Deezer rolled out an AI-generated music feature in its 2026 beta, challenging Spotify and Apple Music’s dominance. The tool leverages a transformer-based LLM with 12 billion parameters, enabling real-time track composition. Deezer’s blog states the system uses end-to-end encryption and integrates with third-party DAWs.
Deezer’s AI-Generated Music Feature Rolls Out in 2026 Beta
Deezer’s new feature, codenamed Mac4Ever, allows users to generate original tracks using a neural processing unit (NPU)-optimized AI model. The system analyzes user preferences through a proprietary recommendation engine, then synthesizes audio using a WaveNet-derived architecture. According to TechCrunch, the beta version supports 48kHz/24-bit output with 128-bit AES encryption.
The tool’s core engine, CompositionNet v3.2, employs a hybrid approach combining autoregressive and diffusion models. Developers at Deezer’s GitHub repository note the system can generate 3-minute tracks in under 8 seconds on devices with ARMv9 architecture. This contrasts with Spotify’s AI-generated playlists, which rely on pre-trained models without real-time composition.
What This Means for Enterprise IT
Enterprise users gain access to Deezer’s AI Studio API, which allows embedding music generation into workflows. The API charges $0.02 per 100ms of generated audio, according to Sajari’s pricing analysis. This creates a potential conflict with Apple’s App Store guidelines, which prohibit “unlimited” in-app audio generation without explicit approval.
Cybersecurity firm CrowdStrike warns that the NPU’s direct memory access (DMA) capabilities could introduce vulnerabilities if not properly sandboxed. “While the encryption is robust, the lack of public documentation on secure enclave implementation raises concerns,” said CTO Dr. Elena Voss in a
statement
.
Technical Breakdown of the AI Engine
Deezer’s system uses a custom Transformer-XL architecture with 12 billion parameters, trained on 150 million tracks from its 70-million-song catalog. The model’s attention mechanisms prioritize tempo and key consistency, according to a 2023 IEEE paper. Benchmark tests show the system achieves 92% accuracy in matching user-specified genres, outperforming Spotify’s SoundSearch by 17%.
The AI engine’s training data includes anonymized user behavior metrics, raising privacy concerns. Boing Boing reports that Deezer’s privacy policy now includes a “creative data sharing” clause, allowing third-party developers to access anonymized composition patterns.
The 30-Second Verdict
Deezer’s move threatens Spotify’s 2025-2026 growth projections, according to PitchBook’s analysis. The AI feature could disrupt music production workflows, particularly for indie creators relying on low-cost tools. However, the lack of cross-platform compatibility with Apple’s Core Audio may limit its adoption.
Ecosystem Implications and Developer Reactions
The Mac4Ever feature’s reliance on Deezer’s proprietary AudioCore 2.0 framework creates potential lock-in effects. Open-source developer Marcus Lee, founder of FreeMusicAPI, criticized the move: “By centralizing AI music generation, Deezer risks fragmenting the ecosystem. Developers should have access to open standards, not vendor-specific tools.”
Apple has not commented publicly, but internal documents obtained by The Verge show the company is accelerating its own AI music project, MusicGen Pro. The system uses a Diffusion Transformer architecture with 24 billion parameters, according to a leaked roadmap.
Spotify’s response remains unclear, but its recent acquisition of Studio AI suggests it’s preparing countermeasures. The startup’s TrackForge platform uses a Bayesian Neural Network to generate music, according to TechCrunch’s review.
Comparative Analysis: Deezer vs. Spotify vs. Apple Music
| Feature |
|---|
