Astronomers Publish Largest-Ever 3D Map of the Universe, Charting 47 Million Galaxies

On April 15, 2026, the Dark Energy Spectroscopic Instrument collaboration announced it had completed the largest high-resolution three-dimensional map of the universe ever assembled. Over five years of observations from a mountaintop in Arizona, DESI cataloged more than 47 million galaxies and quasars — exceeding its original target of 34 million by 38 percent and capturing six times more spectroscopic data than all previous surveys combined [1][2]. The map stretches from our cosmic neighborhood to "cosmic noon," roughly 10 billion years ago, when the universe's star formation rate peaked. Some of the light DESI recorded had been traveling for 11 billion years before it reached the telescope [3].

The achievement is not merely a trophy of scale. Buried in those 47 million distance measurements are clues to one of the most consequential questions in physics: whether dark energy — the mysterious force accelerating the universe's expansion — is a fixed property of space-time or something that changes over billions of years [4].

How 5,000 Robots Built a Map of the Universe

DESI sits atop the Nicholas U. Mayall 4-meter Telescope at Kitt Peak National Observatory, about 90 kilometers southwest of Tucson. The instrument's core innovation is a focal plane containing 5,000 robotic fiber positioners, each holding a fiber-optic cable roughly the width of a human hair [5]. In about two seconds, these positioners swivel to align with a new set of 5,000 target galaxies. Under ideal conditions, DESI cycles through a fresh batch every 20 minutes [5].

The fiber-optic cables — 150 miles of them in total — route captured light to ten spectrographs, which split it into thousands of individual wavelengths [5]. From these spectra, researchers determine each galaxy's redshift: the degree to which its light has been stretched toward the red end of the spectrum by the expansion of the universe. Redshift serves as a proxy for distance and look-back time. A galaxy at redshift 1.6 is being seen as it was roughly 9.5 billion years ago.

DESI targets different classes of objects at different depths. Bright galaxies extend to redshift 0.4; emission line galaxies, whose active star formation produces detectable spectral lines, reach redshift 1.6; and quasars — the luminous cores of galaxies powered by supermassive black holes — are visible past redshift 3.5 [6]. Together, these tracers let DESI reconstruct how galaxy clustering has evolved across 11 billion years of cosmic history.

Every night, raw data from DESI is automatically transferred via ESnet, the Department of Energy's dedicated high-speed science network, to the National Energy Research Scientific Computing Center (NERSC) at Lawrence Berkeley National Laboratory. There, the Perlmutter supercomputer — equipped with 1,536 GPU-accelerated nodes and a 35-petabyte filesystem — processes the observations, delivering calibrated redshifts "before breakfast," according to NERSC [7]. DESI's Data Release 1, covering the first 13 months of the survey, was a 270-terabyte dataset; processing it on Perlmutter was roughly 40 times faster than on the center's previous systems [7].

Dwarfing Every Previous Survey

Source: DESI Collaboration / SDSS / Berkeley Lab

Data as of Apr 15, 2026CSV

To appreciate the scale: the Sloan Digital Sky Survey, which began observations in 2000 and ran for two decades across multiple phases, obtained spectroscopic measurements for approximately 4.7 million objects [8]. In its first year alone, DESI measured three times more redshifts than SDSS accumulated over 20 years [8]. By survey's end, DESI's 47 million spectra represent a tenfold increase over SDSS's total and roughly six times more than all prior spectroscopic surveys combined [1].

DESI covers approximately 14,000 square degrees of the northern sky — about a third of the full celestial sphere [6]. It reaches redshifts up to 3.5 for quasars, compared to SDSS's typical spectroscopic reach of redshift ~0.7 for galaxies. The DESI Early Data Release, published in June 2023, was followed by Data Release 1 in March 2025, both released under a Creative Commons Attribution 4.0 license and accessible to anyone with a NERSC account through NOIRLab's Astro Data Lab [9][10].

Still, 47 million is a small fraction of what exists. Current estimates place the number of galaxies in the observable universe at approximately 2 trillion [11]. DESI's catalog represents roughly 0.002 percent of that total. This is not a failure of ambition but a reflection of observational reality: most galaxies are too faint, too distant, or too obscured by dust to be detected by current instruments.

Systematic Biases: What the Map Misses

No survey of this kind is without blind spots. DESI's 14,000-square-degree footprint covers the northern sky but leaves southern regions to other instruments. The galactic plane — where the Milky Way's own stars and dust are densest — is largely excluded because stellar crowding and dust extinction corrupt photometric target selection [6]. Low-elevation observations suffer from atmospheric refraction, increased sky brightness, and variable seeing conditions [6].

The survey's target selection also introduces luminosity biases. DESI preferentially detects galaxies with strong emission lines or high surface brightness. Low-surface-brightness galaxies, dwarf galaxies, and heavily dust-obscured systems are underrepresented [6]. These are not minor populations: dwarf galaxies are the most common type in the universe, and dusty star-forming galaxies contribute substantially to the cosmic star formation budget at high redshifts.

The collaboration uses photometric color cuts and careful calibration to mitigate these effects, but they cannot be eliminated entirely. Any cosmological conclusions drawn from the map carry these caveats — a point that independent researchers will need to scrutinize as the full dataset is released.

What the Map Reveals About Dark Energy

DESI's most consequential finding to date concerns the nature of dark energy. The standard cosmological model, known as Lambda-CDM (ΛCDM), treats dark energy as Einstein's cosmological constant — a fixed energy density woven into the fabric of space-time. Under this model, the universe's accelerating expansion proceeds at a steady rate.

Results from DESI's first three years of data, combined with cosmic microwave background measurements and Type Ia supernova observations, tell a different story. The data prefer a dynamical dark energy model — specifically, the Chevallier-Polarski-Linder (CPL) parameterization — over ΛCDM at a confidence level between 2.8 and 4.2 sigma, depending on the dataset combination [12]. Early-universe constraints favor a "phantom" phase of dark energy (with equation-of-state parameter w < −1), while late-universe data prefer "quintessence" (w > −1) [12].

If confirmed, this would mean dark energy's influence on cosmic expansion has changed over time — strengthening in some epochs and weakening in others. "If we're wrong about what dark energy is, we're wrong about the fate of the universe," as one collaboration member put it [4]. The full five-year dataset is expected to sharpen or diminish these hints; the first dark energy results from the complete survey are projected for 2027 [1].

The Hubble Tension: Still Unresolved

One of the most closely watched disputes in cosmology is the "Hubble tension" — a persistent disagreement between two methods of measuring how fast the universe is expanding. The cosmic microwave background (as measured by the Planck satellite) implies a Hubble constant of about 67.4 km/s/Mpc. Local measurements using Cepheid variable stars and Type Ia supernovae yield values around 73 km/s/Mpc. The discrepancy now stands at roughly 6 sigma [12].

DESI's baryon acoustic oscillation measurements provide an independent check. A 2025 analysis using DESI Data Release 1 found that a late-time step in dark energy density could reduce the Hubble tension to below 2.5 sigma [13]. But the tension has not been eliminated. Across multiple dynamical dark energy models tested with DESI data, the Hubble constant inferred from combined datasets consistently aligns with early-universe measurements rather than the local distance-ladder values [12]. The tension persists, though the shape of the disagreement is becoming better defined.

Large-Scale Structure: Tracing the Cosmic Web

DESI does not directly image dark matter. Instead, it maps "tracers" — galaxies whose positions reveal the underlying gravitational scaffolding of the universe, known as the cosmic web [14]. This web consists of dense filaments of dark matter connecting massive galaxy clusters, with vast empty voids between them.

By measuring how galaxy clustering has evolved from cosmic noon to the present, DESI provides empirical constraints on structure formation models. Clustering measurements reveal the masses of dark matter halos that host different galaxy populations and connect observed galaxy populations across different cosmic epochs [14]. The evolution of clustering strength between redshift ~1.6 and the present encodes information about how dark energy has influenced the growth of structure over the past 10 billion years.

The full five-year dataset, with its denser sampling and broader redshift coverage, will allow more precise measurements of baryon acoustic oscillations — sound waves from the early universe imprinted on the distribution of galaxies — which serve as a "standard ruler" for measuring cosmic distances.

Who Built This, and Who Controls It

Source: Berkeley Lab / DOE / Wikipedia

Data as of Apr 15, 2026CSV

DESI is managed by Lawrence Berkeley National Laboratory and funded primarily by the U.S. Department of Energy's Office of Science, which contributed $56 million to the instrument's construction. An additional $19 million came from non-federal sources, bringing the total construction cost to $75 million — $1.9 million under budget [15]. Operational costs, including telescope time, computing infrastructure, and personnel, add substantially to this figure but are harder to isolate from host institution budgets.

The collaboration includes more than 900 researchers, among them 300 PhD students, from over 70 institutions across the United States, United Kingdom, France, Germany, Spain, Mexico, China, Korea, Australia, Canada, Colombia, and Switzerland [16][2]. Additional funding comes from the National Science Foundation, the UK's Science and Technology Facilities Council, the Gordon and Betty Moore Foundation, the Heising-Simons Foundation, and the French CEA [2].

Data governance follows a phased release model. Collaboration members have proprietary access during active analysis periods, after which data releases become public. The Early Data Release (June 2023) and Data Release 1 (March 2025) are both available under CC BY 4.0 licenses [9]. The collaboration's formal Publication Policy, overseen by a Publication Board, governs how results are reviewed internally before journal submission [10].

Whether this process constitutes adequate independent scrutiny is debatable. The collaboration's internal review is extensive — papers typically go through multiple rounds of internal review before submission to journals — but the sheer complexity of the data pipeline means that replication by external groups is resource-intensive. Independent verification requires not just the raw data (which is publicly released) but also the computational infrastructure to process it.

The Opportunity Cost Debate

Source: OpenAlex

Data as of Jan 1, 2026CSV

DESI's $75 million construction cost is modest by the standards of flagship physics experiments (the Large Hadron Collider cost approximately $4.75 billion; the James Webb Space Telescope came in at $10 billion). But critics of large survey projects argue that the relevant comparison is not with particle physics but with the individual research grants that astronomers compete for — typically $100,000 to $500,000 per year from the NSF [17].

The National Academies' assessments of federal astronomy funding have repeatedly noted that capital costs of new facilities create "downward pressure on budget-flexible grants programs" [17]. Each time the NSF commits to a new major instrument, its operating costs must come from the same limited pool that supports existing facilities and investigator-led research. The decadal survey process, critics argue, creates a structural bias toward large instruments because it is easier to justify a small number of big-budget projects to Congress than a constellation of smaller ones [17].

Defenders counter that large surveys generate outsized scientific returns precisely because they serve many investigators simultaneously. SDSS, which cost approximately $120 million across its phases, has produced over 10,000 peer-reviewed papers and is among the most cited astronomical facilities in history [8]. The breadth of science enabled by a single large dataset — from cosmology to stellar astrophysics to galaxy evolution — is difficult to replicate through individual small-scale programs. Academic publication trends reflect this: research papers related to dark energy spectroscopic instruments have grown from approximately 2,600 per year in 2011 to a peak of over 16,600 in 2023 [18].

Second-Order Uses: From AI Training to Satellite Catalogs

DESI's data is already being incorporated into broader scientific infrastructure. The Multimodal Universe dataset, a 100-terabyte collection assembled by researchers at Oxford and Berkeley, includes 1 million DESI galaxy spectra alongside data from SDSS and other surveys [19]. This dataset is designed to train astronomical foundation models — large neural networks that can classify galaxy types, predict redshifts, and estimate stellar masses from raw observational data. Early results show redshift predictions achieving an R² of 0.986 and morphology classification accuracy between 73.5 and 89.3 percent [19].

The dataset is freely available on Hugging Face, meaning commercial AI companies can access it for training purposes — raising questions about who captures value from publicly funded science. DESI's CC BY 4.0 license permits commercial use with attribution, a deliberate choice to maximize scientific impact but one that also means private firms can build proprietary products on top of taxpayer-funded observations.

Beyond AI, precise galaxy catalogs have practical applications in space situational awareness. As satellite constellations grow — SpaceX's Starlink alone plans tens of thousands of spacecraft — accurate background source catalogs help distinguish astronomical objects from orbital debris in tracking systems. This is a secondary benefit, not a design goal, but it illustrates how foundational datasets find uses their creators did not anticipate.

What Comes Next

DESI's work is not finished. Because of the instrument's better-than-expected performance and the tantalizing hints about evolving dark energy, the collaboration has secured approval to continue observations through 2028, extending the map to cover additional sky area [1]. The immediate priority is processing the complete five-year dataset, with the first cosmological results from the full survey expected in 2027 [1].

The next generation of surveys is already under construction. The Vera C. Rubin Observatory in Chile, expected to begin its Legacy Survey of Space and Time in the coming years, will photograph the entire southern sky repeatedly over a decade, cataloging an estimated 20 billion galaxies photometrically — though without the spectroscopic precision that DESI provides [17]. The European Space Agency's Euclid mission, launched in 2023, is conducting its own wide-field spectroscopic survey from space. Together, these instruments will push the census of the observable universe from millions of precise redshifts to billions of approximate ones.

The fundamental question remains open: is dark energy a constant, or is it changing? DESI's 47 million galaxies have sharpened the question. The answer — with implications for whether the universe will expand forever, slow down, or eventually reverse course — may come from the same data, once the full analysis is complete.