On November 11, IBM announced Nighthawk r3, its next quantum processor. 520 programmable qubits, built to run 10,000 quantum operations. In 2027, IBM plans to bolt two of them together into a single 1,040-qubit system.
If you build with cryptography, that announcement deserves ten minutes of your attention. Not because your keys are in danger today. Because of what the announcement actually shows about where the timeline is going, and because most of the coverage got the story wrong in both directions.
What IBM actually announced
The numbers first, because they are the part people keep skipping past to get to the scary headline.
Nighthawk r3 is designed around 520 programmable qubits and 2,034 physical qubits in total, built to execute 10,000 quantum operations in a single circuit. The current chip, Nighthawk r2, has been running since September: 120 qubits, up to 100,000 samples per second, circuits of more than 7,500 operations. Partners have run 34 billion circuits on it. E.ON reported a 15x speedup on an energy grid problem. IBM and Oak Ridge ran a materials simulation 12 times faster than they had on the previous generation.
The part that matters most is not any of those numbers. It is the ten chip-to-chip interconnects IBM calls l-couplers. IBM already linked two earlier Nighthawk processors with four of them and moved a quantum state between chips at 99.3 percent fidelity, working toward 99.9. It also connected two cryogenic modules into one system for the first time and got it down below 15 millikelvin in under five days.
Read that again in plain language: IBM is no longer trying to build one giant chip. It is learning to snap smaller chips together like Lego bricks, and the snapping works. That is the whole ballgame for scaling, because a single chip can only get so big before the physics of cooling it and wiring it up starts fighting you. Modular scaling is how you get from hundreds of qubits to thousands and then to the millions.
And 2027 is not far away. The two-chip, 1,040-qubit system is a plan for next year, not a research paper about someday.
The honest part: this does not break your encryption
Now the part the scary headlines leave out.
Breaking RSA-2048, the workhorse that still protects a huge share of the internet, needs on the order of millions of physical qubits once you count error correction. The best-known estimate, from Gidney and Ekerå's 2021 analysis, puts it around 20 million physical qubits running for about eight hours. IBM is at 520 programmable qubits, aiming at 1,040 next year.
That is not close. It is not "almost there." It is four orders of magnitude away, and every one of those orders of magnitude has to be earned through error correction, which is the hardest problem in the field. A qubit that forgets its state in microseconds is not a computer; it is a very expensive random number generator. Fault tolerance is the wall, and nobody has climbed it yet.
So if someone tells you Nighthawk r3 means your encrypted messages are about to be readable, they are wrong, and you can say so with confidence. Your RSA keys, your elliptic curves, your Signal chats, none of them are in danger from this chip.
But here is the thing I have learned from building a messenger myself: the wrong reaction to a quantum announcement is not panic. It is also not shrugging. It is looking at the direction of travel.
Why the direction of travel matters more than the qubit count
Three years ago, the quantum roadmap was a slide deck. Two years ago, the chips existed but were slow and noisy enough that serious work stayed in simulation. This year, a 120-qubit chip ran 34 billion real circuits for real customers and got measurably faster at real problems, and the company that built it is already talking about linking chips together next year.
That is what a technology looks like when it stops being a lab demo and starts being an engineering program. The questions change. They stop being "is this physically possible" and start being "how fast can we manufacture it, cool it, and wire it up." Engineering programs have a habit of arriving earlier than skeptics expect, because each solved problem makes the next one cheaper to solve.
Meanwhile the other side of the clock is already written down. NIST's final transition report, published October 27, says quantum-vulnerable public-key cryptography is deprecated after 2030 and disallowed after 2035. That is not a prediction about when a quantum computer arrives. It is a deadline for when your systems have to be ready either way. Nine years sounds like a lot until you remember how long a real migration takes. The browser vendors spent years rolling out hybrid key exchange. AWS only added ML-KEM to its key management service last month. Big organizations measure crypto migrations in years, not quarters.
And the threat that makes the deadline real is not a future computer. It is the one running right now, quietly recording traffic. Harvest now, decrypt later is not a theory. Any adversary with storage can capture encrypted traffic today and keep it until a machine exists that can open it. The data with a long shelf life, medical records, state secrets, the messages you send today that you would still care about in ten years, is already exposed to that bet. The question is never "will a quantum computer exist in 2035." The question is "will the things I encrypt today still matter then."
What this means if you actually build things
I build an encrypted messenger. It uses ML-KEM-768 for key exchange, the same algorithm NIST standardized and the same one AWS just shipped in KMS. I picked it because the migration is not something you do after the deadline. It is something you do while the deadline is still far away, when you can afford to get it wrong and fix it.
That is the practical lesson of Nighthawk, and it is boring in the best way. You do not need to panic, and you do not need to wait for a breakthrough headline to start. The work is unglamorous:
Inventory your cryptography. Most teams cannot list every place they use RSA or elliptic curves. If you do not know where the vulnerable crypto lives, you cannot replace it.
Go hybrid first. Pair a classical algorithm with a post-quantum one, the way browsers did with X25519 and ML-KEM. If the new algorithm turns out to have a flaw, the classical half still holds. If the classical half falls to a quantum computer, the new half holds. Hybrid is not indecision. It is the only sane way to migrate a live system.
Treat signatures as the hard half. Key exchange got the attention first because it is the half an attacker can exploit retroactively. Signatures are only now starting to ship in real software, and they are bigger, slower, and harder to fit into certificates and protocols. If your migration plan only covers key exchange, it is half a plan.
And stop waiting for a single moment when "quantum arrives." There will not be one. There will be a press release about a chip, then another about linking two chips, then a paper about error correction, and somewhere in that slow drumbeat the machines will cross from interesting to dangerous. The people who are ready will be the ones who started while it still looked early.
The clock is real, and that is good news
I am not writing this to scare anyone. I am writing it because IBM's announcement is genuinely good news for people who take this seriously. A public roadmap with dates on it is better than a rumor. A chip you can buy time on is better than a slide. Modular scaling that demonstrably works is better than a promise that it will work someday.
Nighthawk r3 will not break your encryption. What it does is put a date on the calendar and show you the machine that is walking toward it, one linked chip at a time. Nine years is a long time to migrate. It is also exactly the amount of time a real migration takes. Start now, while it still feels early, because that feeling is the whole point. Early is when the work is cheap.
Sources: IBM Newsroom, "IBM Quantum Nighthawk Demonstrates New Levels of Speed and Quality," November 11, 2026; NIST IR 8547, "Transition to Post-Quantum Cryptography Standards," October 2026; Gidney and Ekerå, "How to factor 2048 bit RSA integers in 8 hours using 20 million noisy qubits," 2021.
Top comments (0)