The alpha, twice a week
That was 4 min.
Get the best new moments in your inbox every Wednesday and Saturday. It's free.
Check your inbox
We sent a link to your address. Click it to finish signing up. Not there in a minute? Look in spam.
or
Silicon Valley Girl · 4 min · Sep 26, 2025
In AI, build what you are obsessed with
Aravind Srinivas argues that training models takes an ongoing lab-scale effort, while product builders still face competition from labs and Big Tech. His bet is to pursue a problem deeply enough to outlast competitors.
Aravind Srinivas is cofounder and CEO of Perplexity AI.
Held research roles at OpenAI and Google DeepMind before co-founding the AI search engine Perplexity in 2022.
Marina Mogilko, Host · Host, Silicon Valley Girl
Co-founded LinguaTrip and built Linguamarina, an English-learning YouTube channel with almost 9M subscribers.
About this segment
Aravind Srinivas argues that training models takes an ongoing lab-scale effort, while product builders still face competition from labs and Big Tech. His bet is to pursue a problem deeply enough to outlast competitors.
From “$20B AI CEO: The ONLY trait for success in the AI era | Aravind Srinivas, Co-founder of Perplexity”, Sep 26, 2025. Watch the full episode
Thanks. We'll look at and fix what's off.
“So I would just say the bet you can make is do what you truly are obsessed about.”
Aravind Srinivasat 35:10 · Silicon Valley Girl35:10Takeaways
- Building a model takes more than the cost of one training run.
- AI labs and big tech companies will expand into working markets.
- Choose a problem you care about enough to keep pursuing.
In this episode
Full episode33:06In AI, build what you are obsessed with4:15
- One more moment in the full episode
Transcript
Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.
- 33:18
Do you think opportunity still exists aside from, like, everyone's building a JPT router, right? Do you think there is still opportunity for that or people should switch? And what are the areas that they should be looking at? Realistically, it's gotten much harder. Yes, like you would want to build a GPT wrapper because building your own models, like basically way harder. It's almost impossible for a beginner.
- 33:37
And I don't even think people should be looking at it as like building a model once. That's the mistake everybody made. It's whoever tried to build models around the time we started. I was like, yeah, how much does it cost to launch? I even had an investor who now went to Meta and it's like, oh, it's going to cost $200,000 to print a GP3 model, so why can't you do it? I'll give you like a million and you spend 200K on the model.
- 34:01
It's not about that one model. It's about that relentless exercise of iterating the models, building the next cluster, hiring the talent to go figure out what the next model should be, and continually iterating, bringing down the cost, improving the capabilities, improving the reasoning, all that stuff. That requires a lab and a cluster. and the people and the milestones and some way to actually turn all those models into a product and a business and revenue.
- 34:36
And then you end up basically saying, oh, you have to build another open air, another end product. What are you like? Why do you need to exist? Like, okay, where's the talent arbitrage? So that's why I think product is actually at least, you know, a little more statistically probable. But even there, you have to compete with the likes of the labs themselves, which are building their own products, and the big tech companies, and like upstarts like us.
- 35:00
And no one's going to be like, oh, Perplexes is going to do research. I can go do the commerce stuff. Perplexes and commerce. Exactly, right? Everyone's building horizontal products. And that applies to us too. Oh, ChatGPT will never do shopping. ChatGPT will never do travel. We won't do finance. All that is like not true. Like everyone's going to do anything that works. So I would just say the bet you can make is do what you truly are obsessed about.
- 35:32
Because fundamentally, it's a bet on yourself. It's not a bet on the market. It's not a bet on ecosystem, like what competitors will do, will not do. like don't try to be this uh whiteboard uh strategy master like it's completely pointless like when your idea works and gets like 100 million or like billion in revenue always expect existing people to go after it because everyone's looking for that incremental revenue in ai because the capex is so high so the only way to justify all this is to actually turn that into like business profits and then so they'll go after you so the only thing you can bet on is whether you're so obsessed about a topic that you will do it anywhere regardless of all the odds stacked against you and then you'll prove the world wrong because you go so far deep into that and no one
- 36:21
cared about the problem more than you did and for us like for me like that's where the whole you know you asked me what my most important trade is like i love learning and and this is one of the best apps for learning it's designed for us it's designed for our founders like we built it for ourselves key bet was everybody in the world allows everybody in the world allows having an AI that they can summon to go do stuff for them. Everybody wants more time in their lives. So I think build for yourself. Hopefully, that's a thing that a lot of people in the world want.
- 36:52
And therefore, you can turn it into a scalable product and a scalable company. And get acquired at some stage, I guess. Thank you. That's a great answer. My last question, do you use anything aside from perplexity in terms of AI apps? And what are your tops are your favorite then? Well,