The AI landscape doesn't move in one direction — it lurches. Some techniques leap from experiment to table stakes in a single quarter; others stall against regulatory walls, technical ceilings, or organisational inertia that no amount of hype can dislodge. Knowing which is which is the hard part. The State of Play cuts through the noise with a rigorously maintained index of AI techniques across every major business domain — classified by maturity, evidenced by real-world adoption, and updated daily so you always know where you stand relative to the field. Stop guessing. Start knowing.
A daily newsletter distilling the past two weeks of movement in a domain or two — delivered to your inbox while the index updates in the background.
Each dot marks the weighted maturity of practices within a domain — hover for a brief summary, click for more detail
AI generation of complete musical compositions with melody, harmony, arrangement, and production. Includes genre-specific composition and multi-instrument arrangement; distinct from background music which produces functional rather than standalone works.
AI full-composition generation has reached a paradox: the infrastructure works, but almost nobody who makes music professionally wants to use it. Consumer platforms have scaled impressively -- Suno surpassed 100 million users and reached $300M ARR (+404% year-over-year growth), and streaming services receive over 50,000 AI-generated submissions daily. Yet professional adoption for full composition sits below 5%. Carnegie Mellon research found listeners judge AI-assisted compositions as less creative and lower quality than human work, and practitioners show only 20% of AI-generated content immediately usable. Creator resistance is intensifying, not softening: only 20% of professional musicians are regular AI users, and 79% of music creators worry about AI competition. Consumer interest is actively declining -- 44% of U.S. listeners report less interest in AI-generated music versus 24% who report more. The gap is not technical; it is regulatory and cultural. UK government reversed its AI training exemption in March 2026 after artist coalition pressure (McCartney, Bush, Lipa, 220+ signatories). GEMA's landmark lawsuit against Suno (decision expected June 2026) challenges whether AI outputs constitute copyright infringement. Meanwhile, 85% of AI-music streams are flagged as fraudulent, flooding platforms with low-quality slop that degrades user experience and creator revenue pools. Licensing frameworks are stabilizing between major labels and platforms, but pure-AI compositions remain uncopyrightable under U.S. law, and independent creators face uncompensated training risk. This practice remains experimental. The consumer entertainment niche is real, but professional composition -- the domain that would signal true arrival -- is structurally blocked by copyright ambiguity, fraud-driven quality collapse, regulatory backlash, and deepening creator distrust.
The technical stack for AI music composition is production-grade. Suno shaved four months off a product launch by scaling to thousands of GPUs on Modal's infrastructure, and reached $300M ARR with 2M paid subscribers in March 2026. Stable Audio 2.5 offers text-to-audio, audio-to-audio, and inpainting workflows integrated into creative toolchains like ComfyUI. Google's entry with Lyria 3 (February 2026) brought music generation to hundreds of millions of Gemini app users, validating ecosystem expansion beyond startups. Yet this technical maturity has not translated into professional creative adoption. Practitioners report only 20% of AI-generated content immediately usable, with 30% requiring significant editing and 50% discarded entirely.
Professional adoption remains below 5% despite tool availability. Sound on Sound surveyed 1,200 music professionals and found only 20% describe themselves as regular AI users; 33% worry AI undermines creative intent, nearly as many cite ethics concerns, and 25% report insufficient quality for professional contexts. MusicResearch found 78% of professional musicians use AI tools, but only 24% for full-song generation versus 71% for stem-separation (technical utility, not composition). Where producers engage with AI, they reach for narrow technical assists -- stem separation, audio cleanup, restoration -- not full composition. Originality and ethical sourcing remain the primary concerns keeping AI out of the compositional process.
Quality and fraud have become the dominant constraints. IFPI data shows 85% of AI-music streams on Deezer are fraudulent. TIME's investigation documented 50,000 daily uploads to Deezer (34% of all intake), with criminal cases generating millions in fake royalties and platform artist impersonation creating distribution chaos. Carnegie Mellon research confirms listeners still prefer human composition, and independent practitioner case studies show realistic adoption requires significant human curation and artistic vision.
The licensing landscape is stabilizing but remains fragmented. UMG and WMG settled with Suno and Udio in late 2025, introducing tiered royalty structures. However, the UK government reversed its AI training exemption in March 2026 after artist coalition pressure, and GEMA's landmark lawsuit (decision June 12, 2026) may establish that licensing is mandatory globally. Pure-AI compositions remain uncopyrightable under U.S. law, blocking a primary monetization route. Independent artists face uncompensated training risk with no equivalent licensing framework as major-label creators enjoy.
— Deezer reports 75K AI tracks daily (44% of uploads, 650% YoY growth) with 85% fraudulent streams. Apple, Spotify deployed detection achieving 60% fraud reduction. Massive generation scale paired with platform-level quality barriers limiting consumer distribution.
— 33% of Apple Music uploads are AI-generated, yet AI tracks represent <0.5% of listening time—66x supply-demand mismatch. Independent analysis shows AI music underperforms human music by 25-40% on saves and 15-25% on completion rate.
— Independent producer generated 50 full compositions in one month using Suno across multiple genres (Bond, gangsta rap, Bollywood, Afrobeat). Demonstrates capability at scale while documenting technical limitations: AI voices lacked character despite excellence, frequent lyric divergence despite confident assertion.
— 30-year nonprofit (Songs of Love Foundation) deployed Suno for personalized full-composition generation. Custom vocal personas for singers with health challenges; dementia patients hear authentic period music. Expanded from 50K historical songs to production rollout.
— Series D funding at $5B valuation (2.5x growth since November 2025) with 2M paid subscribers and $300M ARR (404% YoY growth). Validates product-market fit despite licensing talks stalled and infrastructure gatekeeping by distributors.
— Major distributor (Believe/TuneCore) and DSP deployed 99%-accurate AI-detection and auto-blocking against Suno tracks. Labeled platforms as 'illegal' and 'pirate studios.' Infrastructure-level gatekeeping cuts off distribution for unlicensed AI music.
— Independent 10-year producer evaluation: v5.5 label-submission ready for mainstream genres but fails niche styles—balanced evidence of capability with material genre-specific constraints.
— Litigation evidence: platforms commercializing at scale but licensing frameworks unresolved; distribution rights remain primary blocker for mainstream professional adoption.
2023-H1: Meta releases MusicGen with SOTA quality, Stability AI open-sources Stable Audio tools, and Boomy reaches 14+ million AI-generated tracks on Spotify, signaling mainstream adoption. Research papers establish affective music generation and emotion-driven composition as research directions. Spotify's crackdown on artificial streaming fraud linked to Boomy highlights platform enforcement challenges. Survey of 1,533 producers shows 37% adoption but 73% fear job displacement and copyright concerns dominate industry discourse.
2023-H2: Stability AI launches commercial Stable Audio product (September) with free and tiered paid plans supporting full composition generation up to 90 seconds; NeurIPS confirms MusicGen's SOTA status in main conference track. Educational integration expands: Berklee College embeds AI composition tools in songwriting curricula with mixed student outcomes. Developer adoption accelerates via tutorials and managed deployment platforms. Research reveals significant training data bias—Global South genres under-represented in composition models. Copyright and artist-likeness protections remain unresolved, constraining professional market adoption despite production-grade tool availability.
2024-Q1: Stable Audio 2.5 emerges with enterprise features and multi-minute generation capability. Large-scale creator survey (Sacem/GEMA, 15,000+ respondents) reveals 35% adoption but 64% doubt benefits and 71% fear income loss. Copyright barriers intensify: major labels litigate against AI model creators; deepfake removals (Drake, Weeknd) signal platform enforcement. Deployment reaches massive scale but primarily through fraud: criminal prosecution of $10M+ streaming scheme using 661,440 AI tracks; Spotify removal of tens of thousands of fraudulent Boomy tracks continues. Research identifies interactive composition gaps and emotional limitations in AI music. Professional adoption remains nascent; early scaling concentrated in entertainment, academic research, and illicit use rather than legitimate composer workflows.
2024-Q2: Stable Audio 2.0 released (April) advancing to 3-minute full-track generation with licensed training data and audio-to-audio editing. Research progresses on compositional control (Instruct-MusicGen editing, attention-head analysis) and identifies systemic limitations: peer-reviewed study documents AI lags humans in emotional impact; Global South genres significantly under-represented in training data. Streaming fraud enforcement tightens (Spotify fees for >90% fraudulent tracks; coordinated DSP policies), constraining illicit scale. Professional adoption barriers remain structural (copyright unresolved, creator skepticism, quality ceilings) rather than technical; technology demonstrates enterprise maturity but deployment concentrated in entertainment, fraud, and academia rather than mainstream composition.
2024-Q3: Suno AI launches iOS app in July; Stability AI releases Stable Audio Open (open weights, trained on licensed CC data). RIAA lawsuit continues; Suno's court filing admits copyrighted training, intensifying legal uncertainty. Warner Music Group pivots to partnership with Suno. Adoption grows to 38% among creators (APRA AMCOS survey of 4,274), but only 5% use consistently and 82% fear revenue loss; Tracklib survey shows 25% producer adoption concentrated in tool-assisted workflows, not pure composition.
2024-Q4: Suno reaches 25M users and $500M valuation with V4 launch; Stability AI maintains commercial Stable Audio product with tiered subscriptions. Analysis of 100,000+ AI-generated songs documents hundreds of thousands of active users and chartable music, confirming mainstream adoption scale. Academic research reveals critical limitations: only 5.7% non-Western genres in training data; professional composer studies cite trust/transparency barriers; CISAC projects $4B+ creator revenue loss by 2028. Copyright litigation continues unresolved; creator anxiety about displacement persists despite commercial traction.
2025-Q1: U.S. Copyright Office rules AI-generated compositions without meaningful human authorship uncopyrightable (January); Suno and Udio face lawsuits and industry ethics consensus strengthens with 400+ organizations demanding licensing and transparency. Market projects $0.57B (28.5% CAGR), yet adoption remains bifurcated: consumer platforms scale to billions of AI streams while professional full-composition generation stays below 5% adoption. CEO backlash and licensing uncertainty signal regulatory/cultural barriers to professional mainstreaming persist despite technical maturity.
2025-Q2: Massive consumer adoption documented: 60 million users globally (IMS), 18% of DSP submissions AI-generated, Luminate reports 33% listener comfort with AI instrumentals and emergence of named AI artists. Stability AI and Arm release Stable Audio Open Small for on-device generation. Professional barriers intensify: UK consumer survey (81.5% want labeling, 78.5% oppose unauthorized use), independent artist lawsuits against Suno, 82% of non-adopter producers cite artistic integrity concerns. Adoption concentrated in tool-assisted workflows (25% of producers) rather than full composition; market remains bifurcated by copyright/licensing gridlock.
2025-Q3: Stable Audio 2.5 achieves three-minute generation in under two seconds and targets enterprise brand audio markets; MIDiA analyst research documents persistent conversion and engagement headwinds despite scale. Spotify enforces voice-clone authorization and removes 75M spam tracks, signaling platform governance maturity. Creator adoption plateaus at 48% (down from 59.5% in 2023) with authenticity concerns and declining participation rates. Major record labels pursue licensing negotiations rather than litigation, while U.S. Copyright Office and regulatory frameworks remain unresolved, cementing regulatory/policy barriers as primary adoption constraint replacing technical capability.
2025-Q4: Stable Audio 2.5 achieves enterprise partnerships with Warner Music Group and Universal Music Group, signaling professional adoption pathway. Market accelerates to $60.44B projected by 2034 (27.8% CAGR) with 50,000+ daily uploads; OpenAI enters space (October 2025) working with Juilliard, validating market while intensifying competition. Professional resistance hardens: 97% of music supervisors demand AI transparency, 49% refuse AI music entirely (December 2025 survey); 87% of producers use AI for technical tasks, only 13% for full-composition generation. Udio settles UMG copyright lawsuit (October 2025) with restricted downloads; Suno faces similar legal risks with 40-50% settlement probability. Copyright barriers, licensing uncertainty, and creator authenticity concerns remain primary adoption constraints for professional composition market.
2026-Jan: Licensing consolidation completes with UMG and WMG settlements; platforms announce tiered royalty structures (human/AI-assisted/fully-AI). Carnegie Mellon research documents AI composition judged less creative and lower quality than human work, affirming persistent creative limitations. Ecosystem integration advances (ComfyUI tooling) validating production maturity. Regulatory evolution: platforms implementing download restrictions in settled licenses; artist resistance hardening (49% refusing AI entirely); copyright barriers persist with pure-AI compositions remaining uncopyrightable. Market projection held at $1.22B by 2029 (28.7% CAGR), but professional full-composition adoption remains below 5%, constrained by policy fragmentation rather than technical capability.
2026-Feb: Deployment infrastructure matures: Suno achieves 4-month product acceleration via Modal's GPU scaling, validating production-grade infrastructure. Adoption friction intensifies: creator concerns escalate (79% worried about AI competition, +5pp vs. 2023); consumer interest declines (44% less interested vs. 24% more interested); professional producers focus on technical task automation rather than full composition. Stability AI advances licensing partnerships (UMG, WMG) normalizing commercial pathways. Practitioner quality assessment remains critical: <1% viability from generation outputs documented. Copyright framework holds: pure-AI compositions remain uncopyrightable, blocking primary monetization pathway. Market dynamics: 50,000+ daily AI submissions, yet adoption barriers (creator trust, consumer discomfort, regulatory uncertainty) increasingly structural rather than technical.
2026-Mar: Commercial scaling accelerates: Suno reaches $300M ARR (+404% YoY) and 2M paid subscribers on 50% quarterly growth; major cloud vendor entry (Google Lyria 3 in Gemini, Feb 2026) validates ecosystem expansion beyond startups. Fraud and quality crisis intensify: TIME investigation documents 50K daily uploads to Deezer (34% of intake), 85% of AI-music streams fraudulent (IFPI); practitioner case studies show only 20% of generated content immediately usable. Professional full-composition adoption remains below 5% despite tool maturity—Sound on Sound survey (1,200 pros) finds only 20% regular AI users; MusicResearch survey (1,525 musicians) shows 24% generate full songs vs. 71% using stem-separation. Regulatory hardening: UK government reverses AI training exemption (March 2026) after artist coalition campaign (McCartney, Bush, Lipa, 220+ signatories); GEMA landmark lawsuit (Munich court) challenges Suno outputs as memorized copies (decision June 12, 2026); both Suno and Udio pursue licensing agreements with major labels (Sony, UMG, Warner, Merlin) signaling shift from litigation to consent-based training. Market remains bifurcated: consumer platforms scale impressively while professional composition blocked by copyright ambiguity, quality constraints, artist resistance, and structural licensing gaps for independent creators.
2026-Apr: Fraud concentration intensified as the dominant systemic risk: IFPI data shows 85% of AI-music streams on Deezer are fraudulent (up from 70%), with Deezer reporting 75,000 AI tracks daily (44% of uploads, 650% YoY growth) and platforms collectively removing tens of millions of AI-generated spam tracks. Professional adoption data from a Water and Music survey of 1,525 musicians confirms the composition gap holds — 78% use AI tools but only 24% for full-song generation, with the majority choosing stem-separation and technical assists over end-to-end composition. Independent producer evaluation of Suno v5.5 found it label-submission ready for mainstream genres but failing niche and experimental styles, illustrating the quality ceiling's genre-specificity. The UK's reversal of its AI training exemption under artist coalition pressure solidified regulatory hardening, and Universal and Sony's move to contest Warner's deal with Suno — demanding court disclosure of distribution rights terms — highlighted that licensing frameworks remain fragmented and contested even where label deals exist. Verification data confirms commercial traction: SIQA dataset (April 15, 2026) documents 1,551 AI music tracks from 828 artists across 57 countries, with Thompsxn Therapy (Gospel) sustaining #1 chart position for 7 consecutive weeks. However, pure AI tracks show 40% higher skip rates and 22% lower Save-to-Spotify ratios. Licensing ecosystem consolidation continues: Kobalt-Udio partnership launches consent-based licensed service with opt-in artist training. Trust barriers intensify: 92% of musicians demand AI training-data transparency; 54% of adopters use AI only for noise reduction, not composition. Professional full-composition adoption remains below 5%, blocked by copyright ambiguity, quality ceiling, creator trust deficit, and fraud-driven market degradation.
2026-May: Commercial maturity and distribution barrier intensification both accelerated. Suno raised Series D at $5 billion valuation (+102% since November 2025) with 2M paid subscribers and $300M ARR (404% YoY growth), validating product-market fit among paying users. Real-world deployment expanded: Songs of Love Foundation (30-year nonprofit) deployed Suno for personalized full-composition generation—custom vocal personas for singers with health challenges and authentic period music for dementia patients (specific outcome: dementia patient recognized emotional connection with 1940s swing composition). Yet infrastructure-level gatekeeping intensified simultaneously: Believe (Paris distributor, parent of TuneCore) and Spotify launched coordinated enforcement with 99%-accurate AI detection, auto-blocking Suno tracks and labeling unlicensed platforms as "illegal pirate studios." Distribution consequence: Facebook groups (135K+ members) reported mass rejections across DistroKid, TuneCore, Believe within days. Adoption metrics reveal structural ceiling: Apple Music disclosed 33% of uploads are AI-generated, yet AI tracks represent <0.5% of total listening time—66x supply-demand mismatch; Deezer maintains 75K AI tracks daily (44% of uploads) with 85% flagged fraudulent; independent analysis shows AI tracks underperform human music by 25-40% on save rate and 15-25% on completion rate. Practitioner case study: independent producer generated 50 complete compositions in one month using Suno across multiple genres, demonstrating capability at volume while documenting technical ceilings—AI singers lacked character and frequently diverged from lyric instructions despite confident assertions. Market bifurcation deepens: commercial scale (billions in funding, 100M+ users, 7M songs daily) concentrated entirely in consumer entertainment and fraud; professional full-composition adoption and legitimate marketplace adoption remain structurally blocked by distribution gatekeeping, copyright unresolved, and listener demand mismatch.