The Quantum Endgame Is Not About Cleaner Qubits, It's About Who Can Mass-Produce with Phone-Chip Fabs
The quantum endgame is not about whose qubits are cleaner, but about who can mass-produce using existing semiconductor fabs; silicon spin qubits reuse TSMC lithography, while ion traps are stuck with a hundred thousand laser beams and superconductors with warehouse-sized refrigerators.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Ion traps and superconductors need miracles first; silicon doesn't
To build an ion trap with a hundred thousand qubits, you'd first have to invent the optical miracle of a hundred thousand laser beams; a million superconducting qubits would require a cryostat the size of a warehouse; a million spin qubits only need standard 300mm silicon wafers, run through the same lithography equipment that TSMC, ASML, and Intel use to make phone processors. Krishna uses this to set the tone for the whole episode: the key to the quantum race is not single-qubit fidelity, but whether the path to scaling already exists within the industrial system.
— KrishnaQuantum computing is no longer about principles, but engineering
In 2000, DiVincenzo proposed five criteria, aiming to dismiss liquid-state NMR quantum computing. In this episode, FFP proposes three of its own criteria: qubit quality, control, and scalability with economics. The rest of the episode scores the four routes—superconducting, ion trap, neutral atom, and silicon spin—item by item. Laying out the evaluation system itself tells you: quantum computing has entered the stage of engineering competition, and you can't just talk about principles.
— KrishnaSuperconducting isn't a technical failure, it's an engineering space problem
Estimates show that a thousand logical qubits would require ten million transmon physical qubits. Control line count is limited by the cooling capacity of dilution refrigerators, frequency crowding occurs, and helium-3 supply becomes a new resource constraint. FFP gives superconducting only 1/10 on scalability and economics, with a total score of 11/30. This is the first time the show separates 'technically strong' from 'engineering infeasible'; a warehouse-sized cryostat means even data centers might not accommodate it.
— HostIon traps win on coherence time, lose on gate speed
Ion trap coherence times reach ten hours, but gate operations rely on phonons and are limited by the Paul trap oscillation frequency, making them about a thousand times slower than superconducting. The result is counterintuitive: even if ion traps could scale to code-breaking sizes, cracking 2048-bit RSA would take about a year, while superconducting would take only about eight hours. Long coherence time doesn't mean more work per unit time; for practical computation, slow gate time is a throughput disaster overall.
— HostThe quantum threat is neither tomorrow nor never
A March 2026 paper from Harvard, Caltech, and Qera says ten thousand qubits might crack internet encryption, but it would actually take several years; even with a hundred thousand physical qubits, under mathematically optimal assumptions, cracking RSA would take about three months. This figure strikes at both extreme narratives: the quantum threat is not tomorrow, but it is certainly not forever distant. To achieve 'crackable within months' requires qubit count, gate speed, and error correction all to meet the bar simultaneously.
— GuestQuantum companies live or die by government contracts
HRL lost key project funding in early 2026 and laid off 376 people. Government funding shifted from basic research to applied research demanding rapid commercialization; IBM and Intel received support from the CHIPS Act, and ultimately IBM acquired HRL. Krishna calls this the debut of quantum computing's MVP: quantum companies' fates hinge on government contracts, and being acquired is not a failure but a way for the technology to enter an industrial system with manufacturing capability.
A million qubits can't be tuned by hand; it's image recognition now
Automated tuning has become key to scaling: a 2023 paper could tune six electrons by hand, but hand-tuning a million qubits would take a human lifetime. Tuning is essentially image recognition—when scanning voltage, tunneling events appear as stripes. Krishna's team uses DETR combined with ResNet and Transformer to automatically recognize stripes and generate parameters. This is also the hidden prerequisite for why silicon spin can reuse semiconductor fabs: no matter how many qubits, software can align them automatically.
— KrishnaSilicon may not be first, but it will be the last to win
FFP's final scores: silicon spin gets 10/10 on qubit quality, 10/10 on control (provided isotope enrichment and valley state issues are solved), and 10000/10 on scalability and economics, for a total of 10020/30; neutral atoms score negative 3000. Krishna admits silicon may not be the first route to achieve fault tolerance, but once it does, it will become the final standard, like transistors replacing vacuum tubes, because it is cheap and manufacturable.
— KrishnaIn their own words · checked verbatim
But if you want a million spin qubits, you just put it on a standard 300 millimeter silicon wafer. You run it through the exact same photolithography machines at TSMC, ASML, Intel.
Krishna0:00
A quantum computer is not a machine that tries every single answer at the same time because a qubit is a zero and a one simultaneously.
Krishna23:39
This is where the startup pitch decks collide with reality. They collide with the laws of thermodynamics and the laws of economics, which are not real laws, which is why the economics Nobel Prize is not a real Nobel Prize.
For a thousand plus logical qubits. Which is not the million we talked about earlier. No. It's just a thousand. Thousand logical qubits. You need 10 million physical qubits of these transplants.
Host1:24:33
Here, even if we achieve scale, because the gate times are so slow, it might take like on the order of a year to decrypt 2048 RSA.
Host1:47:08
the current evidence is not enough. And you need to show, like, you know, I would expect, for example, a statistical analysis of valley splitting across devices.
reviewer3:13:48
I don't think it's going to be the future quantum computer. You know, once it's achieved, I think silicon is going to be the one that actually gets scaled because it's cheap and it's easy to make.
Krishna3:33:40
Figures
| Nature paper co-authors | 250 people | 3:07 |
| Nature paper publication date | July 30, 2026 | 3:07 |
| Superconducting qubit FFP score | 11/30 | 1:25:33 |
| Ion trap coherence time | 10 hours | 1:35:45 |
| Time to crack RSA with 100,000 physical qubits | about three months | 2:04:35 |
| Silicon-28 abundance in natural silicon | 92% | 2:38:43 |
| HRL layoffs | 376 people | 2:50:02 |
| HRL paper CNOT error rate | 0.09% | 3:07:30 |
| Silicon spin FFP total score | 10020/30 | 3:32:31 |
| Neutral atom FFP total score | -3000 | 3:32:31 |
Glossary
- transmon
- A type of superconducting qubit that uses a Josephson junction to introduce anharmonicity, making energy levels uneven to avoid misoperation.
- valley splitting
- The energy difference after the degeneracy of multiple energy valleys in silicon electrons is lifted; it determines the quality of silicon spin qubits.
- 2DEG
- Two-dimensional electron gas: electrons confined to move in a plane; it is the physical basis of silicon quantum dots.
- DETR
- Detection Transformer: an object detection model used here to identify tunneling stripes in tuning images.
- exchange-only qubit
- A qubit that uses only exchange interactions between electrons to implement quantum gates, requiring no microwaves, but each qubit needs three electrons.
How to listen
Investors focused on quantum computing route selection, semiconductor process engineers, and security practitioners wanting to calibrate the timeline for quantum code-breaking.
Listeners already familiar with superconducting and ion trap principles can skip from 57:56 to 1:28:00.