$220 million. That's how much the NFT market cap jumped in the first week of 2026, according to CoinGecko data tracked across hundreds of collections. Nobody I know called that. Everyone I know had already written the obituary sometime around 2023, filed NFTs next to Beanie Babies and Google Glass in the drawer marked "things we were embarrassed about." And yet here we are: gaming NFTs now account for 38% of transaction volume, real-world-asset tokenization has crossed $26 billion on-chain, and the market — depending on which analyst you trust — sits somewhere between $18 billion and $65 billion. Wide range. Nobody actually knows. That's usually the tell that something structural is happening underneath the noise.
I've sat through enough digital transformation steering committees to recognize the pattern. A technology gets pronounced dead by people who were never its real audience. Then it resurfaces wearing a suit, doing something boring and load-bearing, and nobody wants to admit it's the same thing. NFTs never stopped being what they always were: a token standard for proving who owns what, at what time, verifiably, on a ledger nobody controls. The JPEGs were the marketing. The token was the point.
The infrastructure quietly outlived the hype
Here's the bottom line: while retail was mourning Bored Apes, the plumbing kept getting built. Identity NFTs hit 12 million issued this year, used for decentralized IDs, employee credentials, event access. BlackRock and Franklin Templeton have tokenized over a billion dollars in government bonds. Ethereum still handles 62% of NFT contracts, Solana another 18%, and Layer-2 fee compression finally made minting something other than a $40 gas tax on curiosity. None of this trended on anyone's timeline. It didn't need to. Utility rarely does. What strikes me, coming out of PLM and digital-twin work, is how familiar this feels. Every enterprise asset-tracking system I've touched eventually wants the same three properties: unique identifier, immutable provenance, transferable ownership. That's a non-fungible token whether you call it that or not. The industry spent five years pretending NFTs were a culture, when they were actually a data structure looking for the right workload.
Agents don't care about status. They care about settlement
Now the part that should actually keep you up at night. AI agents are moving from "responds when prompted" to "runs continuously, transacts autonomously," and that shift is rewriting tokenomics from the ground up, according to recent industry analysis on agentic AI economics. Continuous inference needs continuous settlement. Settlement needs an identity layer that isn't a human clicking "approve" every time. It needs assets an agent can hold, price, trade, and prove it owns — instantly, machine-to-machine, no wallet-connect popup, no human in the loop at all.
That's an NFT. Not a PFP. A programmable, ownable, transferable unit an autonomous system can generate and trade with another autonomous system, faster than any exchange built for people.
Picture the actual mechanics, because I think people jump straight to sci-fi when the plumbing is almost boringly familiar. An agent generates a design, a melody, a synthetic dataset, a piece of code. It mints the output as a token. Another agent — running a different model, owned by a different principal — evaluates it against its own objective function and bids. Settlement happens on-chain in milliseconds, denominated in whatever stable unit the agent economy standardizes on. A human, somewhere upstream, owns the wallet that both agents ultimately answer to. Royalties flow up. The agents never see a dashboard. They don't need one.
I've watched three separate teams try to build "agent marketplaces" this year with REST APIs and centralized ledgers duct-taped together, and every single one hit the same wall: no shared, trustless way to prove an agent-generated asset was actually created by that agent, at that time, and hasn't been duplicated somewhere else. That's precisely the problem NFTs were built to solve in 2021. Nobody wanted to hear it because the word had been ruined by cartoon apes.
Where this goes dystopian, and where it doesn't
The uncomfortable version: agents mint content faster than any human court system can adjudicate provenance disputes, flooding tokenized marketplaces with synthetic "creative" output that trades among bots in loops designed to launder value back to whoever owns the highest-throughput agent fleet. Wash trading, except the traders are language models with no fatigue and no regulatory fear, running 24/7. Regulators are already behind on human-driven crypto; agent-to-agent volume at machine speed makes MiCA and DAC8-style frameworks look like they're policing a horse-and-buggy economy.
The less dystopian version, and the one I'd actually bet on: this becomes infrastructure for licensing, not culture. Enterprise agents trading tokenized training data, synthetic media rights, or compute-credit derivatives among themselves, with humans collecting royalties passively the way a songwriter collects mechanicals without touching every stream. Less "digital art renaissance," more "automated B2B licensing clearinghouse that happens to run on the same rails NFTs pioneered." Unsexy. Profitable. Exactly the kind of thing that survives past the hype cycle, the way gaming NFTs quietly became the market's largest category while nobody was tweeting about it.
Either way, the framing "NFTs are back because of art" misses what's actually happening. NFTs are back because tokenized, verifiable, machine-readable ownership is the only substrate that scales to an economy where the buyers, sellers, and appraisers might increasingly be software. The apes were never the point. They were just the first thing dumb enough, and human enough, to make everyone notice the token underneath.
What I'm watching for next: whether the first genuinely large agent-to-agent NFT marketplace launches as a standalone protocol, or quietly bolts onto an existing chain nobody currently associates with AI. My money's on the latter. Infrastructure rarely announces itself before it's already load-bearing.
Share: