www.technologyreview.com
On January 8, Nvidia CEO Jensen Huang jolted the stock market by saying that practical quantum computing is still 15 to 30 years away, at the same time suggesting those computers will need Nvidia GPUs in order to implement the necessary error correction.However, history shows that brilliant people are not immune to making mistakes. Huangs predictions miss the mark, both on the timeline for useful quantum computing and on the role his companys technology will play in that future.Ive been closely following developments in quantum computing as an investor, and its clear to me that it is rapidly converging on utility. Last year, Googles Willow device demonstrated that there is a promising pathway to scaling up to bigger and bigger computers. It showed that errors can be reduced exponentially as the number of quantum bits, or qubits, increases. It also ran a benchmark test in under five minutes that would take one of todays fastest supercomputers 10 septillion years. While too small to be commercially useful with known algorithms, Willow shows that quantum supremacy (executing a task that is effectively impossible for any classical computer to handle in a reasonable amount of time) and fault tolerance (correcting errors faster than they are made) are achievable.For example, PsiQuantum, a startup my company is invested in, is set to break ground on two quantum computers that will enter commercial service before the end of this decade. The plan is for each one to be 10 thousand times the size of Willow, big enough to tackle important questions about materials, drugs, and the quantum aspects of nature. These computers will not use GPUs to implement error correction. Rather, they will have custom hardware, operating at speeds that would be impossible with Nvidia hardware.At the same time, quantum algorithms are improving far faster than hardware. A recent collaboration between the pharmaceutical giant Boehringer Ingelheim and PsiQuantum demonstrated a more than 200x improvement in algorithms to simulate important drugs and materials. Phasecraft, another company we have invested in, has improved the simulation performance for a wide variety of crystal materials and has published a quantum-enhanced version of a widely used materials science algorithm that is tantalizingly close to beating all classical implementations on existing hardware.Advances like these lead me to believe that useful quantum computing is inevitable and increasingly imminent. And thats good news, because the hope is that they will be able to perform calculations that no amount of AI or classical computation could ever achieve.We should care about the prospect of useful quantum computers because today we dont really know how to do chemistry. We lack knowledge about the mechanisms of action for many of our most important drugs. The catalysts that drive our industries are generally poorly understood, require expensive exotic materials, or both. Despite appearances, we have significant gaps in our agency over the physical world; our achievements belie the fact that we are, in many ways, stumbling around in the dark.Nature operates on the principles of quantum mechanics. Our classical computational methods fail to accurately capture the quantum nature of reality, even though much of our high-performance computing resources are dedicated to this pursuit. Despite all the intellectual and financial capital expended, we still dont understand why the painkiller acetaminophen works, how type-II superconductors function, or why a simple crystal of iron and nitrogen can produce a magnet with such incredible field strength. We search for compounds in Amazonian tree bark to cure cancer and other maladies, manually rummaging through a pitifully small subset of a design space encompassing 1060 small molecules. Its more than a little embarrassing.We do, however, have some tools to work with. In industry, density functional theory (DFT) is the workhorse of computational chemistry and materials modeling, widely used to investigate the electronic structure of many-body systemssuch as atoms, molecules, and solids. When DFT is applied to systems where electron-electron correlations are weak, it produces reasonable results. But it fails entirely on a broad class of interesting problems.Take, for example, the buzz in the summer of 2023 around the room-temperature superconductor LK-99. Many accomplished chemists turned to DFT to try to characterize the material and determine whether it was, indeed, a superconductor. Results were, to put it politely, mixedso we abandoned our best computational methods, returning to mortar and pestle to try to make some of the stuff. Sadly, although LK-99 might have many novel characteristics, a room-temperature superconductor it isnt. Thats unfortunate, as such a material could revolutionize energy generation, transmission, and storage, not to mention magnetic confinement for fusion reactors, particle accelerators, and more.AI will certainly help with our understanding of materials, but it is no panacea. New AI techniques have emerged in the last few years, with some promising results. DeepMinds Graph Networks for Materials Exploration (GNoME), for example, found 380,000 new potentially stable materials. At its core, though, GNoME depends on DFT, so its performance is only as good as DFTs ability to produce good answers.The fundamental issue is that an AI model is only as good as the data its trained on. Training an LLM on the entire internet corpus, for instance, can yield a model that has a reasonable grasp of most human culture and can process language effectively. But if DFT fails for any non-trivially correlated quantum systems, how useful can a DFT-derived training set really be? We could also turn to synthesis and experimentation to create training data, but the number of physical samples we can realistically produce is minuscule relative to the vast design space, leaving a great deal of potential untapped. Only once we have reliable quantum simulations to produce sufficiently accurate training data will we be able to create AI models that answer quantum questions on classical hardware.And that means that we need quantum computers. They afford us the opportunity to shift from a world of discovery to a world of design. Todays iterative process of guessing, synthesizing, and testing materials is comically inadequate.In a few tantalizing cases, we have stumbled on materials, like superconductors, with near-magical properties. How many more might these new tools reveal in the coming years? We will eventually have machines with millions of qubits that, when used to simulate crystalline materials, open up a vast new design space. It will be like waking up one day and finding a million new elements with fascinating properties on the periodic table.Of course, building a million-qubit quantum computer is not for the faint of heart. Such machines will be the size of supercomputers, and require large amounts of capital, cryoplant, electricity, concrete, and steel. They also require silicon photonics components that perform well beyond anything in industry, error correction hardware that runs fast enough to chase photons, and single-photon detectors with unprecedented sensitivity. But after years of research and development, and more than a billion dollars of investment, the challenge is now moving from science and engineering to construction.It is impossible to fully predict how quantum computing will affect our world, but a thought exercise might offer a mental model of some of the possibilities.Imagine our world without metal. We could have wooden houses built with stone tools, agriculture, wooden plows, movable type, printing, poetry, and even thoughtfully edited science periodicals. But we would have no inkling of phenomena like electricity or electromagnetismno motors, generators, radio, MRI machines, silicon, or AI. We wouldnt miss them, as wed be oblivious to their existence.Today, we are living in a world without quantum materials, oblivious to the unrealized potential and abundance that lie just out of sight. With large-scale quantum computers on the horizon and advancements in quantum algorithms, we are poised to shift from discovery to design, entering an era of unprecedented dynamism in chemistry, materials science, and medicine. It will be a new age of mastery over the physical world.Peter Barrett is a general partner at Playground Global, which invests in early-stage deep-tech companies including several in quantum computing, quantum algorithms, and quantum sensing: PsiQuantum, Phasecraft, NVision, and Ideon.