The world is in a mad scramble for computing power. OpenAI alone has committed to building nearly $1.5 trillion worth of AI data centers. Big tech is following suit. Expected demand briefly pushed chipmaker Nvidia’s market value to $5 trillion in October, making it the first ever company to hit that mark. The fervor is so extreme that tech firms, including Google and Nvidia, would like to send data centers to space.
Notable amid the madness is a generational reversal of fortunes. Intel, a darling of the nineties that made its name in general-purpose chips, is struggling, while Nvidia, a maker of specialized chips, has grown into a historically dominant force.
There’s reason to believe companies and investors swept up in the AI craze may be overdoing it. But if the past is a guide, we’ll always want more computing power—it’s the bedrock of every major economic sector—and from GPUs to wafer-scale and neuromorphic chips, the focus on specialized computing is likely to continue.
There’s also something more radical coming: Quantum computers. Like other specialized chips, quantum computers won’t be practical for general computing. And they won’t run on your laptop or phone. But because they process information in a fundamentally different way, the industry believes quantum computing will be powerful in a range of niche but valuable areas, including AI and drug discovery.
The technology has been on the scene for years at this point, but quantum computers have yet to beat classical computers at useful tasks. They’re too finicky and error-prone. To solve these problems, researchers are laser-focused on two interrelated challenges: error-correction and scaling. Solving the former will yield quantum computers that reliably process algorithms nearly free of errors. The latter is focused on building machines big enough to do work no classical computer could manage.
The two trends depend on one another. As error-correction schemes improve, it’ll take fewer quantum computing components, called qubits, to get the job done. This means makers can get away with building smaller chips and reach their goals of practical machines sooner. But even as error-correction improves, quantum computers still need to scale up from today’s computers to include thousands of qubits or more.
Still, the field is making notable progress.
This month, Quantinuum unveiled a quantum computer called Helios with 98 physical qubits and, more importantly, 48 error-corrected (or logical) qubits. That 2:1 ratio is unique and impressive, UCLA’s Prineha Narang told The Wall Street Journal. Most error-correction schemes use far more physical qubits. Google, IBM, and Amazon—who are all chasing a different type of qubit—require 105, 12, and 9 physical qubits respectively per logical qubit, according to MIT Technology Review. Quantinuum plans to build a fully fault-tolerant machine with hundreds of logical qubits by 2030.
IBM is targeting a similar timeline for its own faul-tolerant quantum computer. But to scale its machines, the company is going modular. This is like adding multiple cores to a classical processor or linking chips in a data center. This month, they announced two new quantum chips that make progress on connectivity, error-correction, and modularity. IonQ, meanwhile, announced qubits with 99.9 percent fidelity.
We’ve also seen movement toward quantum computers that do useful work. In October, Google said its new algorithm, Quantum Echoes, can model the structure of a molecule 13,000 times faster than a top supercomputer. Although this isn’t yet the kind of commercial problem developers are targeting, it is another step in that direction. “It’s a meaningful technological advance,” Narang told The New York Times. “We have heard a lot about hardware advances in the field, and for a while, I worried that the algorithms would not keep up. But they have shown that this is not the case.”
The road ahead is still uncertain. The field’s leaders are pursuing a range of technologies, each with tradeoffs. Google, IBM, and Amazon use qubits made of tiny superconducting loops of wire. Quantinuum and IonQ make qubits from trapped ions. Atom Computing and QuEra use neutral atoms, and Xanadu is pursuing photons. Like IBM and Quantinuum, most players are looking to the end of the decade for fault-tolerant quantum computers that do practical work no supercomputer can handle.
All face technical hurdles between now and then, and no approach is a clear winner.
To better measure the field’s momentum and cut through hype, Darpa is running the Quantum Benchmarking Initiative. The QBI recently announced the first 11 companies to reach its second stage after describing “a utility-scale quantum computer concept that has a plausible path to realization in the near term.” Notably, labs pursuing each of the leading approaches made the cut. Said another way: There are currently multiple technologies that could win the race. To enter the third and final stage, labs will have to outline the roadblocks they’re facing and how they’ll overcome them.
With so many approaches in the running, it’s impossible to know which horse to back. But there’s growing confidence at least one will pay off. Quantum computing is “now less of a science experiment than an engineering problem,” Asa Fitch wrote for the Wall Street Journal, referencing a note from BNP Paribas analyst David O’Connor. O’Connor thinks the problem may be solved in the next three to four years.
A true signal quantum computers have arrived will be when they do the impossible, like crack Bitcoin-creator Satoshi Nakamoto’s $460-billion stash or model a high-temperature superconductor. “For the promise of quantum computers to be unlocked, we need to produce a new drug that we only know about because of quantum computers,” Narang said. “Then you can say that all the investment was worthwhile.”
-1.png?upscale=true&width=1200&upscale=true&name=Exp%20Primer_Email_Banner_Ad%20(2)-1.png)
