The question about AI art that gets asked is whether it counts. The question that actually matters is whether any of it will hold a price. These are different questions, and confusing them is the reason the AI-art debate has been so unproductive for so long. A work can be good, interesting, even historically important, and still lack the structural properties that allow it to sustain a valuation outside the small ecosystem that made it. That is the real valuation problem, and it has very little to do with aesthetics.
The question underneath the price
Art markets are not primarily taste markets. They are scarcity markets that use taste as a filtration mechanism. A painter with a reputation produces a limited number of canvases over a finite career, and the supply curve is genuinely inelastic at the top. Taste decides which names earn the premium. Scarcity decides whether the premium compounds.
The debate about whether AI images qualify as art has absorbed years of column inches and resolved nothing, largely because it was framed wrong. Refik Anadol’s data sculptures are unmistakably art by any operational definition. Beeple made images before the 2021 auction and is making them after. The work exists. The question is what properties it has as an asset, which is a question about supply, provenance, and reproducibility, not about whether anyone enjoys looking at it.
The Anadol and Beeple lineage
Anadol’s work holds price inside a specific market because of what it actually is, large-scale, site-specific, computationally expensive installations tied to physical infrastructure and institutional placement. The machine-hallucination videos displayed at MoMA are not infinitely reproducible because the performance environment is not infinitely reproducible. The scarcity is real, and it is real in a way the traditional market can parse. That matters.
Beeple is a different and more instructive case. The 2021 Christie’s sale did not price the work. It priced a moment, a particular confluence of pandemic liquidity, crypto-adjacent buyer formation, and a narrative in search of a peg. The tape since has printed that distinction. Secondary market transactions on comparable Beeple work have cleared at a fraction of the 2021 levels when they have cleared at all, and the volume is thin enough that a “market” is almost a generous description. We are not making a qualitative judgment on the work. We are reading the footnote.
What happens when images are nearly free
The hinge argument is the one the field has been most reluctant to confront. Image generation is asymptoting toward zero marginal cost for an increasingly sophisticated output. A high-quality still image, rendered from a well-constructed prompt by a contemporary model, is now a commodity in the economic sense of the word. There is no amount of curatorial framing that reverses that basic fact.
This matters because the art market has historically priced the labor-scarcity of the image object, a painting represents hours of a specific human’s time in a specific studio, along with the intellectual position the work advances. When the image is effectively free to produce, the first half of that equation collapses, and the entire valuation burden falls on the intellectual position, which is a much smaller and more fragile peg to hang a market on.
When the image is free and the edition is infinite, what remains is the argument. Markets have historically refused to price arguments at scale.
Scarcity has to be engineered, and the engineering has to be credible
The working answers to this problem, the ones the market is currently trying to price, involve manufactured scarcity. An edition of twenty-five. A one-of-one. A token bound to a specific physical object. A provenance chain with forensic rigor. Each of these is a real mechanism, and each of them has been tested, with mixed results.
The NFT experiment, in retrospect, mostly demonstrated that on-chain scarcity without corresponding off-chain cultural infrastructure does not hold price. The token tells you there is one. The token does not tell you why anyone should care. When the ecosystem that cared contracted, the price contracted with it, because the scarcity was never anchored in anything other than the ecosystem itself.
The more durable examples have tended to bolt digital scarcity onto traditional art-market infrastructure, gallery representation, institutional placement, serious criticism, and physical or experiential components. Lucian Poe has argued for two years that the AI work that survives into the late decade will be the work that can be hung, walked into, or experienced at a specific coordinate. The thesis is testable and, so far, has held.
Who has an actual thesis
The small number of AI-adjacent artists with a credible valuation case share one property. They use the technology as a tool inside a pre-existing artistic practice that was already producing durable objects and serious institutional relationships. The AI is a medium, not the argument.
- Trevor Paglen’s work on machine vision and surveillance imaging sits inside a photographic practice with deep institutional backing. The AI piece is legible as a continuation, not a rupture.
- Holly Herndon’s project, which operates between music, visual art, and voice modeling, has built an audience and an archive that precedes and will outlast the current model cycle.
- Anna Ridler’s tulip-mania work connects generative imagery to a specific historical argument about speculation and scarcity, which is, if nothing else, aesthetically self-aware in a way most of the field is not.
These are not large markets. None of these artists is printing the kind of numbers that the 2021 NFT cohort briefly did. But the markets that exist have been slow, institutional, and, crucially, not dependent on a single ecosystem of crypto-native buyers. That is the shape of a market that might compound. It is also, admittedly, the shape of a market that might simply remain small forever.
The part the press release won’t tell you
The AI-art narrative the major platforms prefer is a continuity argument. The camera was new once. Video was new once. Digital was new once. Each time, the market adjusted, and the new medium took its place in the canon and the salesroom. Give us a decade and we will be in the same position.
The counterargument is uncomfortable because it is specific. None of those prior mediums had the property of near-zero marginal production cost for a sophisticated output. A camera still required a specific photographer in a specific place. A video still required a production. AI image generation, at the current frontier, requires neither. That is not a continuation. It is a genuinely new problem, and the market has not yet figured out how to solve it at scale.
Easton Cain, in one of the rare public remarks he has made on the subject, put the matter plainly. The medium is not the question. The bottleneck is the question. Traditional art markets price the bottleneck. AI image generation has removed the bottleneck. If it returns, through regulation, through institutional gatekeeping, through some form of credentialed authorship, the market returns with it. Until then, we are watching a category try to find its scarcity.
The cultural assessment
We are in a moment where a generation of artists will produce genuinely interesting, possibly important, work using AI tools. That is a cultural fact, not a market fact, and it deserves to be taken seriously on its own terms. The mistake is to assume that because the work matters, the work must hold price. Those are different claims, and the field’s reluctance to separate them is the reason so much of the last five years of commentary has aged so poorly.
The forward view is that AI-adjacent work with physical or experiential anchoring will sustain a modest institutional market into the late decade. AI-native work that lives purely as digital image will not, absent a credible new scarcity mechanism we have not yet seen. The question is not whether these artists are good. Many of them are. The question is whether their work has been engineered to hold a price when the image itself no longer can.
Nothing in this article is investment advice. CreativeSlop is an independent publication. Figures rounded for readability. Names of market participants referenced in good faith from on-the-record and on-background conversations.