On a mountaintop in Arizona, a telescope with five thousand robotic fingers spent three years staring at fourteen million galaxies. Each night the robots reset themselves and pointed (with a precision finer than a human hair) at a fresh patch of sky. The instrument is called DESI, the Dark Energy Spectroscopic Instrument, and its job is dull in the way that all great experiments are dull. Measure the redshifts. Map the positions. Build the biggest three-dimensional atlas of the universe ever attempted, and then ask it a single question.

Is dark energy actually constant?

For 25 years the answer has been a confident yes. Dark energy (the mysterious anti-gravitational pressure pushing the cosmos apart) had a value, that value was a constant, and the constant had a name: lambda. Stick lambda into Einstein’s equations next to cold dark matter, sprinkle on some inflation, and you get the standard model of cosmology. We call it ΛCDM (lambda Cold Dark Matter), and it has been one of the great success stories of modern physics. It explains the cosmic microwave background. It explains how galaxies cluster. It explains, with intense accuracy, the abundances of hydrogen and helium that poured out of the first three minutes.

It was enough. For a while.

Then in March of 2025 the DESI collaboration released its second batch of results, and the answer to the question started to look like … no. When you combine DESI’s measurements with the cosmic microwave background and a handful of supernova catalogs, dark energy suddenly appears to be changing with time. Not constant. Evolving. The statistical preference varied depending on which supernova sample you trust. Not quite a discovery in the gold-plated sense. But no longer noise.

So what?

Here’s the thing: The universe you learned about in school was temporary. That version, with its fixed cosmological constant and a clean, settled story about how things end, was the version that fit the data in 2001. The data we have now is starting to fit a different story—and that story has entirely different endings for our universe.

Consider the script you were probably handed for the fate of the cosmos: heat death. A slow, dignified fade to black as galaxies recede past each other’s horizons, stars exhaust their fuel, and the universe drifts toward maximum entropy over timescales so vast that the word “trillion” stops carrying weight. That ending depends on lambda being lambda. A constant. Forever.

An evolving dark energy throws the script out. Maybe the expansion accelerates harder over the next few billion years and we get the big rip, where dark energy eventually tears apart galaxies, then solar systems, then atoms themselves. Maybe it weakens and the universe slows, stalls, even re-collapses into a big crunch. Maybe the parameter wanders around in ways we have not even thought to model. Pick your apocalypse. We no longer know which one we live in.

And here is the part that keeps me up at night.

The DESI result is not arriving in a vacuum, scientifically speaking. The Hubble tension, that stubborn disagreement between how fast the local universe is expanding (measured from supernovae) and how fast the early universe says it should be (measured from the cosmic microwave background), refuses to resolve. Every new measurement, every clever workaround, makes the tension worse, not better. The “sigma-eight tension,” which tracks how clumpy the universe is at large scales, has been quietly nagging for a decade. Independent surveys keep finding the cosmos less lumpy than ΛCDM predicts. Pick your favorite anomaly. They are plural.

For most of my career, working cosmologists have treated these as separate puzzles. Each one had its own cottage industry of proposed fixes (early dark energy, decaying dark matter, modified gravity, neutrino masses doing gymnastics), and each fix was tuned to handle one anomaly without breaking everything else. That worked for a while. It is not really working anymore.

Because the anomalies are correlated. They are pointing in roughly the same direction—that our standard formulation of cosmology is wrong … or at least, inadequate. I want to be careful here. I have been wrong before, and I will be wrong again, and one of the occupational hazards of being a cosmologist is mistaking the noise for the signal at exactly the moment the signal is starting to arrive. It is possible that DESI Year Five quietly walks the result back. It is possible that some unglamorous effect in the data is doing the heavy lifting in the Hubble tension. It is possible we will look back in 2030 and laugh at this very article.

But I do not think we will.

The shape of what is happening in cosmology right now feels less like a string of unrelated puzzles and more like the bubbling intensity just before a paradigm shifts. We have been here before, in physics, and we will be here again.

If you are reading this, you are watching the early phase of one of those moments. The textbooks make the old revolutions (relativity, quantum mechanics) sound clean and inevitable. They were not. They were messy and contested and lived through in real time by physicists who could not be sure, in the moment, whether they were witnessing a new physics or chasing their own tails. We get to find out which one this is.

So I am keeping my eye on the next DESI release, the Vera Rubin Observatory’s first full survey year, and the next batch of CMB-S4 papers. Not because I know what they will say. Because for the first time in a quarter-century, I genuinely do not.