Pinned Tweet
$IONQ Q3 revenue hit $173.8M. But how much becomes shareholder value?
My new Article examines what's behind the growth, what it costs, and whether machines can build a valuable service business. x.com/i/article/2118…
My new Article examines what's behind the growth, what it costs, and whether machines can build a valuable service business. x.com/i/article/2118…
0
0
0
149
Does quantum advantage depend on the structure of the problem?
The Aaronson-Ambainis conjecture says that for certain problems, exponential quantum speedups require special structure.
PSI claim they proved it, using AI.
This was actually one of the first questions I asked mputing.
Does quantum advantage depend on the structure of the problem, or could quantum computers also outperform existing AI solutions on tasks that AI currently dominates?
The new proof supports the former, at least for the kinds of problems studied in the paper.
There are caveats. It does not mean that every quantum application can be matched efficiently by classical computers.
The proof looks at how many times a computer needs to read the data, not how long the computation takes.
But as an investor, I ask a different question.
What does this mean for the potential market size of quantum computing?
I plan to focus part of my research on finding out.
The Aaronson-Ambainis conjecture says that for certain problems, exponential quantum speedups require special structure.
PSI claim they proved it, using AI.
This was actually one of the first questions I asked mputing.
Does quantum advantage depend on the structure of the problem, or could quantum computers also outperform existing AI solutions on tasks that AI currently dominates?
The new proof supports the former, at least for the kinds of problems studied in the paper.
There are caveats. It does not mean that every quantum application can be matched efficiently by classical computers.
The proof looks at how many times a computer needs to read the data, not how long the computation takes.
But as an investor, I ask a different question.
What does this mean for the potential market size of quantum computing?
I plan to focus part of my research on finding out.
Today Physical Superintelligence PBC is announcing a proof of the Aaronson-Ambainis conjecture, one of the central open problems in quantum complexity, with major implications for quantum computing speedup.
psi.inc/research/aaron…
Thread 👇
0
0
3
187
Beijing E-Town is backing quantum R&D, cloud platforms, and pilot projects.
The goal is to move quantum technology from research to real use.
It looks like a continuation of its 2025 policy.
As an investor, I'm curious which sectors will benefit.
beijing.gov.cn/zhengce/zhengc…
The goal is to move quantum technology from research to real use.
It looks like a continuation of its 2025 policy.
As an investor, I'm curious which sectors will benefit.
beijing.gov.cn/zhengce/zhengc…
0
0
0
39
PsiQuantum builds on single photons and leans on fusion measurements that only succeed some of the time. The optical GKP design proposed by $XNDU goes another way.
Rather than putting a qubit in a single photon, GKP stores it in a carefully structured state of light.
With hand, Clifford operations can run on linear optics and homodyne detection, with no probabilistic fusion involved.
The catch is in preparing those states. Producing them still hinges on detector outcomes, and optical loss wears down their quality.
So the fidelity question shifts.
The issue is not only whether an operation succeeded. It is whether the prepared grid stays sharp enough, after loss and noise, to support error correction.
In Aurora's 2025 error-detection experiment, the states were rough two-peak approximations, and the refineries did not carry out GKP breeding. The experiment was too noisy to show error suppression.
GKP does not make the loss problem go away. It moves the place where the architecture has to solve it.
The next interesting question is how close Xanadu is to preparing states that are good enough.
Part of my deep dive on photonic quantum computing.
Rather than putting a qubit in a single photon, GKP stores it in a carefully structured state of light.
With hand, Clifford operations can run on linear optics and homodyne detection, with no probabilistic fusion involved.
The catch is in preparing those states. Producing them still hinges on detector outcomes, and optical loss wears down their quality.
So the fidelity question shifts.
The issue is not only whether an operation succeeded. It is whether the prepared grid stays sharp enough, after loss and noise, to support error correction.
In Aurora's 2025 error-detection experiment, the states were rough two-peak approximations, and the refineries did not carry out GKP breeding. The experiment was too noisy to show error suppression.
GKP does not make the loss problem go away. It moves the place where the architecture has to solve it.
The next interesting question is how close Xanadu is to preparing states that are good enough.
Part of my deep dive on photonic quantum computing.
Photons hold their quantum state better than almost any qubit, and barely talk to each other.
That trade is the whole bet behind photonic quantum computing.
$XNDU and $QUBT are public, while PsiQuantum, Quandela, QuiX and ORCA are still private.
The first half is why light is tempting. Photons stay quantum at room temperature, though the detectors that count them still need cooling. They travel between chips through ordinary fiber, and the chips can come out of the same foundries that make telecom parts.
The second half is the bill. Photons get lost along the way, and because they barely interact, the standard way to entangle two of them works only some of the time.
The basic blueprint for working around that dates back to 2001.
Xanadu builds on squeezed light, PsiQuantum fuses small entangled groups, and Quantum Computing Inc goes another way, with photonic chips and a machine aimed at optimization problems rather than general computing.
That gap between hardware that existing foundries can build and entanglement that is hard to get is what makes the field interesting from an investor perspective.
So for this deep dive, I want to understand how each approach works, which hurdles new papers are actually clearing, and how far the field still is from a useful machine.
1
0
0
213
Primary papers:
PsiQuantum's fusion-based architecture: nature.com/articles/s4146…
Xanadu's Aurora architecture and error-detection experiment: doi.org/10.1038/s41586…
Xanadu's 2025 optical GKP source: doi.org/10.1038/s41586…
PsiQuantum's fusion-based architecture: nature.com/articles/s4146…
Xanadu's Aurora architecture and error-detection experiment: doi.org/10.1038/s41586…
Xanadu's 2025 optical GKP source: doi.org/10.1038/s41586…
1
0
0
55
The explanation behind this post: what a GKP state is, why ideal ones are physically impossible, and what the hardware has actually shown:
quantuminvestornotebook.com/Notes/what-is-…
quantuminvestornotebook.com/Notes/what-is-…
0
0
0
31
$IONQ more than tripled revenue in Q3 and has raised billions to keep building. What exactly is behind that growth, and what does it mean for shareholders?
I dug into the demand, acquisitions, dilution and the economics of an installed machine base in my next X Article.
I dug into the demand, acquisitions, dilution and the economics of an installed machine base in my next X Article.
Our Q3 2026 Financial Results: investors.ionq.com/news/news-deta…
0
0
4
313
$IBM decided to take the Department of Energy benchmark seriously.
Their own scoreboard against the late-2028 targets now shows 70 of the 100 error-protected qubits, but only 438 of the 100,000 hard operations.
That's about 230x short on the hard part.
In DOE's terms, a hard operation is the scarce resource in a fault-tolerant computation, one that needs many elementary gates and is hardest to run cleanly.
So IBM, ORNL and $AMD put a regular computer next to the quantum one. IBM's 120-qubit Nighthawk now sits beside Umbra, ORNL's machine on AMD GPUs, so classical decoding can help close that gap.
For $IBM investors, hard operations just became the number that matters.
Their own scoreboard against the late-2028 targets now shows 70 of the 100 error-protected qubits, but only 438 of the 100,000 hard operations.
That's about 230x short on the hard part.
In DOE's terms, a hard operation is the scarce resource in a fault-tolerant computation, one that needs many elementary gates and is hardest to run cleanly.
So IBM, ORNL and $AMD put a regular computer next to the quantum one. IBM's 120-qubit Nighthawk now sits beside Umbra, ORNL's machine on AMD GPUs, so classical decoding can help close that gap.
For $IBM investors, hard operations just became the number that matters.
The @QuantumSciCtr at @ORNL, @IBM, and @AMD deploy a new co-design platform for quantum error correction at IBM’s site in Poughkeepsie, New York.
The platform enables low-latency quantum-classical workflows for advancing quantum computing research: ibm.co/6012EV4lM
0
0
2
277
A photon barely notices the world around it. The chips, fibers and detectors around it notice everything.
For photonic bets like $XNDU, that machinery is tested in three ways. SPAM, HOM and fusion.
SPAM is state preparation and measurement. Did the chip put the qubit into the e detector read that state back correctly.
HOM is Hong Ou Mandel interference. Two photons from different sources meet on a beam splitter. If they are truly alike, they leave together.
Fusion is how a photonic machine stitches small entangled pieces into a bigger one. Each fusion is itself a detection event.
The fine print: many of those 99% figures only count runs where a photon was detected, and they do not count loss.
PsiQuantum's February 2025 Nature foundry paper is a clear example. Every headline fidelity there is conditional on detection and does not count loss.
So when you see 99% on a photonic slide, always look for the fine print.
Part of my deep dive on photonic quantum computing.
For photonic bets like $XNDU, that machinery is tested in three ways. SPAM, HOM and fusion.
SPAM is state preparation and measurement. Did the chip put the qubit into the e detector read that state back correctly.
HOM is Hong Ou Mandel interference. Two photons from different sources meet on a beam splitter. If they are truly alike, they leave together.
Fusion is how a photonic machine stitches small entangled pieces into a bigger one. Each fusion is itself a detection event.
The fine print: many of those 99% figures only count runs where a photon was detected, and they do not count loss.
PsiQuantum's February 2025 Nature foundry paper is a clear example. Every headline fidelity there is conditional on detection and does not count loss.
So when you see 99% on a photonic slide, always look for the fine print.
Part of my deep dive on photonic quantum computing.
Photons hold their quantum state better than almost any qubit, and barely talk to each other.
That trade is the whole bet behind photonic quantum computing.
$XNDU and $QUBT are public, while PsiQuantum, Quandela, QuiX and ORCA are still private.
The first half is why light is tempting. Photons stay quantum at room temperature, though the detectors that count them still need cooling. They travel between chips through ordinary fiber, and the chips can come out of the same foundries that make telecom parts.
The second half is the bill. Photons get lost along the way, and because they barely interact, the standard way to entangle two of them works only some of the time.
The basic blueprint for working around that dates back to 2001.
Xanadu builds on squeezed light, PsiQuantum fuses small entangled groups, and Quantum Computing Inc goes another way, with photonic chips and a machine aimed at optimization problems rather than general computing.
That gap between hardware that existing foundries can build and entanglement that is hard to get is what makes the field interesting from an investor perspective.
So for this deep dive, I want to understand how each approach works, which hurdles new papers are actually clearing, and how far the field still is from a useful machine.
2
0
7
548
Sources:
PsiQuantum team, Nature (26 Feb 2025), open access: nature.com/articles/s4158…
Open pre-print version (arXiv 2404.17570): arxiv.org/abs/2404.17570
PsiQuantum team, Nature (26 Feb 2025), open access: nature.com/articles/s4158…
Open pre-print version (arXiv 2404.17570): arxiv.org/abs/2404.17570
1
0
1
40
$QNT put a multi-year Nexus roadmap on earnings day.
Same idea I've been watching at $IONQ. Own the software and the workloads around the machine, so every hardware step makes leaving harder.
Nexus is that bet.
Developer platform tied to Helios today through Apollo in 2029. Nexus already turns a program into an error-corrected one and runs it on Helios.
December is expected to widen hybrid workflows.
More than 200 organizations already use it, up 45% in 2026. Job submissions rose 10x in the 12 months through August.
The question is whether those 200 orgs stick long enough until Apollo finally shows up.
Same idea I've been watching at $IONQ. Own the software and the workloads around the machine, so every hardware step makes leaving harder.
Nexus is that bet.
Developer platform tied to Helios today through Apollo in 2029. Nexus already turns a program into an error-corrected one and runs it on Helios.
December is expected to widen hybrid workflows.
More than 200 organizations already use it, up 45% in 2026. Job submissions rose 10x in the 12 months through August.
The question is whether those 200 orgs stick long enough until Apollo finally shows up.
The next era of quantum computing needs more than extraordinary, fault-tolerant hardware. It needs a platform that puts that power in developers’ hands.
Meet the future of Quantinuum Nexus.
Today, we’re unveiling a multi-year roadmap to evolve Nexus, Quantinuum’s platform as a try’s first developer platform designed for the fault-tolerant quantum era.
Nexus will connect application development and deployment across our hardware roadmap, from Helios to Apollo, our universal, fault-tolerant quantum computer planned for 2029.
Our ambition is simple: make building powerful quantum applications easier as we progress toward fault tolerance. Nexus will progressively handle the complexity of error correction, helping developers build programs that grow in capability as the hardware advances.
With Nexus, you can:
🚀 Develop complete applications with fault-tolerant libraries and tools.
⚡ Optimize performance with increasingly automated error correction and resource management.
🌐 Deploy hybrid quantum-classical workflows across enterprise computing environments.
Build today. Be ready for what’s next.
Learn more and explore the roadmap: quantinuum.com/press-releases…
0
0
2
654
$IONQ closed the Elisity deal yesterday, paying $1.15 billion in new stock, just before their Q3 filing.
Will investors get the chance to see what that stock bought?
Sellers can get up to $150 million more, only if Elisity hits revenue milestones through January 2028.
But investors aim higher. Elisity controls which users and devices can talk to which on a network, and IonQ plans to sell its post-quantum security on top.
What will be disclosed about it today after close is what I will be watching.
Will investors get the chance to see what that stock bought?
Sellers can get up to $150 million more, only if Elisity hits revenue milestones through January 2028.
But investors aim higher. Elisity controls which users and devices can talk to which on a network, and IonQ plans to sell its post-quantum security on top.
What will be disclosed about it today after close is what I will be watching.
Today, IonQ completed its acquisition of @ElisityInc.
Security has always been about protecting what matters most, and quantum makes that more urgent. Attackers are already collecting encrypted data today so they can decrypt it later, and the networks we run now will need to be e-scale quantum machines arrive.
Elisity brings Zero Trust capabilities that control who and what can access critical systems, along with the ability to see and manage cryptographic risk. Combined with our work in quantum computing, networking, and sensing, this gives customers a clearer path to securing their environments today and preparing for the quantum era.
Welcome to the IonQ team, Elisity!
Read the full announcement: ionq.com/news/ionq-comp…
0
0
5
680
$IONQ and Classiq make a fair point.
Magic states got much cheaper, so memory, routing and layout can now make up much of the cost. Counting T gates alone can mislead.
They credit earlier work on cheaper magic states, but the effect itself isn't new either.
FLASQ (2025) and Wang et al. (2026) already found cases where fewer T gates didn't mean lower total cost. The blog doesn't cite them.
Classiq's 41% comes from a single circuit, and they haven't published the full details.
Once a paper with the full details is out, I'll be able to compare it and say what this work adds over earlier studies.
Magic states got much cheaper, so memory, routing and layout can now make up much of the cost. Counting T gates alone can mislead.
They credit earlier work on cheaper magic states, but the effect itself isn't new either.
FLASQ (2025) and Wang et al. (2026) already found cases where fewer T gates didn't mean lower total cost. The blog doesn't cite them.
Classiq's 41% comes from a single circuit, and they haven't published the full details.
Once a paper with the full details is out, I'll be able to compare it and say what this work adds over earlier studies.
Optimizing for fewer T-gates can actually make your quantum program slower.
For decades, the quantum industry assumed magic states were the ultimate bottleneck, consuming over 90% of fault-tolerant resources. But the math has changed.
Recent benchmarks show that cutting T-gates inflate total physical spacetime volume by 41% due to routing, layout, and memory overhead.
In our latest joint paper with Classiq, IonQ’s Zeljko Susnjar and Zheng Cui, Ph.D. break down why full-stack accounting - not isolated gate counts - must define the next generation of fault-tolerant quantum computing.
Read the full paper to see what this means for enterprise roadmaps: ionq.com/blog/from-magi…
#QuantumComputing #FTQC #QuantumHardware #IonQ #MagicStates
0
2
11
945
A record GHZ state of more than a dozen entangled photons from one quantum dot.
This is the claim this paper makes.
But the 13 is an estimate of how fast the entanglement fades. The largest entangled state actually measured had 2 photons.
For $XNDU and other photonic bets, the umbers is the story. And it comes from two problems.
The first is engineering. The setup detects so few of the photons that only 4 in a row could be registered at a usable rate.
That loss isn't only a measurement problem. Photons also get lost while traveling, and a photonic computer needs its photons to arrive.
The second is physics. Photons barely decohere, but the electron spin of the quantum dot does. Each photon takes 0.4 ns to leave the dot, and the spin gets a bit weaker each time.
Stretching the spin coherence is still a challenge under research.
The best current result I was able to find is 14 photons in a GHZ state, created from one rubidium atom in 2022, with fidelity of at least 76%.
Will improvements come from better photonic circuits, from faster emission, or from a more stable quantum dot?
Engineering advances in photonic circuits will be the thing to watch for me.
Part of my deep dive on photonic quantum computing.
This is the claim this paper makes.
But the 13 is an estimate of how fast the entanglement fades. The largest entangled state actually measured had 2 photons.
For $XNDU and other photonic bets, the umbers is the story. And it comes from two problems.
The first is engineering. The setup detects so few of the photons that only 4 in a row could be registered at a usable rate.
That loss isn't only a measurement problem. Photons also get lost while traveling, and a photonic computer needs its photons to arrive.
The second is physics. Photons barely decohere, but the electron spin of the quantum dot does. Each photon takes 0.4 ns to leave the dot, and the spin gets a bit weaker each time.
Stretching the spin coherence is still a challenge under research.
The best current result I was able to find is 14 photons in a GHZ state, created from one rubidium atom in 2022, with fidelity of at least 76%.
Will improvements come from better photonic circuits, from faster emission, or from a more stable quantum dot?
Engineering advances in photonic circuits will be the thing to watch for me.
Part of my deep dive on photonic quantum computing.
Photons hold their quantum state better than almost any qubit, and barely talk to each other.
That trade is the whole bet behind photonic quantum computing.
$XNDU and $QUBT are public, while PsiQuantum, Quandela, QuiX and ORCA are still private.
The first half is why light is tempting. Photons stay quantum at room temperature, though the detectors that count them still need cooling. They travel between chips through ordinary fiber, and the chips can come out of the same foundries that make telecom parts.
The second half is the bill. Photons get lost along the way, and because they barely interact, the standard way to entangle two of them works only some of the time.
The basic blueprint for working around that dates back to 2001.
Xanadu builds on squeezed light, PsiQuantum fuses small entangled groups, and Quantum Computing Inc goes another way, with photonic chips and a machine aimed at optimization problems rather than general computing.
That gap between hardware that existing foundries can build and entanglement that is hard to get is what makes the field interesting from an investor perspective.
So for this deep dive, I want to understand how each approach works, which hurdles new papers are actually clearing, and how far the field still is from a useful machine.
3
0
6
351
Su et al., arXiv preprint, arxiv.org/abs/2610.11445
Thomas et al., Nature 608, 677 (2022), arxiv.org/abs/2205.12736
Thomas et al., Nature 608, 677 (2022), arxiv.org/abs/2205.12736
1
0
1
35
Quantum computers are usually sold by the dream. Connecticut's quantum initiative is buying $IONQ's by the hour, more than 150 of them, for its universities, labs and startups.
The question for me is whether hour-based deals like this grow into stickier revenue. I still think they should evolve into either a service business or a platform.
I lean towards the latter, as the deal also covers workforce development, so today's students could become tomorrow's bigger users.
Q3 comes tomorrow after the close. Will we see any sign of which way it's heading?
x.com/QInvestorNotes…
The question for me is whether hour-based deals like this grow into stickier revenue. I still think they should evolve into either a service business or a platform.
I lean towards the latter, as the deal also covers workforce development, so today's students could become tomorrow's bigger users.
Q3 comes tomorrow after the close. Will we see any sign of which way it's heading?
x.com/QInvestorNotes…
We’re partnering with @CTquantum, Connecticut’s statewide initiative anchored by @UConn and @Yale, to accelerate quantum innovation and expand access to our trapped-ion quantum computers.
The agreement includes more than 150 hours of cloud runtime for Connecticut’s universities, , and enterprises, advancing quantum algorithms, optimization, materials science, and workforce development.
“Regional quantum ecosystems thrive when researchers and innovators have access to advanced quantum computers,” said Scott Millard, IonQ's Chief Business Officer.
Together, we're taking Connecticut’s quantum community from exploration to application.
Learn more: ionq.com/news/ionq-anno…
#IonQ #QuantumInnovation #QuantumIsNow
0
1
9
662
D-Wave says it upgraded its Stride solver.
For $QBTS holders, the footnote matters more than the headline.
The numbers are real. What they compare is narrow.
On traveling salesperson problems with up to 10,000 locations, the new Stride averaged within 0.02% of the best possible result in 15 seconds, on average more than 200x closer than the old version at the same runtime.
A big jump, and the 0.02% is measured against answers another solver proved optimal, so it isn't only a grade against itself.
But the footnote is clear. "Improvement factors compare enhanced Stride solver performance with previous Stride solver performance at the same runtime."
And Stride is hybrid. How much of the gain came from the quantum processor, and how much from a better classical algorithm? The release says D-Wave "enhanced the core algorithm" but doesn't split it.
If it is mostly the classical side, it is still a better product. It is just a different story than a quantum advance.
For value, the test is simpler. Will it bring in new customers, or get existing ones to pay more?
For $QBTS holders, the footnote matters more than the headline.
The numbers are real. What they compare is narrow.
On traveling salesperson problems with up to 10,000 locations, the new Stride averaged within 0.02% of the best possible result in 15 seconds, on average more than 200x closer than the old version at the same runtime.
A big jump, and the 0.02% is measured against answers another solver proved optimal, so it isn't only a grade against itself.
But the footnote is clear. "Improvement factors compare enhanced Stride solver performance with previous Stride solver performance at the same runtime."
And Stride is hybrid. How much of the gain came from the quantum processor, and how much from a better classical algorithm? The release says D-Wave "enhanced the core algorithm" but doesn't split it.
If it is mostly the classical side, it is still a better product. It is just a different story than a quantum advance.
For value, the test is simpler. Will it bring in new customers, or get existing ones to pay more?
Today, D-Wave announced powerful performance enhancements to its Stride™ hybrid solver. Combining quantum and classical computing resources, the updated Stride hybrid solver, now available in the Leap™ quantum cloud service, helps users quickly find high-quality solutions to large optimization problems that can be difficult for classical computing alone to address. In key benchmark testing, the enhanced solver found better solutions in 15 seconds than the previous version found in 120 seconds.
Additionally, D-Wave has also launched a new, three-course Applied Quantum Optimization learning path to accelerate hybrid-quantum optimization adoption. Together, the enhanced solver and training courses further equip developers to apply quantum optimization to complex, real-world business challenges.
Learn more: dwavequantum.com/company/newsro…
0
0
2
162
$XNDU and ASML are exploring how better lithography could reduce light loss in photonic quantum chips.
Tiny irregularities along a chip’s structures can scatter light. Better control over those patterns could help more photons survive.
The broader manufacturing challenge matters for quantum sensing too.
Companies such as $INFQ are working to integrate the optics around their sensors onto chips, helping make those systems smaller.
This Xanadu-ASML collaboration was announced in September, and it is still exploratory.
I’m looking forward to seeing what improvements come from it.
Tiny irregularities along a chip’s structures can scatter light. Better control over those patterns could help more photons survive.
The broader manufacturing challenge matters for quantum sensing too.
Companies such as $INFQ are working to integrate the optics around their sensors onto chips, helping make those systems smaller.
This Xanadu-ASML collaboration was announced in September, and it is still exploratory.
I’m looking forward to seeing what improvements come from it.
Xanadu recently joined forces with semiconductor leader ASML to tackle optical loss with next-gen lithography—fast-tracking the path to utility-scale quantum computing.
Learn more: investors.xanadu.ai/news-releases/…
0
3
15
757
Photonic quantum computers, like $XNDU's and PsiQuantum's, run on many entangled photons, and a source that makes them already linked could save a lot of work.
A GHZ state might be part of the answer.
In September, a paper had $IONQ, $IQMX and $RGTI each try to hold 10 qubits one stayed above 0.5, the line that proves real group entanglement.
This recent Technion study creates the same kind of state from light qubits.
Send photons through a polarizer and they all come out the same: all horizontal, or all vertical. Impressive? Not yet. Photons set up beforehand could do that too.
Turn the polarizer to a tilted angle, and those come out at random. Photons in a GHZ state still match. That's how you know.
The authors ran this test on only 2 photons. It's a long way from 2 to many, but it's a good start.
Part of my deep dive into photonic quantum computing.
A GHZ state might be part of the answer.
In September, a paper had $IONQ, $IQMX and $RGTI each try to hold 10 qubits one stayed above 0.5, the line that proves real group entanglement.
This recent Technion study creates the same kind of state from light qubits.
Send photons through a polarizer and they all come out the same: all horizontal, or all vertical. Impressive? Not yet. Photons set up beforehand could do that too.
Turn the polarizer to a tilted angle, and those come out at random. Photons in a GHZ state still match. That's how you know.
The authors ran this test on only 2 photons. It's a long way from 2 to many, but it's a good start.
Part of my deep dive into photonic quantum computing.
Photons hold their quantum state better than almost any qubit, and barely talk to each other.
That trade is the whole bet behind photonic quantum computing.
$XNDU and $QUBT are public, while PsiQuantum, Quandela, QuiX and ORCA are still private.
The first half is why light is tempting. Photons stay quantum at room temperature, though the detectors that count them still need cooling. They travel between chips through ordinary fiber, and the chips can come out of the same foundries that make telecom parts.
The second half is the bill. Photons get lost along the way, and because they barely interact, the standard way to entangle two of them works only some of the time.
The basic blueprint for working around that dates back to 2001.
Xanadu builds on squeezed light, PsiQuantum fuses small entangled groups, and Quantum Computing Inc goes another way, with photonic chips and a machine aimed at optimization problems rather than general computing.
That gap between hardware that existing foundries can build and entanglement that is hard to get is what makes the field interesting from an investor perspective.
So for this deep dive, I want to understand how each approach works, which hurdles new papers are actually clearing, and how far the field still is from a useful machine.
1
0
10
692
Sources:
Su et al. (Technion and partners), arXiv preprint, Oct 8 2026: arxiv.org/abs/2610.11445
My September GHZ post: x.com/QInvestorNotes…
More on GHZ in our Note: quantuminvestornotebook.com/Notes/what-is-…
Su et al. (Technion and partners), arXiv preprint, Oct 8 2026: arxiv.org/abs/2610.11445
My September GHZ post: x.com/QInvestorNotes…
More on GHZ in our Note: quantuminvestornotebook.com/Notes/what-is-…
Quantum computers have to do more than entangle two qubits. What happens when the benchmark grows to six or ten?
Apparently, it makes a difference.
This paper tests 6- and 10-qubit GHZ states on $IONQ Aria-1, $IQMX Garnet and $RGTI Ankaa-3.
A GHZ circuit starts with one qubit and has to pull all the others into the same shared quantum state. In the paper’s version, that means a series of entangling gates, each one running from the first qubit to one of the others.
That is where connectivity comes in.
Aria-1 can entangle any pair of qubits directly. Garnet and Ankaa-3 have restricted connectivity, so a gate between distant qubits may need extra routing steps first.
Every extra gate is one more chance for an error to creep in.
At 6 qubits, the paper reports GHZ fidelities of about 0.96 for Aria-1, 0.68 for Garnet and 0.07 for Ankaa-3.
At 10 qubits: about 0.84, 0.12 and 0.09.
On this benchmark, Aria-1 is clearly ahead.
I will be honest, though: connectivity is not the whole story. Garnet and Ankaa-3 both have restricted connectivity, and at 6 qubits their reported fidelities are dramatically different.
The paper attributes the superconducting drop partly to routing and accumulated two-qubit errors, but it does not give the per-device compiled GHZ gate counts needed to separate those effects cleanly.
And GHZ remains a very particular test. It stresses multi-qubit coherence, entangling-gate quality and connectivity, all inside one simple, regular circuit.
Real algorithms ask for other things: different interaction patterns, far more depth, measurements, classical feedback.
So GHZ crowns a winner for this workload, not for quantum computing.
1
0
1
54