You need to think about how things could break for others, not just for you."

Still, both Gil and Wilhelm lauded Srinivas’s speed and agility as a founder — one of the traits that was heavily discussed in the tech community recently when Y Combinator founder Paul Graham stirred a debate around the “founder mode.”

“The amazing thing about Arvind is, every time we’d discuss an idea, he’d ping me with a demo two days later,” Gil told me. “He was so fast and iterative, but thoughtful at the same time.”

How Perplexity plans to keep growing moving forward

Perplexity declined to share specifics, but the startup said it has a growing user base in the tens of millions (15 million active users as of March 2024), and claims to process nearly 300 million queries a month in 2024 — versus 500 million queries globally in all of 2023. Its fans include 英伟达 CEO Jensen Huang , who has publicly said that he uses the AI search engine “almost every day,” and Shopify CEO Tobias Lütke , who said it has replaced Google for him.?

But converting this momentum into sustainable growth and taking on search’s Goliath won’t be easy. Other companies have tried and failed to disrupt search, including Sridhar Ramaswamy ’s Neeva, which was eventually acquired by Snowflake after struggling to gain market share. And with Google introducing its own AI-generated answers called “AI Overview” this year, Perplexity is up against a formidable competitor.

“With every company of this type, there has to be an emphasis over time on how do they continue to build something that people love, and then how do they distribute it to more and more people?,” Gil said. “So things like expanding product areas and distribution deals (will) tend to be important.”

Even as it tries to challenge Google’s dominance, Perplexity is taking a page out of its rival’s old playbook — with a new spin. While the bulk of Perplexity’s revenue comes primarily from its $20 per month “Pro” version and is supplemented by licensing its API to other organizations, the company plans to introduce ads in the coming months. It’s reportedly in talks with brands including 耐克 and 万豪酒店 as it looks to chip away at Google’s chokehold over the $300 billion digital advertising industry.

Srinivas maintains that Perplexity’s approach is a departure from Google's auction-based ad model, where marketers bid to have a sponsored link placed against search queries. Instead, in Perplexity’s model, brands will be able to bid for a “sponsored” question, which features an AI-generated answer approved by the advertiser.

"Our ads will come as sponsored follow-up questions, not be embedded in the answers or change the ranking of links," he said. "We’re trying to keep the true essence of search alive — offering the right information at the right time without distorting it or corrupting the answer."

Perplexity is not only building its own ad model, but has also been a savvy marketer itself — running campaigns including a seven-figure fake Hollywood movie trailer during the tentpole NBA finals in June, and most recently, a humorous spot during “Monday Night Football” where coach Jim Harbaugh responds to a press conference with the help of its AI engine.

It’s an unusual approach for a startup as young as Perplexity. But it’s neither incidental, nor a surprise to anyone who follows Srinivas on social media, where he has taken it upon himself to regularly tout the latest stats and updates as the company’s de facto spokesperson — a sharp contrast to his quiet and unassuming real-life demeanor. This was more a necessity than an exercise in vanity, he said, laughing.?

“In the beginning, I would just ask our investors to share the updates — but after a while, I started receiving no responses,” he said. “So I decided it was something I had to do on my own, and it’s certainly helped.”

Still, Srinivas knows that building a sustainable business means Perplexity will have to scale beyond advertising and expand its ARR, or annual recurring revenue — a metric that measures a business's yearly revenue based on extrapolating the most recent month’s sales — beyond the last reported rate of $35 million in August.

That’s why the company is working on a new subscription product tailored to large enterprises called “Perplexity Enterprise Pro,” he exclusively shared with LinkedIn News.?

The new tier gives users expanded search capabilities for both internal and external sources of data — making it easier to search across multiple domains at the same time. An investment team deliberating on a decision, for example, could dig up not just public information about a company, but also any internal data it may have as well as analytics from third-party subscriptions.?

This move marks a full-circle moment for Perplexity, which had initially toyed with the idea of powering enterprise search over specific datasets, before pivoting to a consumer application. That isn’t lost on Srinivas, who is betting that a “hybrid approach,” where strong consumer traction feeds into enterprise sales, will help catapult it to the same heights as companies like Slack and Canva, which leveraged a similar strategy.

“There’s a lot of value in starting with a consumer-driven model, as when consumers are willing to pay for your product directly, they become champions of it within their companies," Srinivas said. “Enterprise revenue is more sticky and hard to change. It can become really valuable.”

Srinivas may have matured, but he’s not done taking big swings?

While he still seeks to disrupt the status quo when it comes to online search, Srinivas seems to have changed his tune, saying that Perplexity isn’t in direct competition with Google — or other consumer chatbots like OpenAI.?

Perhaps, he’s been humbled by the startup’s recent scandals, woken up to the likes of LinkedIn parent 微软 and Google investing heavily in integrating AI into their own search products, or simply matured thanks to better counsel over the years. Either way, Srinivas has certainly tempered down since what Wilhelm called his “brute-force” early-founder days.

“Am I a real competitor to them (OpenAI) if I’m a customer of their models?” he asked me. “And in the context of Google, we are hyper-focused on one thing — search — and for them, this is one project among many other priorities.”

He’s also betting on certain technological trends being a tailwind for Perplexity’s continued growth. The startup has thus far kept its spending under check by relying on both open-source models like Llama and those from OpenAI and Anthropic versus building its own, which can require massive compute and infrastructure costs. This approach of not hitching its wagon to one model also means that its product can take advantage of each of their analytical strengths, and improve its own product's performance over time.

As these LLMs become increasingly commodified, Srinivas sees the potential for Perplexity to evolve into a tool that goes beyond search — assisting users with decision-making, task execution and productivity enhancement, much like a personal AI assistant. This means broadening it beyond a tool that answers basic questions to something that offers more in-depth analysis, recommendations and action-oriented solutions.

"Models are getting cheaper and more capable at the same time — and that’s fantastic for us," he said. "It’s rare for a company to be in such a good spot and be well positioned to benefit from both better AI and a growing user base."

Srinivas may have mellowed down when it comes to declaring an all-out war on competitors like Google, but he hasn’t given up his entrepreneurial ambition, he told me.

Having grown up playing cricket, a beloved sport in India, Srinivas was obsessed with the strategies behind the game. He couldn’t resist comparing himself to Virender Sehwag, an Indian cricketer known for his bold, fearless approach.

"When Sehwag was on 194 (runs), he would hit a six to get to 200 — he wasn’t afraid to go big, even when others would play it safe,” he reflected. “That’s something I try to apply. You can’t just protect what you’ve built — you have to keep swinging for the fences.”

Here’s where we bring you up to speed with the latest advancements from the world of AI.

• OpenAI is plotting a for-profit transition; even as its CTO departs. OpenAI is exploring a plan to restructure its primary business into a for-profit benefit corporation, according to Reuters and other outlets citing anonymous sources. The move would grant CEO Sam Altman equity in the AI startup for the first time — a 7% stake, according to Bloomberg. The discussions are happening just as OpenAI Chief Technology Officer Mira Murati announced just yesterday that she is leaving the company, joining a growing list of executives including co-founder John Schulman, who left the company in August for rival Anthropic. Two other executives are also departing. OpenAI is currently in the midst of a funding round, reportedly seeking to raise $6.5 billion at a $150 billion valuation. Relatedly, in a new blog post this week, Altman claimed that we might have "superintelligence" in a few thousand days, and “be able to do things that would have seemed like magic to our grandparents” in the next couple of decades.
• Meta CEO Mark Zuckerberg touted the company’s first pair of augmented reality glasses “Orion” at the company’s annual Connect conference on Wednesday. The techy specs won’t be on sale for a while — Meta says it's still focusing on “internal development.” But as The Verge reports, Zuckerberg believes AR glasses will one day replace smartphones. Separately, Meta unveiled its new virtual reality headset, Quest 3S, which at $299 is cheaper than the previous version. Other announcements included a partnership to support blind and low-vision people, upgrades to Meta’s Llama artificial intelligence model and new voice prompts for its AI chatbot.?
• Google is paying up to rehire AI pioneer Noam Shazeer .?The company is paying $2.7 billion to essentially get back Shazeer, who had “kicked off the AI boom” before leaving the search giant out of frustration, The Wall Street Journal writes. Shazeer then founded his own company, Character.AI, in 2021, after Google declined to release a chatbot he had developed. Now, he is helping lead the giant's development of the next version of Gemini, Google’s most powerful AI technology. The deal to rehire him includes licensing the smaller company’s technology — a strategy deployed by other tech giants to circumvent regulatory approvals needed for acquisitions.
• Meanwhile, 亚马逊 is apparently struggling with its in-house AI. As the company prepares to launch an AI-powered Alexa device next month, the man leading the company’s AI ambitions, Rohit Prasad , is under mounting pressure to keep pace with OpenAI and Google. But Amazon’s in-house AI models are complicating those efforts, according to The Wall Street Journal. Analysts and Amazon insiders say Prasad’s team has “struggled” to develop and refine the company’s technology, falling behind its competitors in the process. Amazon had initially planned to use its proprietary models for the AI-infused Alexa, but later opted to integrate AI technology from Anthropic and other sources.?
• Two of the world's biggest chip makers are mulling investments in the Middle East. 三星电子 and 台积公司 are considering building huge factory complexes in the United Arab Emirates, The Wall Street Journal reports, citing anonymous sources — a move that could change the landscape of their industry.? Expansion of global chipmaking capabilities could bring chip prices down, while boosting production to meet the needs of the exploding AI industry. Executives from both companies have reportedly visited the Gulf nation, although talks are still in the early stages and may not pan out.
• AI is coming to streaming. A new deal between Warner Bros. Discovery and 谷歌 will see the latter's Vertex AI technology used to generate captions for the Max video platform. The tool will be used for Max's unscripted programming to start, and could cut caption-related costs by up to 50%, the companies estimate. It's also expected to speed up the process significantly. Some in the industry worry that these kinds of applications of AI will erase jobs; in this case, Google and WBD say they'll still employ manual transcribers to check the AI's work.

As always, thanks for reading. Pitch me the interesting investors, founders, ideas and companies powering emerging technologies like AI to reach the inboxes of nearly 1 million subscribers plus thousands more on LinkedIn. Follow me for other tech updates. And click 'Subscribe' to be notified of future editions.