Spotify builds AI-driven personalised audio platform

ET reports that at its 2026 Investor Day in New York, Spotify outlined a roadmap to turn audio into a generative, personalised experience. The company presented features including AI playlists, conversational listening, remix tools, personalised podcasts and interactive audio experiences, ET reports. ET quotes co-CEO Gustav Soderstrom saying, "Spotify's evolution has followed a clear path: first access, then personalisation, now generation." ET reports Spotify is developing a Large Taste Model trained on 3.4 trillion daily signals from users across music, podcasts and audiobooks. ET cites CNBC reporting that co-CEO Alex Norstrom said investments in personalisation and AI are contributing to higher user engagement. ET frames Spotify's advantage as two decades of listener behaviour and cultural data rather than competing directly with frontier AI firms.
What happened
ET reports that at its 2026 Investor Day in New York, Spotify outlined plans to expand beyond recommendation engines into generative and interactive audio. ET reports the company described features including AI playlists, conversational listening, remix tools, personalised podcasts and interactive audio experiences. ET quotes co-CEO Gustav Soderstrom saying, "Spotify's evolution has followed a clear path: first access, then personalisation, now generation." ET reports that Spotify presented a Large Taste Model trained on 3.4 trillion daily signals drawn from user activity across music, podcasts and audiobooks. ET also cites CNBC reporting co-CEO Alex Norstrom saying investments in personalisation and AI are increasing user engagement.
Technical details
ET reports the centerpiece is the Large Taste Model, which the article says is trained on 3.4 trillion daily signals; ET frames this dataset as coming from two decades of listener behaviour across Spotify's catalogue. ET describes the product slate at Investor Day as combining generative elements (remixes, personalised podcasts) with conversational interfaces and contextual personalisation.
Editorial analysis - technical context
Companies with large, longitudinal behavioural datasets can build recommender and generative systems that personalise at multiple timescales; this pattern appears across streaming and ad-driven platforms. For practitioners, building a cross-modal model that ingests music, podcast, and audiobook signals raises engineering tradeoffs: label sparsity for niche content, sequence modeling at long time horizons, and feature pipelines that merge content metadata with implicit signals.
Context and significance
Industry observers have noted that when core content becomes commoditised across providers, AI-driven personalization and unique content generation are common differentiation strategies. ET frames Spotify's emphasis on its historical listener data as a potential moat relative to generalist frontier AI firms such as OpenAI and Google, which ET mentions as points of comparison. ET also cites CNBC reporting management linking AI investments to subscriber engagement gains.
What to watch
For practitioners: monitor technical releases around the Large Taste Model (architectural details, training regime, privacy controls), API or developer-facing features for remixing or conversational listening, and any disclosures about data usage or consent frameworks. Observers should also watch product rollout timelines and third-party partnerships that enable generative audio at scale.
Scoring Rationale
Notable product strategy shift with a large proprietary dataset and generative features that matter to ML practitioners building personalization and audio-generation systems. It is not a frontier-model release, but it signals meaningful engineering and privacy work for platform-scale audio.
Practice with real Streaming & Media data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all Streaming & Media problems

