Pillar E · State as of 2026-09-08

Quantum photonics: what does PsiQuantum promise and what is measured?

As of September 8, 2026, PsiQuantum has published the strongest component-level metrics in photonic quantum computing — led by a 99.22% two-qubit fusion fidelity on 300-mm foundry chips (Nature, February 2025) — and no public system-level result: no algorithm run, no logical qubit, no machine. The same company holds the largest capital stack in the field: a US$1B Series E at a $7B valuation (September 2025), A$940M from Australia's governments (2024), a $500M+ Illinois package (2024) and a $125M DARPA Stage C agreement (July 2026). Rosetta Q tracks both ledgers — the measured and the promised — below.

→ Leer en español

State as of: 2026-09-08

PsiQuantum is the best-funded independent quantum computing company on public record, and one of the least publicly measured. That combination is not an accusation — it is the company's stated strategy: build no intermediate machines and go straight to a fault-tolerant photonic computer of roughly a million physical qubits. This post separates the two ledgers, with a source and a date on every number: what PsiQuantum has measured and published, and what it has promised, to whom, and with whose money. Status date: September 8, 2026.

How is PsiQuantum's photonic approach different?

It skips the intermediate machine entirely. Most vendors ship small processors and scale up in public — superconductors, trapped ions, neutral atoms all have public devices you can benchmark today (our architecture map). PsiQuantum's founding premise is that commercially useful quantum computing requires error correction and therefore "on the order of a million physical qubits" (Series E announcement, Sep 2025), so it offers no public machine at any intermediate size: components first, then utility scale.

The qubit is a single photon in a silicon-photonics circuit, and three properties drive the whole strategy:

Photons are manufactured, not trapped. The chips are fabricated on 300-mm wafers at GlobalFoundries' Fab 8 in New York, in a process with more than 20 photolithography levels (Nature, Feb 2025). The scaling bet is semiconductor manufacturing, not laboratory assembly.
Photons connect over fiber. Cabinets are meant to be networked "with standard optical fiber" (Brisbane groundbreaking release, Jun 2026), and the same Nature paper reports a measured chip-to-chip qubit interconnect at 99.72% fidelity — networking is a strength of light, not an afterthought.
Photons don't wait. A photon does not sit in a trap accumulating decoherence; it either arrives or it is lost. The dominant error is loss, and the two-qubit operations of the architecture ("fusions") succeed only probabilistically, which is paid for with heavy multiplexing (Bartolucci et al., Nature Communications, 2023).

Cooling sits between the extremes: the photonic circuits themselves do not need dilution-refrigerator temperatures, but the single-photon detectors are superconducting and cryogenic — which is why the first hard milestone of the Brisbane site is a cryoplant described as "one of the largest ever built for quantum computing", due in the second half of 2027 (Business Wire, Jun 2026).

What has PsiQuantum actually measured and published?

The strongest public component metrics in photonics — and only component metrics. In February 2025 the company published a peer-reviewed characterization of chips built on its 300-mm foundry process (Nature 641). Every number below is measured on manufactured hardware, with stated uncertainty:

Metric	Measured value	What it is	Source
State preparation and measurement	99.98% ± 0.01%	writing and reading the photonic qubit	Nature, Feb 2025
Chip-to-chip interconnect fidelity	99.72% ± 0.04%	the qubit survives a fiber hop between chips	Nature, Feb 2025
Photon spectral purity	99.5% ± 0.1%	how clean each single-photon source is	Nature, Feb 2025
Hong-Ou-Mandel visibility	99.50% ± 0.25%	two-photon interference quality	Nature, Feb 2025
Two-qubit fusion (Bell measurement)	99.22% ± 0.12%	the entangling operation of the architecture	Nature, Feb 2025
On-chip detector efficiency	88.9% ± 3.5% (avg)	superconducting single-photon detectors	Nature, Feb 2025
SiN waveguide propagation loss	0.5 ± 0.3 dB/m	how fast the chip loses photons	Nature, Feb 2025

There are two honest readings of that table. Read charitably, these are among the best figures ever published for integrated photonic qubits — and they come from a manufacturable foundry process, not a hand-tuned lab chip. Read strictly, every row is a component. The paper reports no algorithm, no logical qubit, no error-corrected memory, no system — and its own authors say so, listing what remains: "further reduce SiN materials and component losses, improve filter performance and increase detector efficiency", plus low-loss fast switches and better fiber-to-chip connections (Nature, Feb 2025).

What has PsiQuantum promised, and with whose money?

Two utility-scale sites, a manufacturing chain, and no published date for a working machine. Every commitment below is public and dated; none of them is a benchmark.

Date	Commitment	Amount	Source
Apr 29, 2024	Australian + Queensland governments (equity, loans, grants) for the Brisbane machine	A$940M	The Quantum Insider
Jul 25, 2024	Illinois Quantum and Microelectronics Park, Chicago: state park budget $500M, cryo plant $200M, incentives $500M+ over 30 years	$500M+	PsiQuantum
Sep 10, 2025	Series E led by BlackRock affiliates, with Temasek, Baillie Gifford and NVIDIA's NVentures, at a $7B valuation	$1B	The Quantum Insider
May 21, 2026	Letter of intent, U.S. Department of Commerce	$100M	The Quantum Insider
Jun 17, 2026	Ground broken at Moreton Bay (Brisbane); "tens of thousands of photonic quantum chips"; cryoplant due 2H-2027	—	Business Wire
Jul 22, 2026	DARPA Quantum Benchmarking Initiative, Stage C — one of two companies in the final phase	$125M	The Quantum Insider

Three details matter for reading this ledger. First, the DARPA row is the closest thing to independent verification that exists: Stage C followed hands-on, component-level testing by DARPA teams, and DARPA leadership now says "it now seems likely that someone will build a utility-scale quantum computer by 2033" (The Quantum Insider, Jul 2026) — a statement about the field, not a delivery date for PsiQuantum. Second, the only hard scheduled milestone for Brisbane is infrastructure, not computation: a cryoplant in the second half of 2027. Third, we found no public document that commits PsiQuantum to an operational date for either machine.

What is the catch with photons?

Loss — and the fact that the only public photonic system is far from fault tolerance. Photons do not decohere while idle, but every waveguide, coupler, filter and connector destroys a fraction of them, and fusions succeed only probabilistically. The architecture's answer is redundancy at industrial scale: "tens of thousands of photonic quantum chips" per site (Business Wire, Jun 2026).

The nearest public data point for a complete photonic system comes from a different company with a different photonic approach: Xanadu's Aurora (Nature, Jan 2025) networked 35 photonic chips in four server racks into 12 physical qubit modes per clock cycle and ran a distance-2 repetition code with real-time decoding — and its authors state that component losses must improve by roughly 20-30× for fault-tolerant operation. That is the honest state of photonics as a system in public: assembled, networked, and orders of magnitude from threshold. PsiQuantum's published component numbers are better than Aurora's system numbers — but a system measurement of PsiQuantum's architecture does not publicly exist.

How does photonics compare with matter-based platforms today?

On measured system results, photonics is behind; on the manufacturing story, it is ahead. Matter-based platforms have published error-corrected logical qubits below threshold, 48 logical qubits from 98 physical ions, and ≈96 logical qubits on 448 neutral atoms — all covered with sources in our 2026 architecture map, the neutral-atoms verdict and the superconductors-vs-ions comparison; we do not re-litigate them here. Photonics' counter-entry is the best fab-and-networking story in the field, with zero public system results. And on the Rosetta Q scale that matters most, every architecture — photonic or not — still shares one number: measured end-to-end advantage on a commercial problem, 0.

What would move this verdict?

A system-level number on photonic hardware, published. Any one of these would do it: a logical qubit with a measured error rate on PsiQuantum hardware; an error-corrected memory that outperforms its physical components; or an algorithm run end-to-end at matched budget against a strong classical baseline — the advantaged-solve standard Rosetta Q applies to every vendor equally. The moment one of those exists with a date and raw data, this post gets refreshed under the same URL with a new verdict. That is how the ledger works.

What we know / what we don't know

What we know (sourced above): the component metrics of the February 2025 Nature paper, with uncertainties; the capital commitments and their dates; that DARPA moved PsiQuantum to the final stage of its Quantum Benchmarking Initiative after hands-on testing; that ground is broken at two sites and a cryoplant is due in 2H-2027.

What we don't know: the state of any internal prototype (DARPA has test access; the public does not); whether the component fidelities survive integration into systems of tens of thousands of chips; any operational date for Brisbane or Chicago, because none is published; the real multiplexing overhead the architecture will pay for probabilistic fusion at scale; and whether foundry yield holds at full volume. We also declare our own limit: Rosetta Q has never run an experiment on photonic hardware, so nothing in our ledger measures these machines directly.

A fairness note: publishing component metrics before any system exists is PsiQuantum's stated strategy — skip intermediate machines, go straight to utility scale — not concealment. "No public system-level benchmark" is a verifiable statement about what is published as of today's date, not a claim about what exists inside the company's labs.