Diraq Says Quantum Computers Could Cut AI’s Energy Use For Data Training - Quantum Zeitgeist

Some new artificial intelligence models are expected to exceed 1 MW, consuming more electricity than an average US home uses in two years, as global data center electricity generation surges toward 1,000 terawatt-hours by 2030. While much attention focuses on optimizing power grids or molecule design, creating high-quality training data for AI, a key driver of this energy demand, is a less discussed application where quantum computing could offer a solution. Diraq argues that quantum computers can fill this training data gap, potentially replacing energy-intensive AI racks with calculations performed with ease, though quantum machines will still work alongside classical systems. AI Training Drives Quantum Computing Exploration The creation of high-quality training data represents a largely overlooked application for quantum computing, potentially addressing a key driver of artificial intelligence’s escalating energy demands. Classical computers struggle with complex simulations, resulting in a scarcity of robust data used to train AI models for tasks like molecular design and power grid optimization, and this lack of data forces reliance on energy-intensive classical compute to generate approximations. Diraq proposes that utility-scale quantum computers could perform these calculations, effectively replacing the racks of GPUs currently used for approximate training data generation, and reducing the overall energy footprint. This approach differs from the more publicized applications of quantum computing, such as molecule design, by focusing on the upstream problem of data creation rather than downstream problem solving. The company’s current system, containing eight qubits, is not intended to outperform conventional GPUs on commercial applications, but rather to demonstrate the feasibility of integrating quantum processing with existing high-performance computing infrastructure, Diraq says. The company states a focus on practical integration rather than immediate computational advantage. The system’s self-contained cryogenic cooling and control electronics allow it to operate within standard data center environments, avoiding the need for specialized quantum facilities. Diraq’s strategy centers on scaling qubit density on silicon chips fabricated using standard CMOS foundry lines, aiming for millions of qubits at a cost of under a dollar per qubit. This contrasts with other quantum architectures that require extensive peripheral infrastructure, as the company argues that increasing on-chip qubit count is the primary constraint, not qubit quality. Founded in 2022 and headquartered in Sydney, Australia, Diraq has raised US$140M+ to pursue this silicon-spin qubit technology. A September 2025 Nature paper demonstrated above 99 percent single- and two-qubit fidelity on industry-fabricated silicon unit cells, surpassing the surface-code fault-tolerance threshold, and validating the potential of this approach, according to the company. The economic viability of quantum computing will ultimately depend on delivering valuable computation without a proportional increase in power consumption, and Diraq’s focus on integrating its processors into existing facilities addresses this concern. Avoiding the construction of new, dedicated quantum facilities eliminates significant embedded emissions and costs. “Quantum computing will ultimately be judged by the problems it solves, but it will be adopted according to the economics of delivering those solutions,” the company asserts, highlighting the importance of practical considerations alongside scientific advancements. Diraq is currently integrating its quantum processor with a Dell high-performance computing cluster for low-latency hybrid workflows, further demonstrating its commitment to seamless integration with conventional computing resources.

Silicon Spin Qubits Enable Compact Quantum Systems Silicon spin qubits offer a pathway to compact quantum systems by minimizing the need to scale peripheral infrastructure. Diraq’s approach, forming qubits by modifying standard computer chip transistors, allows for potentially millions of qubits on a chip comparable in size to those holding eight qubits, the firm reports. This density contrasts sharply with other architectures, like superconducting and photonic systems, which scale by multiplying surrounding equipment and demanding expansive cryogenic or optical plants. The benefit of this approach extends beyond chip size; it directly addresses the escalating energy demands of artificial intelligence. While current AI workflows consume substantial power for tasks quantum computers are able to handle efficiently, Diraq’s systems are designed to draw under 20 kW for the complete quantum system, comparable to a standard AI rack. “Quantum does not automatically mean energy efficient,” the company notes, emphasizing that every architecture must reach millions of physical qubits, but the path to that scale varies considerably. Diraq’s strategy isn’t about immediate computational advantage, but about building a sustainable foundation for future quantum computing. The company argues that the binding constraint on quantum computing is qubit number, not necessarily qubit quality, and that increasing on-chip density is the key to avoiding the infrastructure demands of other approaches. This focus on density is supported by a partnership with Imec, which successfully fabricated SiMOS spin qubits using extreme-ultraviolet lithography, a technique central to high-volume chip production, at a sub-10nm pitch. This integration, alongside a collaboration with Iceberg Quantum on Pinnacle qLDPC architecture, highlights a commitment to seamless interaction between quantum and classical computing. The company’s roadmap projects thousands of physical qubits in a commercial product by 2029, and tens of millions by 2033. Diraq’s 300kW Target for Utility-Scale Quantum Diraq is targeting a 300kW total system power draw for its utility-scale quantum computer, a figure intended to align with the operational constraints of modern data centers. This target accounts for both the quantum processing unit itself, expected to consume around 125kW, and the classical compute infrastructure required for interfacing and control. The company positions this energy efficiency as important for the widespread adoption of quantum computing, particularly as demand from artificial intelligence applications surges, Diraq reports. According to Olivier Ezratty, co-founder of the Quantum Energy Initiative, comparing power consumption across different quantum modalities is essential, and Diraq’s target places it competitively against next-generation GPUs. A key driver for exploring quantum computing lies in its potential to alleviate the energy burden of AI training. This doesn’t envision quantum computers supplanting GPUs entirely, but rather working alongside them, with classical processors handling tasks like orchestration, error correction, and AI workloads not suited for quantum architectures. “Quantum computers will work alongside classical systems, with GPUs and other processors handling orchestration, error correction, AI workloads, and the many computations quantum machines aren’t designed to perform,” the company notes. Currently, Diraq has deployed an eight-qubit system in a commercial data center, drawing under 20kW for the complete quantum system. This installation, while not intended to outperform GPUs on commercial applications, represents a deliberate step in the company’s product roadmap, the company’s account states. The system’s compact design, with self-contained cryogenic cooling and control electronics, allows for deployment alongside conventional computing equipment without requiring a dedicated quantum facility. But as we’ve discussed before, more advanced codes like qLDPC will give rise to up to 10,000 logical qubits for our utility-scale system. Diraq has raised US$140M+, including $100M from Australian Government, UNSW Sydney, and a $20 million investment from the National Reconstruction Fund Corporation. Current AI Power Demands Highlight Quantum Potential This surge in energy use isn’t limited to model operation; creating high-quality training data, a less discussed application, represents a significant driver of AI’s overall energy footprint. Diraq contends that quantum computing offers a potential solution, not by replacing GPUs entirely, but by handling specific tasks currently reliant on brute-force classical computation. Diraq distinguishes between qubit modalities, asserting that not all quantum computers inherently offer energy efficiency. This contrasts with approaches focused on increasing on-chip qubit density, which minimizes the need for extensive external infrastructure. Power draw measurements often focus on the system itself, but Diraq emphasizes that the infrastructure required to house and operate a quantum computer is an integral part of the overall energy equation. Technologies demanding purpose-built facilities will inevitably consume significantly more power than those that can integrate seamlessly into existing data centers, even before the first qubit is initialized. This strategy is supported by recent advancements in silicon spin qubit fabrication, including work with Imec utilizing extreme-ultraviolet lithography, a technique employed in high-volume chip production, to create sub-10nm pitch double-dot systems. Its roadmap targets tens of millions of physical qubits by 2033, positioning silicon spin qubits as a potentially viable solution for energy-intensive AI training tasks. Source: https://www.diraq.com/newsdesk/could-quantum-fix-ais-energy-problem More like thisArtificial IntelligenceA new memristor senses humidity like human skin, aiding AIArtificial IntelligenceNUS and partners launch a Quantum × AI conference with 300 expertsQuantum ApplicationsIonQ tests quantum AI on real satellite image dataArtificial IntelligenceIntel, HPE and Multiverse Computing double speech-to-text speedStay currentSee today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals. Tags: