量子计算的关键挑战在于优化问题

compute

Analysts with McKinsey & Co in April said in a report that quantum computing had “reached a commercial tipping point,” noting that companies in industries like climate and life sciences, travel and logistics, and financial services already are using the emerging technology to optimize processes, test scenarios, and run simulations. They also said investors last year put $12.6 billion into quantum startups, another indication of the commercial embrace of the technology.

Such data shows that quantum computing has moved well beyond the “can it be done” point to the “when will it all happen” stage. It also feeds that message that John Roese, global chief technology officer and chief AI officer at Dell Technologies, and many others in the increasingly entangled quantum and IT industries are sending out, that quantum computing will not be a flip-the-on-switch moment but rather a steady march to increasingly scalable and practical fault-tolerant systems with constantly improving error-corrected qubits, architectures, software, and applications.

Traditional original design manufacturers like Dell, Hewlett Packard Enterprise, Cisco Systems, and IBM are putting the infrastructure parts in place for hybrid quantum-classical HPC datacenters. Cisco is building a quantum-focused networking architecture while Dell, HPE, and others are partnering with quantum vendors. At Dell Technologies World 2026 earlier this year, quantum company Equal1 put its RacQ (below) on display.

Likewise, HPE later unveiled partnerships with a range of quantum-focused companies, including IQM Quantum Computers (its Radiance quantum system is below), Quantinuum, QuEra, and Rigetti, as it builds a platform to integrate their systems with HPE’s Cray supercomputers.

As McKinsey’s data shows, early adopters already are seeing benefits, and those benefits and the numbers of companies seeing them will grow as the technology becomes more capable. Roese tells The Next Platform that users will begin seeing this in the near-term, noting that “before the end of this decade, so in like 2028, ‘29, we will likely have enough error-corrected logical qubits to go after the first batch of real problems, which are essentially molecular-level optimizations.”

By the beginning of the next decade, there should be enough fire power in quantum systems to address a second stage, which will focus on optimizations on a larger scale. Think of what airlines and other transportation businesses need to coordinate or the challenges that financial companies need to juggle. Complex optimization involves myriad computations that even the most powerful classical computers now would struggle to crunch through in a time- and cost-sensitive way and that, if addressed, could have resounding benefits from companies to their end users.

“The whole supply chain of the world is basically suboptimal because we don't have the computing to run large-scale, efficient optimization calculations” Roese says. “This is not little stuff. But in order to do that, it's bounded by the number of qubits. To represent the entire country of the United States and every road, you would need a very big quantum computer. And it doesn't quite map to the number qubits, but that data set is quite large, and it wouldn't fit into a 1,000-qubit system. You might need a million qubits or many millions to do a big optimization problem.”

That said, there are a number of architectures from the likes of Qarakal Quantum and Infleqtion to reduce the number of physical qubits needed to create a logical qubit, which would save on infrastructure and energy costs, lead to less noise, and allow for system designs that can better fit alongside classical HPC systems in datacenters.

Whatever the qubit count, the work that will be done on systems in these early days will have real-world results, even that done at the molecular level, Roese says.

“There are a bunch of foundational areas of science and the industrial base where the trick that makes a battery a battery better or a material better in some capacity is actually how you assemble the molecule that creates that material and a molecule is composed of different atoms those and atoms have different elements,” he says, adding that the ability to understand how to optimize them and put them together counts. “It's not just that you're connecting them, but how far apart should they be? What kind of binding power should they have, because atoms in a molecule are kind of gluey. They connect to each other.”

He points to the issue of rust in devices and systems made of steel, which “at the most rudimentary level” is caused in part because the molecules aren’t not properly oriented and that the problem could be fixed by changing the spacing between them.

“There are materials like stainless steel that are corrosion-resistant and it's not because they're steel,” Roese says. “There are many kinds of steel. Stainless steel is the orientation of the actual atoms in the molecule that are such that they actually produce that effect of being corrosion-resistant. Imagine if you could look at any material and you could say, 'Why does it corrode?' and understand that it's because some small portion at the atomic level is not optimized. Then you change the material and make it corrosion- free. The numbers I've seen, corrosion costs the global economy a trillion dollars a year.”

Traditional supercomputers may be able to work some of this out, but quantum systems with their exponentially larger effective compute power will be able to do it faster and at a greater scale. Other industries, such as drug discovery and batteries – making them longer-lasting – also fall under the category of molecular-level optimizations. Quantum computing is no longer a physics problem, but an engineering one. Solving such problems will be invisible to most people will have a significant impact in a lot of areas.

By 2030, quantum systems will be able to tackle optimizations at a larger scale, beyond molecules. Right now, the largest opportunity is in logistics, Roese says. It’s not an untapped area; D-Wave’s annealing quantum systems are aimed squarely at such optimization opportunities (though the company also is will build superconducting quantum systems), and he adds that Dell also has used annealers to address optimization in its supply chain. However, annealers are specific to optimization issues, though their use has raised the challenges large organizations face. Roese points to what he calls the airline gate problem as an example.

“I have 30 gates, I have 300 flights, I have 100,000 passengers,” he says. “Which combination would give me the best outcome? Making it even more, I have 100,000 passengers that all board those planes, and the planes all have different configurations. It cascades. You can't calculate that easily. You can manually try to calculate it, but the number of permutations you have to calculate would require supercomputers a very long time to do it.”

Transportation companies like FedEx, UPS, and DHL are another example. All use route optimization formulas to reduce the number of left turns drivers have to make on their routes, which saves time and gas costs.

“That is a very rudimentary optimization, but that's the best they could come up with because the other choice is to look at all the permutations of all the streets and all the possible things that could happen and figure out that it's not just take left turns,” Roese says. “There is a happy path that would give them a substantial improvement on the overall logistics of delivering packages, delivering gas, delivering anything. Then you cascade it. The delivery trucks are one layer, but what they're delivering is a commodity that comes from different sources.”

Such optimization capabilities can be applied to current and future environments that will only get more challenging. For example, by 2030, there will be exponentially more AI agents, and the use of autonomous vehicles will increase, creating much more complex environments.

“By 2030, how many autonomous agents do you think are going to be in the world?” Roese asks. “A trillion or something? It's going to be a lot. We'll just have all kinds of autonomous AI, both physical and logical running all over the place. What our experience to date has been is that they are predictable but unpredictable. They can be shaped, but when they start working as systems and teams, it gets very hard to understand them and to control them and to direct them. The problem is I'd have too many moving parts and I can't model it, I can't simulate it properly. You look and say, 'If I have a very large ecosystem of autonomous AI entities and I want them to work together,’ have a quantum system in the middle as the vehicle that tells them what to do as a system.”

The workloads that quantum systems will be able to addresses beyond those two phases will only grow in number and complexity, he says.

SPONSORED FEATURE: HPE and NVIDIA help organizations apply accelerated computing, software, enterprise infrastructure, networking, control plane, and services to the workloads and business outcomes that matter most.

October 1, 2026

September 30, 2026

AMD Buys World Labs For $8.2 Billion To Get AI Model Cred