Ernst Ising was born in Cologne on May 10, 1900. He grew up in Bochum, studied physics and mathematics at Göttingen and Hamburg, and in 1922 began working on a problem suggested by his doctoral advisor Wilhelm Lenz: model a ferromagnet using the simplest possible system. A chain of atomic spins, each either up (+1) or down (−1), interacting only with their immediate neighbours. Calculate whether the system can undergo a phase transition — a spontaneous alignment of all spins at low temperature that would explain permanent magnetism.
Ising solved the one-dimensional case in 1924. His result was correct: no phase transition in one dimension. But he made a fateful extrapolation. He concluded that his model would also fail to produce a phase transition in two or three dimensions. On this basis, he published his thesis, moved on, and never returned to theoretical physics.
He was spectacularly wrong. In 1936, Rudolf Peierls proved that the two-dimensional Ising model does exhibit a phase transition. In 1944, Lars Onsager produced the exact analytical solution for the two-dimensional case — a result widely considered one of the greatest achievements in mathematical physics. The model that Ising had dismissed as trivial turned out to be the simplest system in nature that exhibits a phase transition between order and disorder. It became the standard model of statistical mechanics.
He never returned to his early research. It was not until 1949 that he found out, from the scientific literature, that his model had become widely known.
By then, Ising’s life had been destroyed and rebuilt. As a Jewish scientist, he was barred from teaching when Hitler came to power in 1933. He found work as headmaster of a Jewish boarding school in Caputh — next door to Albert Einstein’s summer house. In 1938, the Nazis destroyed the school. In 1939, Ernst and his wife Johanna fled to Luxembourg, where he worked as a shepherd and a railroad labourer. After the Wehrmacht occupied Luxembourg, he was forced into army labour. In 1947, the Isings emigrated to the United States, where he became a physics professor at Bradley University in Peoria, Illinois. He never published again. He died on May 11, 1998 — one day after his 98th birthday.
Today, roughly 800 papers per year cite the Ising model. It is used in neural networks, protein folding, biological membranes, social behaviour modelling, and — most consequentially — quantum computing. The Ising model was the historical precursor to the Hopfield network (1982), which was the precursor to modern recurrent neural networks. The line from Ernst Ising’s 1924 thesis to the AI systems running today is direct and unbroken.
Why Nvidia Named Its Quantum AI After a 1924 PhD Thesis
On World Quantum Day — April 14, 2026 — Nvidia launched NVIDIA Ising, the world’s first open-source AI model family for quantum computing. The fundamental problem in quantum computing is noise. The best quantum processors in 2026 make an error roughly once in every thousand operations. To become useful, that rate needs to drop to one in a trillion. The gap — nine orders of magnitude — is the single largest engineering challenge in the field.
The solution is quantum error correction: encoding logical qubits in redundant physical qubits so that errors can be detected and corrected in real time. But error correction generates enormous volumes of data — terabytes of qubit measurements must be processed thousands of times per second by classical decoding algorithms. This is, in Nvidia’s framing, an “AI-shaped workload” — a problem that maps directly onto the pattern-recognition capabilities of neural networks running on GPUs.
Nvidia chose the name Ising deliberately. The Ising model is the simplest possible description of a system that transitions between order and disorder — which is precisely what quantum error correction does: it forces a noisy, disordered quantum system into an ordered, usable state. The mathematical structure is the same. The spins that Ernst Ising modelled in 1924 are, in a precise sense, the qubits that Nvidia Ising corrects in 2026.
Ising Calibration is a 35-billion-parameter vision-language model trained on multi-modal qubit data. It interprets the visual and numerical output of quantum processors — oscillation plots, spectroscopy data, coherence measurements — and drives an AI agent that automates the continuous tuning of qubit control signals, replacing the human physicists who currently adjust microwave pulses and laser frequencies by hand. Ising Decoding is a 3D convolutional neural network that processes the syndrome data from error-correcting codes and determines the most likely error configuration in real time — faster than the errors accumulate.
The CUDA Playbook, Applied to Quantum
The strategic logic is unmistakable. Nvidia has run this play before. In 2006, it released CUDA — a free, open programming framework for GPU computing. CUDA did not generate revenue directly. It created dependency. Every researcher, every lab, every company that wrote code in CUDA was locked into Nvidia hardware. By the time AI training became the dominant workload in computing, CUDA was the standard, and switching costs were enormous. Nvidia’s GPU monopoly in AI is, in large part, a CUDA monopoly.
Ising is CUDA for quantum. The models are open-source and free. But they run on Nvidia GPUs, integrate with CUDA-Q, and connect through NVQLink — Nvidia’s proprietary QPU-GPU interconnect. Every quantum lab that adopts Ising builds its operational stack on Nvidia infrastructure. When quantum computing scales — and it will — those labs will need more Nvidia GPUs, not fewer. The classical control plane for a million-qubit quantum computer will require enormous GPU capacity for real-time error correction. Nvidia is positioning itself to supply it.
What It Means
The popular narrative assumes AI and quantum computing are parallel tracks that will eventually converge. Ising reveals the actual relationship: quantum computing is dependent on AI for its operational viability. Qubits are too noisy to function without continuous AI-driven error correction. GPUs are not being displaced by QPUs — they are the permanent co-processor that QPUs cannot operate without. This is a structural dependency, not a transitional one.
The CUDA analogy is precise and actionable. CUDA was released in 2006, seven years before deep learning went mainstream. By the time everyone needed GPUs for AI training, CUDA was already the standard. Ising is being released now, years before quantum computing reaches commercial scale. The labs adopting it today will be locked in when the market matures. The switching costs will be the same as CUDA’s: technically possible, economically prohibitive. The investment question is not “which quantum company will win?” It is “who provides the control plane for all of them?”
And then there is the human story, which is the story of science itself. Ernst Ising solved the simplest possible version of his model, concluded it was trivial, and walked away. Twenty years later, Onsager proved it was one of the deepest models in physics. Fifty years later, it became the foundation of neural networks. A century later, it is the namesake of the AI that controls quantum machines. The lesson is not about Ising. It is about the nature of foundational work: you cannot know, at the time you do it, what it will become. The most consequential model in statistical physics was a PhD thesis its author thought had failed.
Flight Log — Dispatch from Altitude
There is a concept in aviation called a Minimum Equipment List — the MEL. It is a regulatory document that specifies which systems on an aircraft can be inoperative and still allow safe dispatch. A failed weather radar on a clear day: you can go. A failed attitude indicator in instrument conditions: you cannot. The MEL does not tell you what the aircraft needs to fly. It tells you what the aircraft cannot fly without.
Quantum computing just got its MEL — and AI-driven error correction is on it. Not as an enhancement. Not as an optimisation. As a dispatch requirement. Without real-time decoding, the qubits are too noisy to produce useful results. Without continuous calibration, the hardware drifts out of tolerance within hours. Ising is not a nice-to-have. It is a minimum equipment item. The quantum processor does not dispatch without it.
What strikes me most about the Ernst Ising story is not the physics. It is the timing. He solved his model in 1924 and concluded it was a dead end. Onsager proved it foundational in 1944. Ising did not learn of it until 1949 — twenty-five years after his thesis, years he spent fleeing the Nazis, working as a shepherd in Luxembourg, and rebuilding a life in Illinois. By the time the world recognised what he had built, he had moved on.
You do not get to choose when your work matters. You do not get to choose whether it matters. Ernst Ising, born in Köln, fled Europe with nothing and died in Peoria at 98 — and his name is now on the AI that will control the quantum computers of the next century. He did not plan this. He could not have planned this. He simply solved the simplest possible version of a problem, as honestly as he could, and let it go.
Sometimes that is enough.