# Ek sees compute as one signal of AI capability Markdown copy of https://realalpha.app/s/ek-sees-compute-as-one-signal-of-ai-capability/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy. - Speaker: [Daniel Ek](https://realalpha.app/people/daniel-ek/) · Daniel Ek is a Spotify cofounder and executive chairman; Neko Health cofounder. · Co-founded Spotify in 2006 and now guides its long-term strategy as executive chairman. - Host: [Chamath Palihapitiya](https://realalpha.app/people/chamath-palihapitiya/) · Founder & CEO, Social Capital · Led Facebook's growth team; his firm backed Slack and AI chipmaker Groq. - Host: [Jason Calacanis](https://realalpha.app/people/jason-calacanis/) · Founder & CEO, LAUNCH · Early Uber angel; sold Weblogs Inc. to AOL; hosts This Week in Startups. - Host: [David Sacks](https://realalpha.app/people/david-sacks/) · Co-founder & partner, Craft Ventures · PayPal's first COO, founded Yammer; White House AI & crypto czar 2025–26. - Host: [David Friedberg](https://realalpha.app/people/david-friedberg/) · CEO, Ohalo · Founded The Climate Corporation, sold to Monsanto for about $1.1B in 2013. - Show: All-In - Episode: [Daniel Ek: Life After Spotify, Broken Healthcare Incentives, Catching Disease Early & AI's Potential](https://realalpha.app/v/JEUboZzZGM4/), Sep 28, 2026 - Moment: 39:24 to 42:42 of the episode (3 min), playing on https://realalpha.app/s/ek-sees-compute-as-one-signal-of-ai-capability/; the episode from there: https://realalpha.app/v/JEUboZzZGM4/?t=2364 ## Quote > “if I'm using a **hundred thousand GPUs** for something, that's probably going to be a lot more powerful than if I'm running an open source model on my home PC.” > — [Daniel Ek](https://realalpha.app/people/daniel-ek/) at [41:56](https://realalpha.app/v/JEUboZzZGM4/?t=2516) · All-In ## Takeaways 1. Ek expects open and closed models to coexist. 2. Spotify uses frontier models alongside fine-tuned internal models. 3. He sees compute scale as one signal of potential capability. ## About this segment Daniel Ek says Spotify uses frontier and fine-tuned models, and argues open and closed approaches can coexist. He also suggests compute scale may indicate how much capability a model can exert. From “Daniel Ek: Life After Spotify, Broken Healthcare Incentives, Catching Disease Early & AI's Potential”, Sep 28, 2026. [Watch the full episode](https://realalpha.app/v/JEUboZzZGM4/) ## Transcript Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there. [39:26](https://realalpha.app/v/JEUboZzZGM4/?t=2366) And in that sense, do you also think about the benefit and the value of open source AI, open weight models, where I think you can now look at like a $0.13 per million token output cost versus $30. dollars that's an incredible difference that unleashes this technology across a broad spectrum of businesses and individuals that there is no you know concentration of value or wealth amongst a handful of businesses or individuals that if we can really make these open weights and open source models proliferate. But obviously, this regulatory conversation may restrict the expansion of open source and open weight models. And if you have a view on kind of the importance there. I think technology has always gone between open and closed, right? Like we had Windows versus Linux in the first iteration, you know, iOS versus [40:24](https://realalpha.app/v/JEUboZzZGM4/?t=2424) Android, like we've seen this play out so many times. And where we tend to net out is we tend to have both. And, you know, to the extent that I have a view about it, I've very positively advocated for open source models before, and I will keep on doing that because I think that that brings a lot of innovation that will be really helpful for the ecosystem as well. So I completely agree with you, Dave, that I think that, you know, if you look at these now and certainly what we're seeing um you know at let's take spotify as an example where we're obviously using a lot of the frontier models but we we have a bunch of fine-tuned um models ourselves too um and we tend to do both and that is actually making us more innovative and that's actually allowing us to do certain things that we couldn't do uh with the frontier models uh either because of cost and [41:19](https://realalpha.app/v/JEUboZzZGM4/?t=2479) efficiency reasons or sometimes because we just couldn't tune them the way we will want it. And so I personally think it's like, it's going to be both. But one of the things I heard that is a perspective I haven't heard before that maybe I would encourage people, not that I am sort of an expert on all these things, but I'm surprised that people haven't talked about sort of the amount of compute. So if you think about it is for me, this isn't about just the intelligence itself of a single model, but it's still, you know, of the basis of, you know, the amount of computes you're doing can indicate something, right? Like, so if I'm using a hundred thousand GPUs for something, that's probably going to be a lot more powerful than if I'm running an open source model on my home PC. It's probably very unlikely that I'm going to be able to do an enormous cyber thing. [42:16](https://realalpha.app/v/JEUboZzZGM4/?t=2536) Certainly, if the companies have mythos class models to try to, you know, protect themselves, that I should be able to, with an open source model on my home PC, do a lot of damage. So, you know, I just find it to be a fascinating thing to just think about if there's a way of looking at the amount of compute as well, and as one sort of factor in that.