Oratomic raised $300 million on the claim that breaking RSA needs 10,000 qubits instead of millions

Oratomic, a Pasadena quantum computing startup that came out of stealth at the end of March, has raised $300 million in a Series A co-led by ARCH Venture Partners, Spark Capital and Khosla Ventures. Bezos Expeditions, Index Ventures, General Catalyst, Lowercarbon Capital and Bain Capital also put money in. Vinod Khosla wrote on X that it was his firm's "largest initial investment yet."

That is a lot of money for a company with no product, no customers and no plans to sell anything for years. What the investors are actually buying is a resource estimate.

The number comes from a preprint

On March 30, nine authors including Oratomic's founders posted a paper to arXiv titled "Shor's algorithm is possible with as few as 10,000 reconfigurable atomic qubits." The claim in the title is the whole company.

Shor's algorithm is the one that breaks RSA and elliptic-curve cryptography. Running it on real hardware has always been an error correction problem: qubits are noisy, so you spend most of your machine on redundancy. Optimized estimates have historically landed in the millions of physical qubits. Craig Gidney at Google Quantum AI got that down to under a million in a May 2025 paper, and that itself was a 20x cut from his own 2019 figure of about 20 million.

The Oratomic paper says 10,000. It also gives a runtime: with 26,000 physical qubits, a discrete log on the P-256 elliptic curve would take "just a few days," while factoring RSA-2048 runs "one to two orders of magnitude longer." Read that second number carefully. One to two orders of magnitude longer than a few days is somewhere between a month and a year of continuous, error-corrected operation. The 10,000-qubit machine is the minimum viable version, not the fast one.

What the atoms are doing

The hardware is neutral atoms held in optical tweezers, which are tightly focused laser beams that trap individual atoms in a lattice. The trick that buys the overhead reduction is that you can pick the atoms up and move them mid-computation.

That matters because error correction overhead is mostly a connectivity problem. A superconducting chip has qubits soldered into a fixed grid, so each one can only talk to its neighbors, which forces you into low-rate codes like the surface code where hundreds of physical qubits buy you one logical qubit. If you can physically shuttle atoms across the array and pair up any two of them, you can run high-rate codes that encode many logical qubits per block, and you can do logical operations by moving whole blocks rather than by grinding through rounds of local syndrome extraction. The paper stacks that with a more efficient logical instruction set and better circuit design for the arithmetic.

The mechanism is real and the authors are the people who built the field. It is still a resource estimate, made "under plausible assumptions," about a machine nobody has built.

Demonstrated versus calculated

The paper is honest about the gap, and cites three experimental results as the foundation: universal fault-tolerant operations below the error correction threshold, computation on arrays of hundreds of qubits, and trapping arrays of more than 6,000 coherent atoms.

Note that those are three different experiments. Trapping 6,000 atoms is not the same as computing with 6,000 atoms, and the systems that have done fault-tolerant logic did it on a couple of orders of magnitude fewer qubits. CEO Dolev Bluvstein told TechCrunch the company has "already experimentally demonstrated all of the core components required of that computer at a slightly smaller scale." Slightly is carrying weight there. The paper's own phrasing is that "substantial engineering challenges remain."

The specific challenges are the boring ones that kill neutral-atom scaling: atoms get lost from the trap and have to be reloaded, moving them takes milliseconds while gates take microseconds, and mid-circuit measurement has to happen without heating everything nearby. None of that is exotic physics. All of it is why the machine does not exist.

The man who named the era they are skipping

The team list is the strongest thing on Oratomic's website. Bluvstein, Manuel Endres, Harry Levine, Madelyn Cain, Qian Xu, CTO Hsin-Yuan Huang, and John Preskill.

Preskill coined the term NISQ, for noisy intermediate-scale quantum, to describe the current generation of machines that are too small and too noisy to do anything useful. Oratomic's stated plan is to build no NISQ hardware at all: no prototypes for research customers, no cloud access to a noisy chip, nothing to sell until the fault-tolerant machine works. Preskill is on the founding team of a company whose entire thesis is that his era should be skipped.

"You would have not previously been able to convince any of us to start a quantum computing company, because we just thought it was way too far away," Bluvstein told TechCrunch. "Only when we made this recent breakthrough did we simultaneously all change our minds."

The field is crowded with people making the same promise

PsiQuantum, valued around $7 billion last September, is also bypassing NISQ and says it will deliver a million-qubit machine by the end of next year. Atom Computing, which is also doing neutral atoms, raised its own $300 million in June. QuEra is working with AWS toward a fault-tolerant machine in 2028. Quantinuum filed to go public in May at a $12.7 billion valuation on trapped ions. Google and IBM are still on superconducting qubits.

Every one of them has a date. None of them has a fault-tolerant computer. Oratomic's differentiator is that its date requires two orders of magnitude fewer qubits than PsiQuantum's, which is either the best argument in the sector or a very expensive way of saying the calculation was easier than the construction.

The thing to watch is not the valuation or the timeline. It is whether Oratomic can put a few thousand atoms into an array and run error-corrected logic on all of them at once. That is a measurable event, and it either happens or it does not.