By Stuart Kerr, Technology Correspondent, LiveAIWire
Christie’s closed its dedicated digital art department in September 2025, citing a slump in NFT trading. Six months earlier, the same auction house had run its first sale devoted entirely to artificial intelligence, titled Augmented Intelligence, taking in 728,784 dollars at an 82 percent sell-through rate. The two events are not the contradiction they look like. Christie’s is not retreating from AI art, it is folding AI art into its mainstream sales rather than treating it as a curiosity, and that shift is the clearest sign yet of a market moving from the margins toward the centre without fully replacing what was there before.
The scale behind that shift is real. The AI-generated art market was valued at 3.2 billion dollars in 2024 and is projected to reach 40.4 billion dollars by 2033, a compound annual growth rate of 28.9 percent, according to Grand View Research. In June 2026, DATALAND, billed as the first museum built entirely around AI art, opened in Los Angeles, co-founded by artist Refik Anadol, whose piece Unsupervised was acquired by MoMA in October 2023 in what the art press treated as AI art’s institutional coming-of-age moment.
The market is growing and it is deeply contested at every level. Only 9 percent of gallery professionals consider AI-generated art a legitimate new medium, according to Artsy’s February 2026 survey of more than 300 gallery respondents. Twenty-five percent describe it as a destabilising force for authorship and value, and 41 percent say AI art rarely comes up in conversation with collectors at all. Nearly 60 AI-related copyright lawsuits were tracked as of late October 2025. The market is growing faster than the legal and institutional frameworks that would settle whether it deserves the prices it is fetching.
Table of Contents
Who Is Actually Buying AI Art and Why
Museum acquisitions function as the market’s legitimacy signal. MoMA’s acquisition of Anadol’s Unsupervised, the Getty Museum’s first AI-generated photograph in January 2025, and the Centre Pompidou’s 2023 purchase of 18 blockchain-linked artworks each represent a curatorial authority deciding that AI-generated work merits preservation alongside human-made art, a decision collectors watch closely when deciding what to buy.
At the collector level, the Artsy survey found that 15 percent of galleries see curiosity-driven interest, collectors who ask questions about AI art without buying it, while 16 percent report collectors who actively avoid AI-assisted work altogether. The segment with genuine demand is the speculative top end, high-value auction lots from established names like Anadol and autonomous generative projects like the collective Botto, where buyers are betting on the appreciation of an asset class they expect to keep growing.
Christie’s own registration data from Augmented Intelligence shows who these new buyers are: 48 percent of registered bidders belonged to the millennial and Gen Z generations, a far younger profile than the auction house’s typical client base. That demographic shift matters because younger bidders are more likely to have come from crypto and NFT communities already comfortable with digital ownership, and less likely to be anchored to the connoisseurship norms that have historically driven long-term price stability in the traditional art market.
Two distinct collector logics are emerging from that split. Institutionalised artists, meaning those with museum acquisitions and major gallery representation behind them, carry a risk profile closer to an emerging artist in the traditional market, backed by outside validation. Purely autonomous or generative projects carry no such backing. Buying one is a speculative bet on the redefinition of authorship itself, with a correspondingly higher ceiling and a much higher chance of going to zero.
What This Means for You
If you are considering buying AI art, the two collector logics above are the first filter to apply to any purchase. An institutionalised name gives you comparables, a price history, and a curatorial record to lean on if you ever need to sell. An autonomous or purely generative work gives you none of that. You are pricing in the possibility that the entire category is later judged not to have been art in any sense collectors continue to value, and that risk does not show up on the auction house estimate.
The resale market compounds that risk. Auction houses have built price discovery for the handful of AI works that cross six figures, but the equivalent of the ordinary commercial gallery market, the mid-range resale infrastructure that lets a collector exit a position without a headline sale, barely exists yet for AI art. Anyone buying at current prices is taking on illiquidity risk on top of the underlying uncertainty about whether the work will still be considered valuable in a decade.
The Copyright Question Nobody Has Resolved
The legal foundation under the AI art market remains unsettled. Nearly 60 AI-related copyright suits, tracked by law firm BakerHostetler’s litigation tracker, were pending as of late October 2025. The Andersen v. Stability AI trial is scheduled for September 2026. Disney and Universal are suing Midjourney for 150,000 dollars per alleged infringement, with more than 150 allegedly infringed works named in the complaint, meaning total damages could exceed 20 million dollars if the studios win.
The EU AI Act, fully applicable from August 2026, will require providers of general-purpose AI models to publish a detailed public summary of the copyrighted material used in training, a disclosure obligation with real consequences for tools trained on scraped images without artist consent. No equivalent disclosure requirement exists yet in the United States, which means American courts are being asked to settle the underlying fairness question before regulators have defined what transparency should look like.
A March 2026 class action by journalist Julia Angwin against Superhuman Platform, the owner of Grammarly, alleging misuse of writers’ names and identities to build its Expert Review feature, extended the same underlying question from image generators to language tools trained on professional writing. In every case the courts face the same question: does training an AI system on human creative work without permission or payment, then using it to generate output that competes commercially with the people whose work trained it, constitute infringement under existing law? The answers arriving from 2026 onward will decide whether the AI art market rests on solid legal ground.
What It Means for Human Artists
A Stanford Graduate School of Business study published in 2025, examining more than 3.2 million images and 62,000 artists on a major image marketplace, found that after the platform began allowing AI-generated images, monthly image volume rose 78 percent and the number of active sellers rose 88 percent, while sales by non-AI artists fell 23 percent. The effect was not evenly distributed. Artists with distinctive personal styles and established client relationships were far less affected than those whose income depended on high-volume production of images in popular, easily replicated styles.
Roughly four in five artists surveyed say AI-generated work lacks the emotional depth of human-made art, according to a 2025 Scientific American report on attitudes toward AI-driven art. That belief may well be accurate, and it is still not, by itself, protection against a market where consumer testing has repeatedly found that ordinary viewers, in some studies just over half, cannot reliably tell AI-generated abstract art from human-made work on sight. Artists whose output is hardest to distinguish from AI generation on price and visual effect alone are the most exposed. Artists whose work carries a legible, personal point of view are the least.
The Legitimacy Question That Remains Open
Whether AI art carries the expressive weight of human-made art is not a question that sales figures can answer. That AI-generated work now sells for hundreds of thousands of dollars at major auction houses tells us about the speculative dynamics of the current market. It does not settle whether algorithmic output, however sophisticated visually, constitutes creative expression in the sense that work produced by a human consciousness engaging with lived experience does.
DATALAND’s opening is a bet that it does, that AI art is a genuine new category deserving its own institutional infrastructure. The Artsy survey’s finding that only 9 percent of gallery professionals agree represents the establishment’s scepticism. The market’s growth trajectory represents the buying public’s increasingly permissive stance. For readers following AI’s effect on creative industries, LiveAIWire’s coverage of how generative AI learned to tell stories and our analysis of AI-driven content creation in media traces the same legitimacy question through literature and journalism.
Why Long-Term Value Is Still a Gamble
The long-term value of AI art as a collectible depends on questions nobody can currently answer with confidence. Researchers including Ilia Shumailov documented, in a 2024 Nature paper, that AI models trained repeatedly on their own output degrade in a pattern known as model collapse. If that degradation shows up measurably in AI art generation by 2028 to 2030, works made during the current period of peak capability could gain a scarcity value similar to a first edition, while later, technically superior work could paradoxically depress the value of what came before it by making it look primitive.
Christie’s own bidder data adds a further wrinkle: at the Augmented Intelligence sale, 37 percent of registered bidders had never bid at the house before in any category. A meaningful share of the AI art market is being built by buyers new to art collecting altogether, drawn by novelty rather than by the connoisseurship that usually anchors long-term value in the traditional art market, and that novelty-driven demand is exactly the kind that historically proves least durable when a category stops being new.
What Artists Can Do
The evidence points toward differentiation as the more durable strategy than competing with AI generation on price or volume. Artists with a distinctive voice rooted in specific personal or cultural experience are less exposed to AI substitution than those whose work mainly demonstrates technical proficiency in a replicable style. The Stanford research bears this out directly: overall platform sales rose while lower-selling, less distinctive artists absorbed the loss, and established artists with a recognisable point of view were largely unaffected. Technical mastery still matters. It is no longer sufficient by itself, because AI can now produce technically accomplished work in a matter of seconds.
The artists and writers most actively contesting these questions in court, including Julia Angwin and the plaintiffs in Andersen v. Stability AI, are asserting rights the legal system has not yet confirmed they hold. For related coverage of that fight, see LiveAIWire’s reporting on the legal battle over AI training data and our analysis of artists using AI as a tool of creative activism. How those cases resolve will matter well beyond the art market, to every creative professional whose work has already been used to train a model without being asked.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.