DATALAND: When Architecture Learned to Dream

I. From a Projection to a Building

Eight years ago, Refik Anadol turned an entire building facade into skin that could dream. From September 28 to October 6, 2018, to mark the LA Philharmonic’s centennial, he and his studio took more than a century of the orchestra’s digitized archive — compressed into roughly 45 terabytes of recordings, footage, and archival material — reprocessed it through machine-learning algorithms, and fired the result, recalculated live each night, through 42 large-scale projectors onto the stainless-steel skin of Frank Gehry’s Walt Disney Concert Hall, calling it WDCH Dreams. Gehry himself, watching the projection, said the building now looked like it had grown a “living skin.”

A lot happened in those eight years. The seed Blade Runner planted in Anadol at age eight, watching it for the first time in Istanbul, eventually carried him to Los Angeles. He first arrived in 2012 to pursue a graduate degree in Design Media Arts at UCLA, where he has since taught for a decade; Efsun Erkılıç moved to Los Angeles in 2014, and the two co-founded the studio there. On June 20, 2026, directly across the street from the Concert Hall, inside another building by the same architect, they turned that line into an entire museum. DATALAND sits inside The Grand LA, Gehry’s roughly billion-dollar downtown development built by Related Companies — sharing a cultural corridor with MOCA, The Broad, and the Concert Hall itself — the world’s first museum built specifically for AI art. It’s a “full-circle moment” for Anadol and Erkılıç’s careers: what they inherited was a concrete shell Gehry had left completely empty for them — Anadol calls it “an open canvas for free imagination.” Erkılıç, in conversation with Hans Ulrich Obrist, recalled that the two of them had originally been searching for an existing space to realize the technology they were developing, until, at some point, the question changed: “What would happen if we designed the space we were looking for ourselves?” This time, Anadol wasn’t projecting onto an exterior wall — he turned the building’s entire interior into the projection itself. “Rather than concrete, glass, and steel, we wanted to shape this environment through light, data, sound, and human presence.”

II. How Architecture Gets “Kissed”

That shift can be read directly against architectural theorist Sylvia Lavin’s Kissing Architecture. Lavin uses “kissing” to describe the increasingly intimate relationship between building surfaces and new forms of media imagery: a projected image meets architecture’s skin without swallowing it — you can still see the wall through the image. She calls this layering “superarchitecture,” citing artists like Pipilotti Rist and Doug Aitken, who “kiss” projected images onto the surfaces of existing buildings, to name what’s ambiguous about the relationship: image and architecture stay in a state of “contact without possession.”

DATALAND does something else entirely. It doesn’t kiss an image onto a building’s surface — it builds the building’s entire interior volume as the container in which the image is continuously generated and never repeats. There’s no gap left through which you could “see the wall itself,” because the wall itself has been designed to carry nothing but this system. What actually matters in Lavin’s argument isn’t how clever the “kissing” metaphor is — it’s how she uses it to reopen an older, more fundamental question: what is architecture’s own medium, exactly. Once an image can be directly layered onto a building’s skin and swapped out at will, architecture’s medium specificity is no longer determined solely by brick, glass, and structure — it becomes a condition that light and data can temporarily take over and hand back at any time. In 2011, that argument mostly described temporary, exterior media events happening to a building. What DATALAND forces is a further question: once that “reskinnable” condition moves from a building’s exterior to its entire interior volume, and becomes permanent rather than occasional, does the question “what is architecture’s medium” simply get replaced by a blunter one — is this still architecture at all.

That reads especially sharply against Gehry’s own body of work. Gehry is Gehry because of an unmistakably personal, sculptural formal language — the twisting titanium skin of the Bilbao Guggenheim, the rippling facade of the Concert Hall itself, both direct evidence of “the architect’s hand.” But across both collaborations — WDCH Dreams and DATALAND — what Gehry handed Anadol was, each time, the exact moment his own signature gets deliberately muted and pushed into the background. The Concert Hall’s titanium skin, during the projection, became nothing more than a curved substrate for carrying someone else’s image. DATALAND’s “completely emptied concrete shell” is described outright as a “blank canvas” — precisely the neutral, self-effacing kind of space Gehry is least known for. An architect famous the world over for an unmistakably loud formal language has, twice now, voluntarily muted that language and handed the entire stage to a different author. That raises a question internal to architecture itself: when the architect with the loudest form willingly lets his own building degrade into a container, who, right now, inherits the authorship that building itself once took such pride in.

III. Putting Technology Into the Skeleton

This is also an extreme sample of a broader shift museums are already going through. More and more institutions treat technology as a curatorial tool or an exhibition gimmick, but in most cases, technology is still a guest “invited in” for the run of a show, and it leaves when the show closes. DATALAND inverts that logic: it builds the entire skeleton of the building as a real-time-running computer. The museum’s Large Nature Model (LNM) is, in the studio’s own words, “the world’s first open-source generative AI model based solely on nature data,” trained on more than 1.2 billion ecological data points and over 500 million images. The data doesn’t come from one institution — it’s sourced through a set of licensed partnerships with the Smithsonian, the Cornell Lab of Ornithology, Getty, the citizen-science platform iNaturalist, and London’s Natural History Museum, among others. The studio also traveled to sixteen rainforest ecosystems directly, using LiDAR scanning, photogrammetry, and ambisonic audio recording to capture first-hand material.

Two officially named systems hold all of this together. Data.Link connects the building, the artworks, and the audience in real time, translating visitors’ biosignals and ambient environmental data into a language the LNM can interpret as emotion. Connectome, built with NVIDIA’s support, is the museum’s centralized high-performance computing system — what the studio calls its “neural center,” giving every installation the capacity to process and respond live. The visual apparatus is its own piece of custom engineering: the largest gallery, the Data Pavilion, runs on 84 Epson 4K projectors, each rated at 20,000 lumens, together generating more than 700 million pixels, automatically stitched, blended, and color-matched by Scalable’s proprietary software so seams disappear and curved surfaces look flawless. Across all five galleries, DATALAND totals 1,577 custom 10K LED panels (roughly 3,500 square feet of LED surface) and 28 LG displays, adding up to the studio’s stated total of 1.5 billion pixels. Which means “curating,” here, no longer means selecting, hanging, and arranging works — it means training a model, setting its parameters, and deciding which direction this building-scale computer should “dream” in.

IV. Two Rooms, Two Postures

Of the museum’s five galleries, the two that most clearly demonstrate the system’s two different postures toward the question of where its data comes from are Gallery C, the Infinity Room, and Gallery D, the Sanctuary.

The technical lineage of Infinity Room: The Dream of Ruwe Pinu is actually older than DATALAND itself. Anadol first conceived the Infinity Room in 2014 as a graduate student at UCLA; it debuted in physical form in 2015 as a 12x12-foot mirrored cube animated by projected black-and-white data, then toured 35 cities worldwide over the following decade, experienced by more than ten million people. DATALAND’s version layers AI-generated scent and “World Models” — a class of generative AI that understands real-world physics and spatial dynamics — onto that lineage, the first immersive space to use the technology. At its center is a translucent hummingbird, Ruwe Pinu, carrying the last breath of a “tree of wisdom.” The image originated in a dream Anadol had during his own travels in the Amazon: a luminous bird circling the edge of a dying forest. The Yawanawá spiritual leader Nixiwaka later recognized the image within his own cosmology and gave it its name. The official wall text closes with a specific credit line: “Developed in partnership with the Yawanawá community of the Acre region of Brazil, with deep gratitude to Chief Nixiwaka Yawanawá. The story of Ruwe Pinu is shared with their permission and remains their cultural heritage.” The room’s sound design closes on the 1987 recording of the last known male Kaua’i ’Ō’ō, an extinct Hawaiian honeyeater.

Gallery D, The Sanctuary: Machine Hallucinations — Rainforest, is positioned in the official text as the museum’s one deliberately contemplative space — dedicated to stillness and sensory reflection, a counterweight to the intensity of the other galleries. Its centerpiece is a thirty-foot Data Painting, produced by an officially named robotic system, QUALIA, which converts the biosensor readings gathered from visitors across every gallery in the museum — not just this final room — into what the studio calls a “collective emotional temperature,” rendered as brushstrokes and color. In other words, what Gallery D presents isn’t one room’s live reaction; it’s the accumulated biometric ledger of an entire visit. Each cycle closes with the molecular scent of the Amazonian moonflower, a rare bloom that opens for a single night each year.

V. Whose Knowledge Counts as Data

Read together, these two rooms make DATALAND’s own published policy more worth close-reading than the outside criticism. According to official materials, the museum partnered with Empatica, a medical-grade wearable company, to build a custom consumer version of its EmbracePlus biosensing wristband for every visitor, capturing heart rate, skin conductivity, and skin temperature. DATALAND states explicitly: “Visitors may opt out at any time before, during, or after their experience. Captured data remains anonymous unless voluntarily claimed through a My.Dataland account. Visitors may delete their data at any time. No data is ever shared with third parties, and all unclaimed data is automatically erased after 30 days.” That’s a fairly specific, checkable policy. But press reports describe multiple visitors finding wristband readings wildly inconsistent with their own smartwatches (114 bpm against a simultaneous 60 bpm reading), which the museum later acknowledged were “artistic interpretations” rather than medical measurements; other visitors reported data collection that felt “disturbingly” extensive, including, they claimed, data logged during a restroom visit. Exactly where the gap opens up between the stated policy and what visitors report — poor execution, or simply visitors never being clearly told what “anonymous” and “opt out” actually mean in practice — can’t be settled here. But that’s the only honest way to evaluate any “responsible data” claim: not by what the policy says, but by what visitors can actually confirm they experienced.

Equally worth asking is a deeper question: whose knowledge counts as data in the first place. The LNM’s publicly credited partners are, without exception, established Western scientific institutions — the Smithsonian, Cornell, Getty. The Yawanawá community’s ecological knowledge and the story of Ruwe Pinu entered the system through an entirely different channel — the official text uses the word “entrusted,” a transmission based on personal relationship and permission, not a routine data-partnership agreement. In other words, the museum runs two parallel logics of what counts as trustworthy data at once: one institution-to-institution, formalized by contract; the other person-to-person, granted through trust. That’s not a verdict on which is right. But it demonstrates, precisely, the real answer to whether art can resist data’s discriminatory tendencies: not a blanket stance like choosing “nature data” over “human data,” but — as with Winds of Yawanawá — leaving the right to collect, profit from, and name the data with the people it came from in the first place. Gallery D’s anonymized, opt-out, auto-deleting visitor policy reads like an attempt to reach that same standard — one visitor experience hasn’t yet fully confirmed it has.

VI. Who’s Paying, and Where the Power Comes From

DATALAND’s list of partners is long: NVIDIA, Google Cloud, founding olfactory partner L’Oréal Luxe, Epson, L-Acoustics, Empatica, LG Electronics, architecture firm Gensler, sustainability and engineering consultancy Arup, projection-blending technology provider Scalable, dessert studio Valerie Confections, and Constellation Immersive, an affiliate of the talent agency CAA — that last one worth noting specifically: a top Hollywood talent agency showing up as “strategic support” on a museum’s partner list is itself a signal that DATALAND is being run as a content and IP business, not simply a nonprofit institution. Another checkable signal on the funding side: Biome Lumina, DATALAND’s founding digital collection released in February 2025 — 1,000 unique AI data sculptures that sold out in 34 minutes, the earliest and most direct market validation the project had, before the physical museum ever opened.

On energy, the official line is consistent: the LNM runs entirely on Google Cloud servers in a low-CO2 compute zone in Oregon, drawing roughly 87% of its power from carbon-free renewable sources; officials repeatedly state that “re-iterating” the experience for each visitor uses roughly the energy of one smartphone charge. That number sounds reassuring, but it only ever describes the power mix of that one Oregon data center — not the full energy and water cost of running Connectome, 84 projectors, 1,577 LED panels, and a museum-wide biosensor network continuously. One of generative AI infrastructure’s most widely documented costs is the water consumed by chip manufacturing and data-center cooling — a figure that never appears in any DATALAND official text. Interestingly, DATALAND’s own materials already demonstrate a more honest way to talk about “partial” efficiency: the press release notes that L-Acoustics’ L-SMART technology, built into the museum’s 250-speaker immersive sound system, delivers “an average 30% greater power efficiency, with zero compromise to the sound” — a specific, checkable claim, explicitly scoped to one subsystem: the audio. If the energy narrative were held to that same granularity — not a blanket “sustainable,” but a specific accounting of projection, compute, climate control, and biosensing individually — at least some of the “environmental paradox” criticism would lose its footing. Right now it hasn’t, and that’s the actual problem: not that DATALAND is lying, but that “sustainable” is being used to describe a much narrower fact than the one it’s implicitly claiming.

DATALAND’s ambitions clearly extend past ticket sales. In October 2025, the museum launched an AI Artist Residency in partnership with Google Arts & Culture; in July 2026, it announced the first cohort, four artists selected from more than 500 applicants — Maya Man, who examines internet identity culture; Lucia Rebolino, an artist and architect working across computational design; Penny Slinger, whose feminist and mystical practice spans more than fifty years; and Jess Starns, focused on history and disability. Each resident receives $25,000 in project support, dedicated studio mentorship, and access to Connectome, the LNM, and Google’s cloud tools, over a residency running June through December 2026, culminating in a public showcase that December. The jury carries real weight: Hans Ulrich Obrist; Paola Antonelli, MoMA’s senior curator of architecture and design; critic Nora Khan; architect Gabriel Kozlowski, whose own project ForestGens maps the human footprint across Amazonia; and Google’s VP of Technology & Society, Mira Lane, among others. An institution willing to fund a residency and staff its jury with curators from MoMA and the Serpentine looks, structurally, more like a serious institution than a photo-op attraction — a piece of evidence worth weighing directly in the “new museum model” question.

VII. Is This Actually a New Museum Model

Multiple visitors in its opening month have compared DATALAND, unfavorably, to teamLab — operating for more than two decades, with venues in Tokyo, Abu Dhabi, and Macau — calling it “a cheap copy with a bigger marketing budget and a smaller soul.” Underneath that comparison are two different business and technical logics: teamLab runs mostly on real-time interaction within preset rules — flowers that wilt when touched, light that gathers and scatters with the crowd — at roughly $30 a ticket. DATALAND is led by a single artist’s brand, substantively backed by capital and compute from NVIDIA and Google, priced at $49–79 standard and up to $129 for premium access, whose core pitch is that every experience is independently generated by the AI model and never repeats. Whether it’s a genuinely new kind of institution — where the building itself is the computer and curating is model training — or a repackaging of an already-proven “immersive digital art” model under an AI narrative and a higher price tag, remains unresolved.

What actually complicates the question is how DATALAND has escalated interactivity from a single exhibition feature into the default condition of the entire institution — the biosensor wristband goes on at the entrance and stays active across all five galleries, not just Gallery D. Anadol states the motive plainly: “For 5,000 years, humans have been emotionally moved by artworks, but the relationship has always flowed in one direction. We asked ourselves: is it possible for artworks to feel us back?” That’s an honest enough statement of DATALAND’s biggest ambition — and exactly where it’s currently weakest. When the scope of data behind that “response” requires a visitor to go dig through the website’s policy language to confirm, what the interactivity delivers isn’t so much participation as a low-grade unease — being displayed and recorded without necessarily understanding the details in the moment. Real interactivity needs more than sensors and real-time rendering. It needs a data contract that’s clearly explained, can survive being questioned, and is actually felt by visitors as such — which is precisely where DATALAND still falls short, even though its own policy text, on paper, is considerably more responsible than most people would assume.

Location: Los Angeles
Links: https://dataland.art/ https://refikanadolstudio.com/
Author: Kaiho Yu

Installation view of Machine Dreams: Rainforest, DATALAND, Los Angeles, CA, June 20, 2026 – January 31, 2027. © 2026 Refik Anadol Studio on behalf of DATALAND. Photo: Refik Anadol Studio

Back to all COLLECTA projects