90,000 AI Tracks Per Day: The Music Industry’s Liquidity Fragmentation Problem

Last week, Deezer dropped a number that should terrify anyone who still believes content scarcity is the foundation of creator economics: 90,000 AI-generated tracks are being uploaded to its platform every day. That’s not innovation. That’s a denial-of-service attack on the concept of provenance. I’ve spent years auditing smart contracts where the exploit wasn’t a code bug—it was a governance vacuum. The same logic applies here. The exploit isn’t the AI model. It’s the absence of any functional framework to distinguish signal from noise. Deezer, a French streaming service with roughly 10 million subscribers, is not Spotify. But its data point is a canary in the coal mine of digital content ownership. Ninety thousand per day. That’s 63,000 tracks per week. Over 32 million per year. And this is just one platform. Multiply across Apple Music, Spotify, and you’re looking at an industrial-scale flood of synthetic audio. The context: AI music generation tools like Suno, Udio, and open-source AudioCraft have reached commodity price and latency. A user with zero musical training can generate a passable lo-fi beat in 10 seconds. The barrier to entry for “music creation” has collapsed to zero. What follows is an avalanche of supply that overwhelms any human curation mechanism. Crypto Briefing ran this story—a blockchain news outlet—and the subtext is obvious: “You need Web3 to solve this.” Smart contracts for automatic royalty splits. NFTs for immutable provenance. Decentralized storage for permanent attribution. It’s a neat pitch. But as someone who has audited over 50 projects claiming to “fix” digital ownership, I can tell you: most of these solutions are simply shifting the problem to a different layer of abstraction. Let me perform a forensic autopsy on this 90,000 figure. First, the detection problem. Deezer presumably uses some combination of spectral analysis, metadata heuristics, and user reporting to flag AI content. But every detection model is a cat-and-mouse game. The exploit isn’t in the track—it’s in the gap between model release and detection update. I’ve seen the same pattern in DeFi: when a new vulnerability is discovered, the window between exploit and fix is where all the damage happens. With AI music, that window is shrinking because the generators evolve faster than the detectors. Second, the economic distortion. Streaming platforms distribute royalties based on pro-rata shares of total play time. If AI-generated tracks soak up 10% of total plays on a platform, that’s 10% of the royalty pool diverted away from human creators. That’s not a fair market—that’s a liquidity fragmentation problem. In DeFi, liquidity fragmentation across dozens of Layer-2s creates inefficiency. Here, “attention liquidity” is being sliced into millions of synthetic grains, each claiming a micro-share of the payout. The result: the same poverty of meaningful value, just with a higher transaction count. Third, the legal vacuum. Copyright law is territorial, slow, and built on the assumption of identifiable human authors. AI-generated works exist in a gray zone. The US Copyright Office has refused registration for AI-only works. The EU AI Act requires transparency but not retroactive cleansing. So these 90,000 tracks per day are entering a system with no law, no jurisdiction, and no mechanism for appeal. In crypto terms, it’s a permissionless network without slashing conditions. Now the contrarian angle. The bulls on blockchain-based music solutions will say: “We can timestamp every track on-chain, register the creator’s key, and enforce smart contract royalties. The problem is solved.” They’re half right. The tech stack exists. But standardization fails when it ignores human chaos. Real-world adoption requires all major platforms to agree on a single protocol, all artists to register their keys before uploading, and all detectors to trust a global registry. That’s not a technical problem—it’s a coordination nightmare. I have audited projects that tried to create “universal music metadata” layers. They end up with governance token splits, fork wars, and liquidity pools that drain faster than a flash loan attack. Moreover, the volume itself becomes a triage issue. If 90,000 tracks arrive daily even in a scenario where 100% are registered on-chain, who verifies the verifier? Gas costs on Ethereum would make registration prohibitive. L2s reduce fees but add finality delays. And once a track is on-chain, removing it for copyright infringement violates immutability. Code is law until someone wants an exception. You didn’t think the easy answer was just “put it on a blockchain,” did you? Liquidity is a mirror, not a vault. The mirror reflects the underlying incentives. If the incentive is to flood the platform with cheap synthetic content, the mirror will show a fractured, unplayable mess. No smart contract can fix a broken incentive. What I find most revealing is the silence from the major labels. Universal, Sony, Warner—they’re all running their own AI experiments, but none have publicly called for a global registration standard. Why? Because they want to sue first, then license. They want to control the legal precedent before the technology settles. That’s the same playbook they used against Napster and Spotify: litigate, then pivot to partnership. But the speed is different this time. In the 2000s, a 10-year legal cycle gave time to adapt. Now, 90,000 tracks per day means the data pile grows by more than the total number of songs released in 1990 every single week. The blockchain remembers, but the auditors forget. We keep building technical audits as if the threat is code bugs. The real threat is volume—unsupervised, unlabeled, unremovable. From my experience auditing Layer-2 bridges, I learned one thing: finality is a social construct, not a technical one. A bridge may confirm a transaction on the source chain, but until the destination chain accepts it, the user is in limbo. For AI music, the social finality is missing. We don’t yet collectively agree what “owning” an AI-generated track means. Until we do, every technical solution is like a smart contract that calls an oracle nobody trusts. So what’s the takeaway? If you are a project building a “blockchain music provenance” solution, ask yourself: who will use it? Will Deezer and Spotify integrate your protocol? Will they enforce royalties that you define? If the answer is “we hope they will,” you are building on hope, not code. Hope is not a security parameter. I predict we will see two camp emerge. The first camp will build private permissioned chains where labels control ingestion and only pre-vetted content gets on-chain—essentially a database with a consensus algorithm. The second camp will build open platforms that treat every AI track as a valid asset, accepting the noise in exchange for permissionless innovation. Neither solves the problem. The first centralizes censorship power; the second turns music into spam. In code, silence is the loudest vulnerability. The quiet is coming from the industry’s lack of urgency. Last week, Deezer shouted. But the rest of the ecosystem is still debugging its own incentive structure. Trust nothing. Verify everything. And if you’re building a solution, start with the question: who suffers if this fails? If the answer is “the creators,” you have moral responsibility. If the answer is “no one,” you’re building a toy. I’ve seen too many toys dressed as infrastructure. The exploit wasn’t in the code. It was in the assumption that technology alone can govern human expression. Stop assuming.