Appendix A · Different leads, no overall winner
The September architecture comparison, including its underlying references.
QUANTUM LITMUS / SPOTLIGHT #1 / 1 SEPTEMBER 2026
Appendix A-1
Appendix A · Different leads, no overall winner
Different leads, no overall winner
Spotlight #1 · Sept 1st, 2026 · Company knowledge graph snapshot 31 August 2026
Commercial readiness: Early · estimated window 2030–2033
Spotlights are occasional themed readings of the Company knowledge graph. They are not the daily Reality
Check. This one compares the main ways companies on that graph are trying to build a quantum computer,
and explains what happens when a program runs on each kind of machine.
There is not one winner. Different approaches are ahead at different jobs. Commercial readiness remains
early. Broad, dependable use in ordinary organizations is still more plausible in the 2030–2033 window than
it is this year. Nothing in this article changes that estimate.
What a quantum computer is trying to do
Ordinary computers are already extremely good at the work most organizations need. They move money,
route packets, train models, and simulate large parts of the physical world by reducing those jobs to
arithmetic on bits. A bit is a stable distinction between two states. The entire digital economy is an
engineered agreement that those distinctions can be copied, stored, and operated on without the hardware
noticing the difference between a payroll file and a weather field.
A quantum computer is an attempt to compute with a different kind of distinction. Its basic unit, the qubit,
can be prepared so that 0 and 1 are not yet settled. Until the device is measured, the state of each qubit —
and, more importantly, of the collection of qubits — is described by amplitudes: quantities that can
reinforce or cancel when the machine is steered through a sequence of operations. Measurement ends that
stage of the calculation. What comes out is ordinary bits. The quantum part of the work, if it happened at
all, happened in the interference of those amplitudes. A transistor is easier to picture. A control voltage
either crosses a threshold or it does not. Current is on or off. That clean switch is the bit. A qubit is not that
kind of device. While the program runs it is not sitting at 0 or 1. The operations only tilt the odds. When you
look, you still get only 0 or 1. The simple case is a yes-or-no question. You run the same program a
thousand times. If you get 700 ones and 300 zeros, the answer is yes. If you get 510 ones and 490 zeros,
you do not have an answer. You have a coin flip. One run never decides it. Most jobs are not a vote. In a
chemistry estimate each run produces a whole string of bits, such as 0101101. That length is not the
number of operations. It is the number of qubits you measured: seven qubits, seven bits. The operations
only steered those qubits before the look. That string is one snapshot of how the electrons were arranged,
not a yes or a no. An ordinary computer already knows what energy that particular arrangement would
have. After a thousand runs it has a thousand snapshots and a thousand little energy scores. It averages
those scores. That average is the energy number. Counting how many 1s appeared anywhere in the pile
would answer the wrong question. That number is what a chemist can put next to a flask: which of two
candidate catalysts has the lower barrier, and therefore which compound is worth months at the bench. An
ordinary computer can guess at that barrier only by throwing away some of the electron arrangements that
matter. The quantum machine is being asked to keep those arrangements in the calculation long enough
that the averaged number matches the reaction — or shows, cheaply, that the reaction is not worth
running.
QUANTUM LITMUS / SPOTLIGHT #1 / 1 SEPTEMBER 2026
Appendix A-2
That is a narrower idea than the usual slogan that a quantum computer “tries every answer at once.” If you
load many candidate answers into a set of qubits and measure immediately, you sample from a spread of
outcomes. You do not receive a ranked list. The methods that matter are the ones that shape the
amplitudes so that the outcomes you want become likely and the ones you do not become rare. In that
sense the machine is closer to an instrument you tune than to a warehouse you search.
The reason anyone bothers is not mystery. It is cost. Some objects that scientists and engineers need to
describe — molecules, magnets, certain optimization landscapes, certain cryptographic functions — have
an internal structure whose complete ordinary description grows violently with size. An ordinary computer
can still approximate them, often very well, by throwing away structure and using huge amounts of time
and memory. A quantum device is built from the same sort of physical degrees of freedom as the system
under study, or can be programmed to imitate the arithmetic that system requires. The bet is that some of
those problems will fit the device more naturally than they fit a data centre full of bits.
Three jobs that make the bet concrete
The oldest of these jobs is physical simulation. Richard Feynman put the point plainly in the early 1980s:
nature is not classical, and a classical computer is an awkward tool for keeping track of it. Electrons in a
molecule do not occupy neat boxes. They occupy a shared state whose relevant arrangements multiply as
atoms are added. Chemists already have a mature toolbox of approximations for this. Those
approximations fail, or become expensive, in the cases people actually care about — a catalyst whose
activity depends on several correlated electrons, a battery material whose useful state is not the simplest
one, a magnetic material whose phases are hard to tell apart. A quantum processor will not “discover
drugs.” The more honest target is a number that is now slow or untrustworthy: an energy difference, a
reaction barrier, a correlation function. If that number can be produced more cleanly than it can be
produced today, a laboratory can spend fewer cycles on dead ends.
A second job is structure in large combination spaces. Assigning crews and aircraft, placing facilities,
cutting waste on a production line, and balancing a power grid are all problems of choosing among
arrangements. Ordinary algorithms already solve many of these at industrial scale by exploiting structure
and accepting good-enough answers. Quantum methods do not repeal that fact. What they offer, in theory,
is a different way to concentrate amplitude on better arrangements, or to sample from a distribution that is
hard to build classically. In practice this is the part of the field most often confused with a product. A
machine can be used inside a planning workflow and still be doing work that a conventional solver could
have done. The interesting evidence is not that an optimization ran. It is that the same class of job was
worth repeating, under operational constraints, after the novelty wore off.
A third job sits further out and is easier to state than to perform. A large share of public-key cryptography
assumes that certain reversible-looking operations — factoring large integers, extracting discrete
logarithms — remain slow on ordinary hardware. Shor’s algorithm showed that a sufficiently large,
sufficiently accurate quantum computer would not find them slow. That result is why banks, governments,
and standards bodies are already moving toward algorithms that do not rely on those assumptions. It is
also why cryptography is a poor illustration of current hardware. An attack of that kind is a long, precise
calculation. It requires error-corrected operations at a scale no company on the Company knowledge graph
has shown. When a laboratory revises the estimated size of such a machine, it is revising a planning figure.
It is not reporting that the attack exists.
Other proposed uses follow the same pattern. Some sampling tasks, some linear-algebra subroutines inside
larger programs, and some methods for producing randomness that can be checked as unpredictable are
all attempts to make amplitude do work that bit arithmetic does expensively. Each of them fails in the
same place if the device cannot hold and steer its state long enough.
QUANTUM LITMUS / SPOTLIGHT #1 / 1 SEPTEMBER 2026
Appendix A-3
What a run actually is
Almost every program written for the machines in this Spotlight still has the same outer shape.
An ordinary computer first translates the user’s question into operations this particular device can perform.
That translation is not clerical. Devices have limited connectivity, limited duration, and a native set of
operations that do not match the textbook. A short description on paper can become a long job once
information has to be moved across a chip, an ion chain, or an atom array. Every extra movement is
another chance to lose the calculation.
The device is then prepared in a known state, usually a field of zeros. The translated sequence is applied.
The device is measured. The measurement yields a string of bits. The string is almost never treated as the
answer. The loop is repeated until the ordinary computer has a distribution it can interpret. In the chemistry
and optimization examples above, the quantum processor is often only the inner kernel. The outer program
— choosing the next experiment, updating a model, deciding whether the result is good enough — remains
classical.
Two exceptions appear later because they change that picture rather than decorate it. Some
superconducting systems, D-Wave’s among them, do not take a list of operations. They encode a problem
as an energy landscape and change that landscape until the hardware settles. Some photonic systems do
not hold a long-lived register of matter qubits at all. They turn the program into a pattern of measurements
on light. Those are different ways of making amplitudes do work. They are not different marketing names
for the same box.
Why the hardware argument starts here
Amplitudes are not software. They live in a physical object, and every object has a preferred way of failing.
Heat, charge noise, stray magnetic fields, imperfect control pulses, lost atoms, and lost photons are not
metaphors. They are the reasons a calculation that exists on a whiteboard dies in a laboratory.
Error correction is the proposed industrial answer: store one reliable unit of information across many
unreliable physical qubits, keep measuring the symptoms of error, and repair as you go. That answer is
real, and it is expensive. The extra qubits are not a larger product. They are the cost of making the product
hold together. A processor can therefore look large in a company announcement and still be unable to
finish a program whose structure matches the three jobs above.
This is the point of comparing technologies instead of ranking press releases. Companies are choosing
different objects to carry the same kind of information because those objects demand different factories,
fail in different ways, and force programs to be carried out differently. Superconducting circuits, trapped
ions, atoms in laser traps, spins in silicon, and photons are not five brands. They are five answers to a
single engineering question: where can amplitudes be prepared, steered, and read with enough control
that the cancelling-and-adding process outruns the damage.
Superconducting systems
In these machines the qubit is a tiny electrical circuit on a chip. The chip is cooled far below any
temperature used in ordinary industry. On the Company knowledge graph this group includes IBM, Google,
Rigetti, IQM, Oxford Quantum Circuits, QpiAI, SEEQC, QuantWare, Alice & Bob, Amazon’s quantum
research, Fujitsu, and D-Wave. D-Wave belongs here because it also uses superconducting circuits, even
though customers program it differently.
QUANTUM LITMUS / SPOTLIGHT #1 / 1 SEPTEMBER 2026
Appendix A-4
How a step-by-step program runs
You begin with a sequence of operations, or a software tool writes that sequence for you. Another program
then rewrites it so that it fits this chip. On these chips a qubit can usually interact only with its neighbours.
If two qubits that need to work together sit far apart, the software inserts extra steps to move information
across the device. Those extra steps make the job longer and create more chances for error.
Each operation is a short pulse of microwave energy sent down a wire. One kind of pulse changes a single
qubit. Another kind briefly links two neighbouring qubits. To read an answer, the machine checks a nearby
sensor. The sensor’s behaviour tells the ordinary computer whether that qubit came out as 0 or 1. The
whole sequence is then run again. The collection of results, not a single pass, is what the outer program
interprets.
If the company is trying to make the result reliable, the sequence you wrote is not the sequence that
actually runs. Each useful unit of information is spread across many physical circuits. A large share of the
machine’s time is spent checking for damage rather than doing the original calculation.
How a cat-style superconducting machine differs
Alice & Bob, and Amazon’s related research, still use a cold superconducting box and still send microwave
signals. They store the information in a different microwave state, chosen so that one common type of
error is already unlikely. The user still thinks in terms of a sequence of operations. The hope is that fewer
physical circuits will be needed to protect each useful unit of information.
How a D-Wave-style program differs
D-Wave machines do not take a list of quantum operations in the same way. You describe a problem as a
landscape of better and worse answers. The chip is wired to represent that landscape. The machine
changes the landscape slowly and the circuits settle into a pattern. You read that pattern. An ordinary
computer may adjust the problem and try again.
This is a different kind of program. It can be a reasonable fit for some planning and optimization tasks. It is
not a substitute for a general algorithm from a textbook. We keep D-Wave in the superconducting group
because the hardware is still a superconducting circuit. We do not treat it as a separate contest.
Why this group is ahead in industry and supporting equipment
This is where the largest technology companies have serious programs. It is also where specialists already
sell pieces of the stack: chips and packaging, cooling systems, and control electronics. On the Company
knowledge graph that supporting layer is thicker here than in any other group. That does not mean the
machines are ready for everyday work. It means this approach looks most like an industry.
Recent evidence moved money, delivery plans, a new control-electronics company, and a repeated
optimization job at a telecom operator. None of that showed that general useful computation is available,
or that the 2030–2033 estimate should change.
QUANTUM LITMUS / SPOTLIGHT #1 / 1 SEPTEMBER 2026
Appendix A-5
Trapped-ion systems
In these machines the qubit is a single charged atom held in place by electric fields inside a vacuum
chamber. Companies in this group on the Company knowledge graph include IonQ, Quantinuum, and Alpine
Quantum Technologies. Honeywell appears as related exposure through Quantinuum.
As of the 31 August 2026 snapshot, the populated market-value field for IonQ is the largest among quantum-first companies on the Company knowledge graph. Quantinuum’s market value is not populated on that snapshot, so this is a comparison of filled fields, not a ranking of company value.
How a program runs
You still begin with a sequence of operations. Light, or a microwave signal, changes one atom. To connect
two atoms, the machine makes them share a vibration and then removes that vibration, leaving the two
atoms linked. In a single row of atoms, many pairs can be connected directly. The software therefore often
inserts fewer extra movement steps than it does on a superconducting chip.
On larger machines the atoms themselves are moved from one zone to another. Moving an atom is part of
running the program, not a side task.
The answer is read with light. One state of the atom glows under a probe. The other state stays dark. A
detector turns that glow or darkness into an ordinary 0 or 1. As with the other gate machines, the loop is
repeated and an ordinary computer interprets the collection of strings.
Why this group leads on fidelity and several logical-error records
These systems have a long public record of precise operations and flexible connections between qubits.
Quantinuum’s Helios processor is a 98-qubit trapped-ion machine with published component fidelities at
the front of the public table, and associated work reports encoded computations across tens of logical
qubits, including encodings near two physical qubits per logical qubit. That is a protection-quality record,
not only a hardware-quality record. It is why this Spotlight does not give neutral atoms an unqualified lead
on “protected qubits.”
A recent item in our daily series showed only that, in a model, an ordinary processor could keep up with the
extra checking work that error correction would require on one ion design. That is a point about control and
software. It is not evidence that ion machines can be manufactured at very large scale, or that they are
ready as a product.
The main open problem remains packing. It is still much harder to place millions of ions than it is to print
millions of circuit elements on a chip.
Neutral-atom systems
In these machines the qubit is an uncharged atom held in a spot of laser light. The spots can be moved.
Companies in this group on the Company knowledge graph include Pasqal, Infleqtion, QuEra, and Atom
Computing. Hitachi appears at the edge of the group through software and mixed research.
QUANTUM LITMUS / SPOTLIGHT #1 / 1 SEPTEMBER 2026
Appendix A-6
How a step-by-step program runs
Atoms are loaded into a grid of light spots. Some spots miss. The machine slides spare atoms into the
empty seats. A focused beam changes one atom. To connect two atoms, the machine briefly puts one atom
into a very large, fragile orbit. While that is happening, a nearby atom cannot enter the same orbit. That
rule links the two atoms.
Between stages of the program, the machine can move atoms so that the next pair that need to work
together are sitting next to each other. The arrangement of the atoms is part of how the program is carried
out.
At the end the machine photographs the whole grid. Whether an atom is bright, dark, or missing becomes
the result. Missing atoms matter. In this kind of machine a qubit can disappear, not only change from 0 to
1.
How a program can run without a list of steps
Some jobs do not use a written sequence of operations. The operator sets the pattern of lights and field
strengths, waits, and then photographs the atoms. The group of atoms changes together as one system.
That method is used for some simulation work and some planning work. It is not the same as sending a
standard quantum program to the machine.
Why this group leads on array scale and rearrangeability
In the past year this approach has produced some of the largest public logical-qubit experiments, including
work associated with QuEra, Harvard, and MIT that reports deep logical circuits with on the order of 96
logical qubits. Being able to rearrange atoms is part of that story, because the layout required by an
error-correcting code can be built by moving atoms rather than only by adding extra operations. That is a
lead in scale and reconfigurability. It is not an unambiguous lead in protection quality. Trapped-ion
hardware, especially Helios, holds several of the strongest published fidelity and repeated-encoding results.
Recent evidence moved funding, including Pasqal’s public listing, and improved day-to-day control,
including faster correction of a laser that had drifted. A national research machine in Japan also came
online. None of that showed that customers are getting repeatable useful results, that the economics work,
or that the readiness window has moved.
Silicon spin systems
In these machines the qubit is the magnetism of an electron, or of a carefully placed atom, inside silicon.
Companies in this group on the Company knowledge graph include Diraq, Quantum Motion, Equal1, Silicon
Quantum Computing, and Photonic Inc. Hitachi also has related work in this area.
How a program runs
You begin with a sequence of operations. A voltage change or a microwave signal turns one spin. To
connect two neighbouring qubits, the machine lowers the electrical barrier between them for a short,
measured time so they influence each other, then raises the barrier again. A qubit usually talks only to its
neighbours. If information needs to travel farther, the software adds extra steps.
The answer is not read by looking at the spin directly. Only one spin state is allowed to jump onto a sensor.
The sensor notices a tiny change in charge, and the ordinary computer infers whether the qubit was 0 or 1.
QUANTUM LITMUS / SPOTLIGHT #1 / 1 SEPTEMBER 2026
Appendix A-7
Several qubits can share one sensor and take turns. Silicon Quantum Computing has shown a version of
that sharing with four qubits.
Photonic Inc. is trying to add another step: copy the information from a spin onto light, send the light
through fibre, and catch it in another module. That would turn one program into a program spread across
more than one box. That idea is still research.
Why this group has the clearest semiconductor-compatibility thesis
The case for silicon is not that it already has the most qubits. It does not. The case is a manufacturing
thesis: spin qubits can be built in processes that sit close to the lines used for ordinary chips, including 300
mm foundry flows. That thesis is real, and it is incomplete. Published work is clear that ordinary transistor
processes still have to be customized. Superconducting qubits have also been fabricated with industrial
300 mm methods. Compatibility is not a monopoly.
Diraq’s plan to place an eight-qubit machine in an Equinix data centre in Sydney is an announced operations experiment. Public material reviewed for this report does not include measured uptime in that setting, a customer workload result, a comparison showing advantage over ordinary computers, or published operating economics. Absence of those figures in the material reviewed is not a finding that the installation failed.
Photonic systems
In these machines the qubit is a property of light, such as which path a particle of light took. Companies in
this group on the Company knowledge graph include PsiQuantum, Xanadu, Quantum Computing Inc.,
ORCA, and Quandela.
How a program can run
There are several versions, and they are easy to confuse.
In the most familiar version, a sequence of operations is turned into settings on an optical arrangement:
splitters, delays, and path changes. Light travels through that arrangement. Detectors click. The clicks
become 0s and 1s. Connecting two qubits while the light is in flight is difficult, which is why many groups do
not rely on that method for large machines.
In the version that large photonic projects often describe, the written program is translated into a
measurement recipe. The machine tries to build a large web of linked light from many small pieces. Some
links fail. The design is supposed to work around missing light. The program is carried out by choosing what
to measure and in what order. The information does not sit in a tray of long-lived matter qubits. It sits in the
light that has not been measured yet.
In another version, used by some Xanadu-style work, the program is a set of optical settings rather than a
list of bit operations. Light goes through a mesh. Detectors count. An ordinary computer finishes the
calculation. Some problems can be posed that way without ever looking like a standard qubit program.
QUANTUM LITMUS / SPOTLIGHT #1 / 1 SEPTEMBER 2026
Appendix A-8
Why this group leads on optical networking, not on “room temperature”
Light already travels through fibre. That networking point is strong. The phrase “ordinary temperatures” is
not. PsiQuantum’s published design still cools single-photon detectors to about 2–4 K. That is much warmer
than the millikelvin refrigerators used for superconducting chips, and it can use industrial helium plants
rather than a chandelier-style dilution fridge. It is still cryogenics. Parts of a photonic system can sit at room
temperature. The detectors, for now, cannot.
The gap is the middle of the journey. Compared with superconducting and trapped-ion machines, it is
harder for a customer to point to a public, intermediate photonic computer and run the same kind of test.
Canada’s manufacturing support for Xanadu moved the factory plan. It did not show working production
yield, useful computation, or customer economics.
The main physical cost in this group is loss. A missing particle of light is not simply a wrong answer that you
correct later in the same way as a flipped bit. It is a piece of the calculation that never arrived.
Why the supporting companies matter
A quantum computer is not only the qubits. It also needs cooling, control electronics, software that
suppresses errors, equipment that corrects errors in real time, and factories that package chips.
On the Company knowledge graph that supporting layer is densest around superconducting systems. That
is one reason superconducting systems look like an industry.
Some of the most important supporting companies, including firms that sell control systems and
error-correction software, say their tools can work with more than one kind of qubit. As long as those
suppliers refuse to bet on only one approach, the market is not forcing a single winner. That is a second
reason this article describes different leads rather than one crown.
Microsoft’s topological effort is on the graph. On the public evidence used here, it is a research roadmap inside a large company, and it is not ahead on the jobs compared above.
Software companies that write programs without choosing a qubit type also do not decide this contest. A
compiler does not choose the hardware.
Scorecard
Question Lead cited in public evidence What that does not prove
Who has the most complete industry around the machine?
Superconducting That these machines can do useful general work for customers
Who has the strongest public fidelity and several logical-error records?
Trapped ion That ions can be manufactured at very large scale
Who leads on array scale and rearrangeability in public logical-qubit experiments?
Neutral atom An unambiguous lead on protection quality, or that customers are buying the result
Who has the clearest semiconductor-compatibility thesis?
Silicon spin That ordinary transistor lines can be reused unchanged, or that a planned colo install is a data-centre lead
QUANTUM LITMUS / SPOTLIGHT #1 / 1 SEPTEMBER 2026
Appendix A-9
Question Lead cited in public evidence What that does not prove
Who best matches optical networking and warmer-than-millikelvin cryogenics?
Photonic Room-temperature operation end to end, or a public middle-size machine customers can test in the usual way
Who has supporting suppliers you can already name?
Superconducting That parts and cooling equal readiness
What changed recently, and what did not
Pasqal’s public listing added money to the neutral-atom group. It did not show better processor
performance.
Diraq’s plan with Equinix strengthened the silicon argument about data centres. The material reviewed does not show measured uptime in that setting.
QuEra showed faster correction of a drifting laser. That is an operations improvement, not a completed
product.
IonQ showed, in a model, that ordinary computing hardware might keep up with error-checking work on an
ion design. That is not the same as a finished, error-corrected service.
Xanadu received a manufacturing commitment from Canada. That is a factory plan, not proof of yield or
customer results.
A new company, S-Transistors, added to the superconducting control-electronics layer. D-Wave showed a
repeated optimization job inside a working telecom network. Both belong to the superconducting story.
Neither changes the overall readiness estimate.
What did not change is the part that matters for readers of the Company knowledge graph. We still do not
have useful computation as something an organization can order across these groups. We still do not have
durable customer economics. We still estimate broad commercial readiness in the 2030–2033 window.
Closing
These companies are not all building the same machine. Superconducting work looks most like an industry
with suppliers. Trapped-ion work has the strongest public record for high-fidelity gates and several leading
logical-error demonstrations. Neutral-atom work leads on array scale, rearrangeability, and some of the
largest public logical-qubit experiments. Silicon work offers the clearest compatibility thesis with
semiconductor manufacturing. Photonic work offers the clearest connection to optical networking and
cryogenics that are demanding, but less demanding than millikelvin refrigerators.
That is specialization. It is not arrival. Until one approach can do useful work, keep customers, and survive
the ordinary demands of a real facility, there is no overall leader. There are only different leads on different
jobs. None of those leads is commercial readiness.
Assessment
Evidence confidence: Moderate. Impact confidence: Low. Commercial-readiness effect: None. QCRI
movement: None. This is an architecture-specific evidence ledger, not a change to the 2030–2033 window.
Primary sources
QUANTUM LITMUS / SPOTLIGHT #1 / 1 SEPTEMBER 2026
Appendix A-10
Trapped ion — Ransford et al., “A 98-qubit trapped-ion quantum computer with all-to-all connectivity,”
Nature (2026), doi:10.1038/s41586-026-10676-4; preprint arXiv: 2511.05465. Quantinuum, “From Memory
to Logic,” on Helios encodings including 48 logical qubits at about 2:1 and larger encoded demonstrations.
Neutral atom — Zhou et al., “Low-overhead transversal fault tolerance for universal quantum
computation,” Nature 646, 303–308 (2025), doi:10.1038/s41586-025-09543-5. “A fault-tolerant
neutral-atom architecture for universal quantum computation,” Nature 649, 39–46 (2025),
doi:10.1038/s41586-025-09848-5. Ismail et al., “Transversal architecture for megaquop-scale quantum
simulation with neutral atoms,” PRX Quantum; arXiv: 2509.18294.
Photonic cryogenics — PsiQuantum, “Introducing Omega,” on 2–4 K detector modules and industrial
cryoplants. Alexander et al., “A manufacturable platform for photonic quantum computing,”
arXiv: 2404.17570 (SNSPDs operated near 2 K).
Manufacturing — Van Damme et al., “Advanced CMOS manufacturing of superconducting qubits on 300
mm wafers,” Nature (2024), doi:10.1038/s41586-024-07941-9. Steinacker et al., “A 300 mm foundry silicon
spin qubit unit cell exceeding 99% fidelity,” arXiv: 2410.15590. Imec notes on Si-spin and superconducting
300 mm flows that still require process customization.
Company knowledge graph snapshot, 31 August 2026. Reality Checks on Diraq/Equinix, S-Transistors,
Pasqal, Xanadu, QuEra, IonQ, and D-Wave/NTT DOCOMO (late August 2026).
See the holdings disclosure beneath the report byline. Independent editorial
analysis; not investment, procurement, or engineering advice.
Quantum Litmus · Spotlight #1 · Sept 1st, 2026
A themed reading of the Company knowledge graph, not the daily tape.