# What Is a Megaquop Machine, and How Far Away Is It?

One error in a million per [logical gate](https://quantumintel.tech/glossary/logical-qubit). That is the headline number in John Preskill's megaquop specification — and it sits three to four orders of magnitude beyond the best physical two-qubit gates built today. [Google Quantum AI](https://quantumintel.tech/companies/google-quantum-ai)'s Willow chip delivers roughly one error per 300 two-qubit operations. The gap is not closed by better fabrication; it is closed by error correction, and that distinction is the entire point of the megaquop concept.

Preskill introduced the term at the Q2B conference in Silicon Valley on 11 December 2024, and published the formal treatment the following March as *Beyond NISQ: The Megaquop Machine* in ACM Transactions on Quantum Computing (DOI 10.1145/3723153). The specification is explicit: approximately one hundred [logical qubits](https://quantumintel.tech/glossary/logical-qubit), circuits of approximately ten thousand layers deep, and a logical gate error rate of order one in a million. Those three numbers together define roughly a million reliable quantum operations — hence the name.

The same researcher who coined [NISQ](https://quantumintel.tech/glossary/nisq) — noisy intermediate-scale quantum — in December 2017 is now retiring it. His argument is that NISQ described an era by what it lacked; megaquop describes one by what it can do.

---

## Where the Term Comes From

Preskill coined NISQ at a keynote in December 2017 and formalised it in arXiv:1801.00862 in January 2018. The label stuck because it was accurate: machines with dozens to hundreds of noisy physical qubits, no error correction, shallow circuits. For several years that framing was useful.

The problem is that NISQ aged poorly as a forward-looking category. A machine with more noisy qubits is still a noisy machine; the label offers no natural scale for progress. Others had proposed successors that preserved the ISQ suffix, appending letters or swapping the first word. Preskill explicitly rejected that approach, writing that he would rather leave ISQ behind as the field moves forward. The objection is not stylistic — a label that describes absence cannot measure progress.

The megaquop framing provides a ruler instead. A gigaquop machine runs a billion reliable operations; a teraquop runs a trillion. Each step is a factor of a thousand. At the far end sits what Preskill calls a FASQ machine — Fault-Tolerant Application-Scale Quantum — crediting the acronym to Andrew Landahl and describing the goal itself as "a rather distant" one. Every milestone between here and FASQ belongs to the megaquop era and its successors.

---

## The Specification in Detail

The megaquop target is not a vague aspiration. Preskill is explicit that "mega" is approximate — "not precisely a million but somewhere in the vicinity of a million" — but the underlying circuit task is checkable: roughly one hundred logical qubits running circuits of roughly ten thousand layers. A machine either executes that without collapsing into noise, or it does not.

The logical gate error rate required is of order one in a million. Preskill notes that the rate could be somewhat larger if error mitigation continues to carry weight alongside error correction. That caveat matters: mitigation's sampling overhead grows exponentially with circuit size, so it shifts the specification rather than relaxing it in any fundamental sense.

On physical qubit overhead, Preskill offers a rough guess — tens of thousands of high-quality physical qubits could suffice — while labelling it a guess, not an estimate. The true number depends on which error-correcting code is used, what the underlying physical error rate is, and how much of the machine is consumed producing the magic states a [fault-tolerant quantum computing](https://quantumintel.tech/glossary/fault-tolerant-quantum-computing) architecture requires.

---

## The Hardware Gap Today

The source material provides two concrete reference points for physical gate fidelity. Two-qubit gates on Google's Willow chip fail roughly once in 300. The best superconducting two-qubit gate reported so far, on an MIT fluxonium device, fails roughly once in 1,280. The megaquop target is one failure per million logical gate operations. That is three to four orders of magnitude of improvement, and the path there runs through [below-threshold](https://quantumintel.tech/glossary/below-threshold) error correction, not through hardware refinement alone.

This is the inversion at the heart of fault-tolerant quantum computing. In the NISQ regime, adding more physical qubits does not help: every additional gate adds noise and the circuit must stay shallow to preserve any chance of a useful answer. Error correction turns that relationship around. Once the code suppresses errors faster than it introduces them — the below-threshold condition — spending more physical qubits per logical qubit makes the logical qubit more reliable. Scale becomes an asset rather than a liability.

Three things must hold simultaneously for that inversion to pay off in a megaquop-class machine, as the source identifies:

1. **Error suppression must scale with code size.** The promise of QEC is that adding more physical qubits per logical qubit keeps reducing the logical error rate. Whether that suppression continues as codes grow to practically useful sizes remains an open experimental question.

2. **A decoder must keep pace with the hardware.** A fault-tolerant processor generates syndrome data that must be interpreted in real time. A decoder that falls behind the physical clock rate accumulates a backlog that undermines the correction it was meant to provide.

3. **Correlated errors must remain rare.** Standard error correction models assume errors are largely independent. Rare but correlated failure events — hardware crosstalk, cosmic rays, fabrication defects — can defeat a code even when average error rates sit below threshold. Nobody has demonstrated that this floor is cleared at scale.

---

## What a Megaquop Machine Would Not Do

The most commercially loaded question about any advance in quantum hardware is whether it threatens public-key encryption. The source is unambiguous: breaking RSA-2048 is a gigaquop problem, one thousand times further up the scale than the megaquop target. A megaquop machine leaves current encryption intact.

Preskill is equally plain about the application question. He says he cannot name the first useful application a megaquop machine will enable, and expects the early payoff to be scientific rather than commercial. The most plausible early targets involve simulating quantum systems — chemistry, materials, condensed matter physics — rather than optimisation or cryptanalysis problems of direct enterprise relevance.

This is a useful calibration for investors evaluating near-term commercial narratives. The megaquop specification describes a machine that would represent a genuine transition from the NISQ era, but Preskill's own framing locates the commercial payoff further down the scale.

---

## Industry Implications

The megaquop framing does something tactically useful for the field: it gives hardware teams a single, unambiguous intermediate target that is harder than anything built so far but does not require solving every problem at once. A hundred logical qubits at one-in-a-million fidelity with ten-thousand-layer circuits is a specification you can benchmark against, unlike vaguer claims about quantum advantage or quantum supremacy.

For enterprise buyers and investors, the specification also clarifies what hardware announcements should be evaluated against. Physical qubit counts, raw gate fidelities, and even demonstrations of error suppression on small codes are all necessary way-stations — but none of them, individually, constitutes a megaquop-class machine. The relevant question for any hardware announcement is: how many logical qubits, at what logical error rate, sustaining what circuit depth?

The answer to that question, across every platform currently in the field, remains well below the megaquop threshold. The three unsolved problems — scalable error suppression, real-time decoding, and correlated error floors — are active research fronts, not engineering timelines. Preskill, who has more standing than most to estimate when the era arrives, says he does not know.

---

## Key Takeaways

- **The megaquop specification**: ~100 logical qubits, ~10,000 circuit depth, ~1-in-a-million logical gate error rate — together constituting roughly one million reliable quantum operations.
- **The hardware gap**: Google's Willow delivers ~1 error per 300 two-qubit gates; the megaquop target is 3–4 orders of magnitude better, achievable only through fault-tolerant error correction, not hardware refinement alone.
- **What it will not do**: Breaking RSA-2048 is a gigaquop problem — 1,000× beyond the megaquop threshold. Current encryption is not threatened by a megaquop-class machine.
- **The naming logic**: Preskill retired the NISQ framing because it describes absence rather than capability. Megaquop, gigaquop, teraquop, and ultimately FASQ provide a measurable scale of progress.
- **Three open problems**: Scalable error suppression, real-time decoding that keeps pace with hardware, and demonstrating a floor on correlated errors — none is solved at scale.
- **Application outlook**: Preskill himself cannot name the first useful application and expects early payoffs to be scientific, not commercial.
- **Andrew Landahl** is credited by Preskill with the FASQ acronym.

---

## Frequently Asked Questions

**What is a megaquop machine?**
A megaquop machine is a quantum computer capable of running approximately one million reliable quantum operations. Defined by John Preskill in a December 2024 keynote and a March 2025 paper in ACM Transactions on Quantum Computing, the concrete specification is roughly one hundred logical qubits, circuits roughly ten thousand layers deep, and a logical gate error rate of approximately one in a million.

**How is megaquop different from NISQ?**
NISQ — noisy intermediate-scale quantum — described machines by what they lacked: error correction. Megaquop describes a machine by what it can do: execute a million reliable operations. The shift from absence-based to capability-based naming also provides a natural scale — megaquop, gigaquop, teraquop — that NISQ could not.

**Will a megaquop machine break RSA encryption?**
No. According to Preskill's own framing, breaking RSA-2048 is a gigaquop problem — one thousand times further up the scale than the megaquop target. A megaquop-class machine would leave current public-key encryption intact.

**How close is the best hardware today to the megaquop specification?**
Google's Willow chip achieves roughly one two-qubit gate error per 300 operations. The best superconducting result reported so far, on an MIT fluxonium device, is roughly one error per 1,280. The megaquop logical gate target is one error per million — three to four orders of magnitude beyond current physical gate performance.

**What are the three main obstacles to building a megaquop machine?**
Preskill identifies three: (1) error suppression must continue to improve as error-correcting codes grow in size; (2) a classical decoder must process syndrome data fast enough to keep pace with the quantum hardware in real time; and (3) the rate of rare correlated errors — which can defeat a code even when average error rates are below threshold — must be demonstrated to be sufficiently low. None of these has been cleared at practically relevant scale.

**When will a megaquop machine exist?**
Preskill says he does not know. He offers no timeline, and the source material does not provide one.