Most quantum companies publish a qubit count. PsiQuantum published a fab process. Omega, announced on February 26, 2025 alongside a Nature paper titled "A manufacturable platform for photonic quantum computing", is a photonic chipset made on a **300-mm** GlobalFoundries line with more than **20** photolithography levels.
That's the whole bet in one sentence: if qubits are photons on silicon photonics chips, then a million-qubit machine is a manufacturing and cooling problem, not a lab problem. This full guide tests that bet against the evidence as of October 15, 2026.
Here's the shape of the piece: what PsiQuantum Omega actually is, the benchmark numbers and their footnotes, the GlobalFoundries and barium titanate supply chain, the Brisbane and Chicago buildouts, the money and government validation, how the plan compares with IBM and Xanadu, and the honest limits. For the other hardware camps, keep Inside Deep Tech's Quantinuum Helios guide and IonQ Tempo guide open.
Last updated: October 15, 2026.
Key Takeaways
- **PsiQuantum Omega** posts **99.98%** single-qubit SPAM, **99.22%** fusion and **99.72%** chip-to-chip fidelity, all conditional on photon detection and excluding loss.
- Omega is built at GlobalFoundries in New York; PsiQuantum grows its own **300mm** barium titanate wafers in California.
- A **$1 billion** Series E at a **$7 billion** valuation (September 2025) funds the Brisbane and Chicago sites.
- Brisbane moved to Moreton Bay, broke ground June 18, 2026, and waits on a Linde cryoplant due 2H 2027.
- Honest limit: 2027 means "operational" (hardware-ready), not a finished computer, and loss is still the open problem.
What PsiQuantum Omega actually is
Omega is a chipset, not a computer. PsiQuantum calls it a quantum photonic chipset purpose-built for utility-scale quantum computing, meaning it contains the component types a fault-tolerant photonic machine needs: photon sources, interferometers, detectors, switches, and chip-to-chip links.
The qubits are single photons in a dual-rail encoding. The Nature paper benchmarks monolithically integrated modules that generate, manipulate, network and detect heralded photonic qubits. The compute model is fusion-based quantum computing (FBQC), where small entangled resource states are joined by measurements called fusions.
Three engineering choices make Omega different from a superconducting or trapped-ion chip:
- Fab-native. The baseline stack was developed with GlobalFoundries and fabricated in its 300-mm high-volume foundry, with hundreds of processing and in-line measurement steps.
- New materials in a standard flow. PsiQuantum added a superconducting material for single-photon detection and barium titanate (BTO) for low-loss, high-speed optical switching.
- Fiber networking. Photonic qubits travel over standard telecom fiber, and PsiQuantum says it has shown high-fidelity interconnects over distances up to **250m**.
That means the chip is only cold where it has to be. MIT Technology Review reports the cabinets operate at **2 K** because the single-photon detectors need it, not because every component does.
The Omega benchmarks, with every footnote attached
The headline numbers are strong. The footnote is stronger. Every figure below is reported in the Nature paper as conditional on photon detection and not accounting for loss.
| Metric (Nature, Feb 2025) | Reported value | What it measures |
|---|---|---|
| Single-qubit SPAM fidelity | **99.98%** ± 0.01% | Prepare and measure one dual-rail qubit |
| Hong-Ou-Mandel visibility | **99.50%** ± 0.25% | Indistinguishability of photons from separate sources |
| Two-qubit fusion fidelity | **99.22%** ± 0.12% | The entangling measurement FBQC is built on |
| Chip-to-chip interconnect | **99.72%** ± 0.04% | Qubit sent between chips over fiber |
| Pump suppression | About **115 dB** | Laser light kept away from the detectors |
| Baseline fiber coupling loss | About **3 dB** (grating couplers) | Light lost getting on and off chip |
Source: PsiQuantum Team, Nature, published February 26, 2025. Fidelities are conditional on detection and exclude loss, as the paper states.
Here's why that footnote matters. In photonics, the dominant error is a photon that never arrives. A fusion that succeeds at 99.22% fidelity when both photons show up says little about how often they don't. The paper says so directly: the baseline platform's performance is still not sufficient for useful photonic quantum computing.
The next-generation components that target loss
The second half of the Nature paper is the more important half for a buyer or investor. It previews the parts meant to close the gap between the baseline and fault tolerance.
- Silicon nitride waveguides: multimode loss as low as **0.5** ± 0.3 dB per meter and single-mode loss of **1.8** dB per meter.
- Splitters and crossings: **0.5** and **1.2** millidecibels of loss, which PsiQuantum says is about two times away from target.
- Chip-to-fiber edge couplers: **127** mdB to SMF-28 fiber and **52** mdB to high-numerical-aperture fiber, against roughly 3 dB on the baseline grating couplers.
- Photon-number-resolving detectors: **98.9%** median single-photon efficiency with up to four-photon resolution.
- BTO switches: a loss-voltage product of **0.33** dB·V, aimed at the fast switching needed to multiplex probabilistic sources.
That means the roadmap is specific and checkable. Component losses are reported with error bars. Waveguide losses are on target. Splitters and crossings are not yet. The open question is whether all of these improvements land on the same wafer, at yield, at the same time.
For the classical cousin of this technology, Inside Deep Tech's silicon photonics for AI data centers full guide covers the same foundry ecosystem from the interconnect side.
The GlobalFoundries and barium titanate supply chain
PsiQuantum's chips are designed in-house and manufactured at GlobalFoundries' Fab 8 in New York, a high-volume commercial foundry. The two companies have partnered on quantum photonic chipset production since 2019.
The unusual part is the switch material. PsiQuantum manufactures **300mm** BTO wafers at its own California facilities, which are then integrated with wafers from GlobalFoundries. MIT Technology Review describes the in-house BTO decision as an expensive, agonizing call because the material was not available at scale.
The scale claims are about measurement as much as chips. PsiQuantum says it has characterized millions of devices on thousands of wafers and performs around half a million measurements each month.
On the flip side, owning BTO adds a single-supplier dependency inside PsiQuantum itself. The September 2026 **$100 million** CHIPS Act award from the U.S. Department of Commerce targets exactly these parts: optical switches, higher-temperature single-photon detectors, and advanced packaging.
Cabinets, not chandeliers: how Omega gets cooled
PsiQuantum dropped the dilution-refrigerator "chandelier" for a cuboid cabinet closer to a data center server rack, first deployed at its UK facility at STFC Daresbury. The company says each cabinet can cool hundreds of quantum chips.
The test path runs through three U.S. sites. PsiQuantum first siphoned liquid helium from SLAC's cryoplant in Menlo Park, then built its own system in Milpitas. As of July 2026, Milpitas had connected three cabinets with **250** chips each.
A commercially useful machine is expected to need roughly **100** cabinets connected together. So the honest progress metric today is three of about one hundred, with error correction still to be shown across them.
For the error-correction background behind those cabinets, see Inside Deep Tech's fault-tolerant quantum computing explainer and the 2026 guide to logical qubits.
Brisbane: from airport site to Moreton Bay Central
The Australian and Queensland governments committed almost **A$1 billion** to PsiQuantum to build a fault-tolerant quantum computer in Brisbane, with up to **400** ongoing jobs.
The site changed. On May 21, 2026, PsiQuantum said it would anchor the project at Moreton Bay Central, the former Petrie Paper Mill site, instead of the Brisbane Airport location announced in 2024. Groundbreaking followed on June 18, 2026, with new CEO Victor Peng and co-founder Jeremy O'Brien on site.
The critical path is cold, not chips. A Linde Engineering cryoplant ordered in late 2024 is set for delivery in the second half of 2027. After commissioning, the site starts accepting cabinets filled with photonic chips and networked by fiber. A Test and Validation Lab at Griffith University opened in May 2026 to test chips and subsystems in the meantime.
Chicago: the IQMP anchor and the DARPA test system
PsiQuantum broke ground on September 30, 2025 at the Illinois Quantum and Microelectronics Park on the former U.S. Steel South Works site, as anchor tenant. The first phase is not the million-qubit machine. It's PsiQuantum's largest intermediate-scale test system, to be evaluated directly by DARPA's experts.
Construction is visible. By June 29, 2026, the shell of the **80,000**-square-foot building was complete, with Clayco crews expected to spend the next year on fit-out. That puts usable Chicago space around mid-2027 on the reported schedule.
| Site | Role | Status (as of Oct 15, 2026) | Next gate |
|---|---|---|---|
| Milpitas, California | Multi-cabinet test (PsiFactory) | 3 cabinets x **250** chips connected | Error correction across cabinets |
| Moreton Bay, Queensland | First utility-scale system | Broke ground June 18, 2026 | Cryoplant delivery 2H 2027 |
| Chicago, Illinois (IQMP) | Intermediate test system, then full system | Shell complete, **80,000** sq ft | Fit-out through about mid-2027 (IQMP) |
| Malta, New York | Chip fab (GlobalFoundries) | Omega in production flow | Next-gen SiN, PNRD and BTO at yield |
| Brisbane (Griffith Univ.) | Test and Validation Lab | Opened May 2026 | Subsystem qualification |
Source: PsiQuantum releases on Moreton Bay, Chicago and funding; IQMP / Crain's Chicago Business, June 29, 2026; MIT Technology Review, July 14, 2026.
Money and validation: $1B, DARPA Stage C, and CHIPS
The capital stack is large for a pre-revenue hardware company. The September 10, 2025 Series E raised **$1 billion** at a **$7 billion** valuation, led by BlackRock, Temasek and Baillie Gifford, with NVIDIA's NVentures among new investors. PsiQuantum also announced a collaboration with NVIDIA on algorithms, GPU-QPU integration and its silicon photonics platform.
Government validation is arguably worth more than the cash. On July 22, 2026, PsiQuantum signed a **$125 million** expanded agreement with DARPA under the Quantum Benchmarking Initiative, as one of two companies in Stage C. It follows a **$31.8 million** Stage C agreement from September 2025. MIT Technology Review names Microsoft as the other Stage C company.
DARPA's QBI director Micah Stoutimore has said it now seems likely that someone will build a utility-scale quantum computer by 2033. Note the word someone. Stage C is a rigorous evaluation, not a delivery guarantee.
Then there's domestic supply. Beyond the **$100 million** Commerce award finalized September 8, 2026, PsiQuantum says it spent approximately **$200 million** with American suppliers across **38** states in 2025.
How PsiQuantum Omega compares with IBM and Xanadu
PsiQuantum is not the only company with a dated fault-tolerance plan. IBM says Starling will be delivered by 2029 in Poughkeepsie, running **100 million** quantum operations on **200** logical qubits. MIT Technology Review notes IBM's earlier large-scale target was 2028 and appears to have moved to 2030; IBM's own June 2025 release says 2029.
Within photonics, Xanadu took a different encoding and published a networked system first. Its Aurora machine used **35** photonic chips, **84** squeezers and **36** photon-number-resolving detectors to synthesize a cluster state with **86.4 billion** modes. Xanadu itself calls Aurora a sub-performant scale model and names optical loss as the dominant hurdle to fault tolerance.
| Program | Qubit type | Public proof point | Dated target |
|---|---|---|---|
| PsiQuantum Omega | Photonic (dual-rail, FBQC) | **99.72%** chip-to-chip fidelity, excl. loss | Brisbane "operational" end of 2027 |
| IBM Starling | Superconducting (qLDPC) | **200** logical qubits planned | By 2029 |
| Xanadu Aurora | Photonic (continuous-variable, GKP) | **35** chips networked over fiber | No dated FTQC system in the paper |
| Quantinuum Helios | Trapped ion | 98 qubits (IDT guide) | See Helios guide |
Source: Nature 2025 (PsiQuantum), IBM newsroom, June 10, 2025, Nature 2025 (Xanadu Aurora), MIT Technology Review, and Inside Deep Tech's Helios and IBM roadmap coverage.
That means the comparison is not apples to apples. IBM and Quantinuum publish incremental system results. PsiQuantum publishes components and builds facilities. Scott Aaronson told MIT Technology Review that PsiQuantum is very hard for an outsider to evaluate.
Inside Deep Tech's take
Inside Deep Tech's take: PsiQuantum Omega proves the fab half of the thesis. Chips with four benchmark figures above **99%** (before loss), a tier-1 foundry, millions of measured devices, and fiber links all exist on paper and in Milpitas.
The systems half is unproven. Nobody outside PsiQuantum and DARPA has seen error correction run across networked cabinets. The Brisbane cryoplant date and the Chicago fit-out set the earliest window for that proof to around late 2027 at best.
For you, the practical stance is simple. If you're building quantum algorithm capability, PsiQuantum's software and partner programs are a reasonable place to start. If you're underwriting a 2028 production workload, don't.
Honest limits
First, loss. The Omega fidelities exclude loss, and the paper says the baseline is not yet sufficient for useful computing. Splitter and crossing losses are still about two times from target.
Second, timelines. PsiQuantum says the 2027 Australian milestone means the facility is operational and hardware-ready, with no promise about computer size. Some news coverage reported 2027 as the date for a full-scale computer; the company calls that a misreading.
Third, opacity. PsiQuantum publishes fewer system-level results than IBM, Google or Quantinuum. DARPA's Stage C access is deep but mostly private.
Fourth, cryogenic capex. MIT Technology Review reports that helium cooling systems will consume a significant chunk of the **$1 billion** round, and the Brisbane site moved once already.
Fifth, applications. An Airbus fluid-dynamics study showed only a moderate speedup, which PsiQuantum's own applications VP agreed was modest. The killer apps still wait on the full machine.
FAQ
What is PsiQuantum Omega?
It's PsiQuantum's photonic quantum chipset, announced February 26, 2025 with a Nature paper. It integrates photon sources, interferometers, superconducting detectors, BTO switches and chip-to-chip links, made at GlobalFoundries.
How many qubits does PsiQuantum Omega have?
Omega isn't sold as a fixed-qubit processor. It's the component set for a million-qubit-scale machine. The near-term metric to watch is cabinets: as of July 2026, Milpitas had three with **250** chips each, and a useful machine needs about **100**.
When will PsiQuantum have a working quantum computer?
No dated full-computer promise exists. The Brisbane site aims to be operational, meaning hardware-ready, by the end of 2027, after a cryoplant arrives in 2H 2027.
Where are PsiQuantum's quantum computers being built?
At Moreton Bay Central in Queensland and the Illinois Quantum and Microelectronics Park in Chicago. Chips come from GlobalFoundries in New York; multi-cabinet testing runs in Milpitas.
How much funding has PsiQuantum raised?
Its latest round was a **$1 billion** Series E at a **$7 billion** valuation in September 2025. Government support includes almost **A$1 billion** from Australia and Queensland, a **$125 million** DARPA agreement and a **$100 million** Commerce award.
Is photonic quantum computing better than superconducting?
It trades different problems. Photonics gets fab manufacturing and fiber networking, but loss is the dominant hurdle. Superconducting players like IBM target **200** logical qubits by 2029. See the IBM roadmap guide.
What to watch next
Three signals will tell you whether Omega becomes a computer. First, a published multi-cabinet error-correction result from Milpitas or Chicago. Second, the Linde cryoplant arriving at Moreton Bay on its 2H 2027 schedule. Third, next-generation silicon nitride, PNRD and BTO parts reported together on one production wafer.
Until then, compare roadmaps with Inside Deep Tech's IBM quantum roadmap guide, the Quantinuum Helios guide and the IonQ Tempo guide. For more frontier coverage, return to the Inside Deep Tech homepage.
