Perplexity CEO: SMBs Could Drive the Next Economic Boom

Summary · 5 min read instead of 55 min

Aravind Srinivas is the cofounder and CEO of Perplexity, which the description says is valued at about $20B with more than $2B in reported funding from backers including NVIDIA, SoftBank, IVP, NEA, Accel, Jeff Bezos and Elad Gil. Perplexity started as an AI answer engine in 2022 and has grown into a platform for agents, research and knowledge work, including Perplexity Computer.

Srinivas argues that future US economic growth may come from small and mid-market companies that use AI agents to stay lean and capital efficient, and that Perplexity Computer is built to serve them. He says Perplexity's revenue is growing 5x year over year because agents need search, a browser and multi-model orchestration, which Perplexity already had. He sees AI products as having recursive effects rather than network effects, so depth of usage matters more than user count. Perplexity is betting on local and hybrid AI with NVIDIA and Apple instead of building data centers, which he also presents as its answer to security and privacy concerns. He says the real bottleneck for engineers is problem solving rather than writing code, and he wants Perplexity to help turn everyone into a scientist.

Minute by minute

1. 1:22

   Perplexity's brand celebrates curiosity. Srinivas says Perplexity's brand is built to celebrate curiosity, which he thinks will be the human quality people can be proudest of as AI improves. Citing David Deutsch, he argues that once AI can answer everything, the human role is to ask the next questions without fear of being judged. He points to Apple and NVIDIA as companies whose brands are about human progress rather than a specific technology.

2. 4:41

   Great inventions start with one question. Meeting rooms at Perplexity are named after academics, an idea Srinivas says he copied from Demis Hassabis at DeepMind. He argues every great invention starts with a question, citing the few people who asked whether something could beat vacuum tubes, which led to the junction transistor. His favorite namesake is Ramanujan, whom he sees as the spirit of human intuition.

3. 8:16

   A trillion-dollar company can be US-only. Srinivas points to Berkshire Hathaway and JP Morgan, worth about 940 billion, as companies with largely US customer bases. He says US customers have high lifetime value and a constant entrepreneurial spirit, which is why Perplexity built Perplexity Computer to let new companies delegate work to agents. He recalls that he hired his first 10 to 20 people through week-long trials because he did not know how to recruit.

4. 12:27

   The next big companies will be tiny. Srinivas argues that many future companies will be smaller yet make a lot of money. He says growth investors favor Fortune 500 revenue because it seems durable, but US GDP growth may come more from small and mid-market companies that are capital and headcount efficient, AI native and less bureaucratic. Perplexity, he says, builds tools for people like its own team.

5. 13:38

   The American dream rewards outsiders. Asked whether the American dream is still alive, Srinivas says the US is the only country he knows where an outsider can arrive, try to improve things and get funded to do it. He describes coming from a middle-class family in India hoping just to get a job, then finding that others' ambition here pushed him to dream bigger.

6. 16:18

   Stop blaming users for bad prompts. Srinivas names Larry Page as his entrepreneurial hero and credits him with the philosophy that the "user is never wrong". He argues people should not need to learn prompting or which model is good at what; the product should handle it. He says forward deployed engineers are needed mainly because products are bad, comparing it to an iPhone that needed an FDE to onboard everyone.

7. 22:51

   Agents need search, so Perplexity grew 5x. Srinivas traces Perplexity from an answer engine in 2022 to a multi-step search agent in late 2023, deep research in 2024 and browser agents the following year. He argues agents doing real computer work need web access, a browser and multi-model orchestration running in cloud sandboxes, which Perplexity already had. That is why, he says, revenue is up 5x year over year, with enterprise growing faster than prosumers.

8. 26:47

   Recursive effects beat network effects. Srinivas says one solopreneur spends upwards of a million dollars a year on Perplexity while making three million a year automating newsletters for others. He argues AI products have recursive rather than network effects: depth of usage that makes users money matters more than a billion users. He calls frontier labs contributors rather than competitors, since new models add capabilities and their competition lowers serving costs.

9. 29:14

   The data center is coming to your laptop. Srinivas says chat is best served from data centers and would be 30x more expensive locally, but for long contexts like 100k the data center advantage drops to about 3x or close to 1x. He cites slow data center build-outs and back-of-envelope math of 50 watts across 100 million Apple Silicon devices as gigawatts of untapped power. He compares cloud to a bus and local hardware to owning a car, adding privacy concerns about frontier labs.

10. 34:24

    Orchestrating up with NVIDIA, down with Apple. Srinivas describes a fully local portable computer with NVIDIA that can escalate to the cloud, and a version with Apple that starts in the cloud and runs sensitive local files on Apple Silicon. He says finance, legal and healthcare show interest. Sovereign funds want on-prem deployments Perplexity cannot see, which he says limits feedback but builds trust.

11. 39:12

    No plans to build data centers. Srinivas says Perplexity has no plans to build its own data centers and is not betting against them. He argues the shortage is in permits, HBM supply, grid access, interconnection queues and gas turbines, problems he does not want to deal with. He says Perplexity is most incentivized to orchestrate across models, chips and devices, since even Microsoft is tied to Azure and OpenAI.

12. 41:20

    Ask AI to think, not to code. Srinivas argues the best engineers are not bottlenecked on writing code, which AI already makes simple, but on problem solving, specs and constraints. He wants an AI that tells him when building a clone is a bad idea and proposes what to build instead. With more people strong at clarity of thought, he says Perplexity could be 10x more in market cap.

13. 45:43

    Local runtimes as the security answer. Srinivas says the best security answer is letting customers run Perplexity on hardware they own, even air-gapped, rather than trusting a vendor's gateway. On the Hugging Face incident, he expects slower model rollouts because multi-agent risks are missed in single-agent tests. Perplexity is releasing open-source monitors, Bumblebee for devices and Numbat for servers, that audit logs and flag threats.

14. 50:37

    Turning everyone into a scientist. Srinivas wants Perplexity to handle long-horizon research such as building prediction models and testing hypotheses in the world. He says models are not yet long horizon, persistent or parallel enough, especially where truth cannot be verified as in math or code. His hottest take is that AI products are meant to have "recursive effects, not network effects", so user quality matters more than quantity.

Published Oct 12, 2026.