Quantum computers made of single atoms now reach 12,001 spots

What happened

Think of an egg carton, where each dip holds one egg. Now shrink it, and make the dips out of tightly focused laser beams. Put one single atom in each dip. That is a neutral atom quantum computer.

Each trapped atom holds one bit of quantum information. To make two atoms talk, you poke one with a laser so its outer electron swells up. The swollen atom then reaches its neighbour and changes it, and that is how one operation gets done.

A new review paper takes stock of this whole field. It is a survey rather than an experiment, because it gathers what many labs have published and reports where the numbers stand. Two numbers stand out. Arrays now reach 12,001 trap spots, and one atom held its bit for 119 seconds.

The test

There is no test, and that is the first thing to say. The paper is called “Neutral atom quantum computing” and Mark Saffman wrote it alone. He is at the Department of Physics at the University of Wisconsin-Madison, and also at Infleqtion in Madison, in the United States. The paper went on arXiv on 31 August 2026. No journal or conference is named, and it runs no new experiment.

What it does instead is collect the record. The idea was proposed more than 25 years ago, and a full set of working operations arrived about 10 years after that. However, the big jump in size and accuracy came only in the last few years.

The review reports the best published results for each part of the job. Trapping atoms. Holding their state. Reading them out. Doing operations. Correcting errors. Therefore, judge it as a map of the field, not as evidence for any one claim.

The result

Start with size. One array reported 12,001 trap sites for caesium atoms. Another held 2024 rubidium atoms in 2025 sites, so almost every spot was full. A ytterbium array had 1225 sites, filled 99 percent of the time.

Now memory. The record for a single atom keeping its bit is 119 seconds, which is a long time in this business. With extra correction pulses, teams have held a full quantum state for 40 seconds in strontium and 12.6 seconds in caesium.

Now accuracy. Single atom steps driven by microwaves pass 0.9999. Two-atom steps, the hard ones, have passed 0.99 with four different kinds of atom. However, the paper is clear that this is still not good enough, and still far from what physics allows.

Error correction has started to work. The largest demonstration used up to 96 logical qubits, and making the code bigger improved the result by a factor of 2.14.

What it means

Here is the useful shape of the news. The pieces work, but nobody has put all of them in one machine at once. You cannot add the numbers above together, because each one comes from a different lab, with a different atom, on a different table.

The review is honest about the wall ahead. Getting one array past 100,000 atoms looks very hard. This is because the devices that shape the laser beams have a limited number of pixels. Getting rid of the heat from all that laser power is also already a problem at around 10,000 sites.

Therefore, the plan for going bigger is to link several arrays with light, the way you join computers with cables. The paper says those links exist but are early, and still lag behind what happens inside one array.

Business ideas from this paper

1. A plain, honest briefing on how far away quantum really is. You take one industry, list the tasks people claim quantum will do, and give a straight answer on each, using accuracy and error correction numbers from reviews like this one. Who buys it: heads of technology at banks, chemical makers and drug firms who keep getting quantum sales calls. A price to test: 2,000 dollars for a two-hour session and a written note. A one-week test: offer it to ten firms that got a quantum pitch this quarter, and count how many book a slot.
2. Heat control kits for atom trap benches. The review names two hardware walls. Beam shaping devices run out of pixels, and dumping the heat from laser power is already noticeable at 10,000 sites. Sell a cooled, pre-aligned mount that fixes the second one. Who buys it: university atom array groups and quantum hardware startups. A price to test: 3,500 dollars per kit. A one-week test: email 20 groups a one-page spec sheet and count how many ask for a quote.
3. A short hands-on course in optical tweezer work. Two days on beam alignment, vacuum handling and imaging, aimed at physics graduates who want a job. Who buys it: quantum hardware firms that cannot hire trained hands fast enough, and the graduates themselves. A price to test: 1,200 dollars per seat. A one-week test: put up a waiting list page, share it in two atomic physics forums, and count sign-ups.

How sure can you be?

Be careful here. THIS PAPER IS A REVIEW, NOT AN EXPERIMENT. It reports no new measurement, because every number in it is somebody else’s published result, chosen and summarised by one author.

A single author also means a single view of what matters. The paper names no journal or conference, and as far as the fetched version shows, it has not been through peer review. There is no code or data to check.

The author states two affiliations. One is a university physics department, and one is a quantum computing company. That is disclosed in the paper, so read the outlook with that in mind.

The paper flags the hard parts itself. Scaling one array past 100,000 qubits faces steep engineering problems, and two-atom accuracy needs to improve further. Links between separate arrays are early and lag behind. What would settle the question is one machine that holds thousands of qubits, runs a long error-corrected program, and beats an ordinary computer at a task somebody actually wants done.

Do this today

The next time somebody sells you anything quantum, ask three questions. How many qubits. What is the two-qubit accuracy. Was error correction switched on. Those three answers separate real progress from a slide.

Source: Neutral atom quantum computing, August 2026. arXiv:2608.30783 · arxiv.org (preprint · not yet peer reviewed).

Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.