Perplexity CEO Aravind Srinivas Talks AI Boom

**SPEAKER_1** (0:02)
Bloomberg Audio Studios, Podcasts, Radio, News.

**Caroline Hyde** (0:07)
It's not just chip makers pitching the future of computing at Computex. Perplexity took the stage with Intel to unveil what it calls the world's first hybrid local server-agentic inference orchestrator, a phrase that sounds like it was generated by AI itself. Here with more Perplexity CEO Aravind Srinivas. Aravind, it's great to have you back on the show. I spent all morning thinking about how do I explain this?
Basically, the orchestrator is there. It's a piece of software to decide whether A or a part of an AI workload is best done locally on device at the edge, or if it needs the superior computing of Cloud Server. Is that right? Have I nailed what you're trying to solve for here?

**Aravind Srinivas** (0:53)
That's correct. Thank you for having me again.
That is exactly correct. You don't want all your compute centralized in gigantic servers and everything running through the largest frontier models. You're already reading reports of how people are freaking out about their token costs. Some people are spending half a billion dollars per month per engineer. What you actually want is efficient token value per watt per user.
And that requires orchestrating privacy, accuracy, intelligence and costs all together in one single unified system. And that orchestration capability requires hybrid model between server side and the local. And that's what we demoed today with Intel and we're actually chip agnostic. So our solution works with Intel. It works with NVIDIA RTX. So just like how we've been model agnostic, we plan to be chip agnostic here.

**Caroline Hyde** (1:57)
That's interesting. So my next question was going to be why Intel?
What is it about Intel's role in the AI PC market and on the server side that makes it work? But if you're agnostic, what's the breakthrough that you've cracked? Like if you've written the software, what is it you've managed to achieve in how those workloads are diverted? Okay, go a bit further.

**Aravind Srinivas** (2:18)
So like I said, you want one single system to route across models, files, tools, chips, servers and decide when to use which model or when to use your local file system, your local sub-agent model, your local LLM or when to use a frontier model for depending on the task and the prompt or depending on the confidentiality and sensitivity of your files or apps.
That requires you to make clever orchestration decisions, balance trade-offs between accuracy and cost. And that's what we're doing in our software. And then the computer is essentially an operating system that balances all these different objectives simultaneously.

**Ed Ludlow** (2:59)
I mean, you sit in such an interesting place as an orchestrator, whether it be letting people use your own in-house models, whether it's using a mix of third-party models. And the third-party models are up to a lot right now. Aravind, I just want to get your take on how you feel about competitive moats or competitive threats if these big companies, Anthropic Open AI, SpaceX, will go public in the next few months.

**Aravind Srinivas** (3:23)
We actually love Anthropic Open AI, XAI, all these frontier labs. Every time any of their AI gets better, our unified system also gets better, because we route across all of them. We basically think of Perplexity computer as taking the best of all AI and putting it together in one single unified interface and system. So all of you know how much Anthropics models have improved since the beginning of the year.
What has it led to for us? Our revenue actually tripled since the beginning of the year. It's just been five months in the year and our revenue has already tripled.

**Ed Ludlow** (4:02)
To what?

**Aravind Srinivas** (4:04)
We are actually very happy with all these companies' progress and they completely deserve their IPOs, so we are very excited for them.

**Ed Ludlow** (4:12)
Can I follow up? Are you able to discuss what that revenue has jumped to? There were reports in the FT that you are up to about $450 million just in the month of March.

**Aravind Srinivas** (4:21)
Yeah, we crossed that. I think I publicly tweeted that we crossed $500 million about somewhere around mid-April. We are announcing new numbers yet, but we are doing really well.

**Caroline Hyde** (4:32)
Aravind, I've been thinking a lot about where Perplexity sits in the suite of available tools and technologies, right? Research seems to be a really interesting place with Perplexity.
And I'm wondering how you measure the engagement on the platform, so like it's not just like one query and done, but do you kind of track the time that an individual desk or user would stick with one query as sort of indication of success, you know, how the platform is being used, the behaviors of the user base?