Quantum could solve AI energy drain, says tech founder

Quantum Field Test in a Data Centre

Australian startup Diraq has placed a small silicon‑based quantum processor inside a commercial data centre in Sydney. The system, which houses eight physical qubits and draws under 20 kW of power, is the first of its kind to be operated alongside conventional servers in a real‑world environment. Diraq’s founder, Andrew Dzurak, says the device demonstrates that quantum hardware can run at modest power levels and could eventually relieve the massive electricity demand of AI workloads.

What the Prototype Can Do Today

• Eight physical qubits on a silicon chip, cooled by an integrated cryogenic system.
• Power consumption below 20 kW, comparable to a small rack of classical servers.
• Field‑testing in an Equinix data centre, providing early data on reliability, integration, and operational overhead.

The current machine is not error‑corrected; it operates with raw physical qubits and therefore cannot yet run fault‑tolerant algorithms. No benchmark results or performance metrics for AI tasks have been published, and the claim that it will “solve” AI’s energy drain is speculative.

Roadmap and Ambitions

Diraq outlines an aggressive scaling plan:

| Target Year | Qubit Count (Physical) | Intended Power Use | |-------------|------------------------|-------------------| | 2029 | 150 000 | ~200 kW | | 2031 | 2 000 000 | ~200 kW |

If achieved, such a system could meet DARPA’s 2033 benchmark for cost‑effective quantum advantage. However, the roadmap hinges on breakthroughs in qubit yield, error rates, and large‑scale integration that have not yet been demonstrated in peer‑reviewed literature.

Potential Impact on AI Energy Consumption

Dzurak argues that many AI problems are “better suited to quantum calculations,” implying that quantum processors could replace power‑hungry classical clusters. While quantum algorithms (e.g., for optimisation or sampling) may offer asymptotic speed‑ups, practical energy savings depend on:

1. Error‑corrected logical qubits that can run deep circuits reliably.
2. End‑to‑end software stacks that translate AI workloads into quantum‑friendly forms.
3. Demonstrated cost‑per‑operation that beats specialised classical accelerators.

At present, Diraq’s eight‑qubit device cannot address these criteria, so any claim of immediate AI energy relief is premature.

Outlook

The field test is a noteworthy engineering milestone—showing that silicon quantum chips can be housed in a commercial data centre and operate with modest power. It provides valuable data for scaling efforts and for the broader quantum‑AI dialogue. Nevertheless, the headline promise of solving AI’s energy crisis remains unsubstantiated until larger, error‑corrected systems are built and benchmarked against real AI workloads.

--- The article is based on an AFP interview and a Phys.org news release; no peer‑reviewed study or independent verification was cited.