Google pays publishers for AI answers
Benedikt Langer
4 min read Google's “AI contribution pilot” pays around 100 digital publishers for the extent to ...
Read Article
7 min read
Artificial intelligence keeps getting more capable and is driving electricity demand in data centers. Quantum computing is still early, but it is often framed as a possible way out of the efficiency dilemma. IBM sees the key in diamonds instead of helium.
At a glance
- Situation: According to the IEA, data centers accounted for about 1.5 percent of global electricity use in 2024 and are growing clearly faster with AI – not 30 percent, but with real local grid pressure.
- Leverage: Quantum machines often draw only tens of kilowatts in operation (cooling included), yet they solve only narrow problem classes. Diamond-based approaches aim to avoid costly helium cooling.
- Consequence: Steer in parallel: hybrid HPC roadmap, data-center energy KPIs and post-quantum crypto – instead of waiting for one quantum leap as the AI power rescue.
RelatedPost-quantum cryptography is pressing enterprise IT / Token OPEX: inference drives the bill, not seat budgets
Finnish startup IQM Quantum Computers opened Munich’s first quantum data center with rentable compute capacity in June 2024, according to Computerwoche. At the time, two quantum computers were mainly available for research, later also for business and government. Plans called for scaling to twelve systems. Anyone wanting access had to budget about 1,800 euros per hour, according to co-founder and CEO Jan Goetz.
Quantum computing is making progress. It still needs far more than the usual few hundred to around 1,000 physical qubits to run more complex calculations with low noise and approach supercomputer performance, as Digital Chiefs already argued in 2024. At the Quantum Summit in late 2023, IBM broke its own record and presented a comparatively low-noise processor with more than 1,000 qubits (Condor with 1,121 qubits), as covered by cloudmagazin. The roadmap at the time envisioned systems with 2,000 qubits and more than a billion logic gates; IBM has since refined that path toward error-corrected logical qubits.
Double-digit kW
Many superconducting quantum systems draw only low double-digit kilowatts in operation – orders of magnitude below classical supercomputers. GPU AI stays on classic clusters today.
Digital Chiefs framing – quantum computing and AI energy demand
What is quantum computing? Quantum computing uses superposition and entanglement of qubits to attack selected optimization, simulation and cryptography problems differently from classical bits. In practice, error-corrected logical performance matters more than raw physical qubit counts. For enterprise IT the lever stays hybrid: chosen workloads next to HPC and GPUs as targeted complements to AI infrastructure.
Many opportunities, and new risks
The D-Wave quantum annealer that went live in Jülich in 2022 already offers more than 5,000 qubits. These quantum bits behave differently from “normal” bits, which as the smallest unit of computation can only be on or off, mathematically 1 or 0. Qubits can, through superposition and entanglement, occupy many intermediate states between 0 and 1, which can lift processing speed for the right problems onto a new level – if error rates can be controlled.
Logical qubits that bundle several physical qubits are meant to cut error rates sharply. In the wrong hands they could also help break conventional encryption. According to the BSI development report from late 2023, attacking a 2048-bit RSA system would require about 4,098 logical qubits; the related estimate for physical qubits then stood around 20,000 and is highly model-dependent in newer studies – from under 100,000 into the millions. For CIOs the takeaway is clear: post-quantum migration runs in parallel with the hardware story, not only “when the quantum leap arrives”.
A new application area: AI
As the field matures, quantum computers could open new options in artificial intelligence – and for tightly defined workloads help ease its energy hunger. Putting large language models into production already consumes visible power and capex because of the number of GPU accelerators required; reports of scarce Nvidia capacity and rising inference costs have been a standing theme in enterprise IT since 2023.
In an April 2024 white paper on high performance computing (HPC), market researcher IDC expected quantum computing to gain ground in HPC and classical hybrid environments. VINCI Energies and subsidiaries Axians and Actemium also positioned themselves in a joint project with D-Wave and QuantumBasel in uptownBasel, helping build Switzerland’s first commercially used quantum hub. One goal was maximum performance and usability with as little energy and material overhead as possible.
AI is driving data-center power demand sharply higher
The often-quoted claim that data centers already use four to five percent of global energy and could soon reach 30 percent with AI does not hold up. The International Energy Agency (IEA) estimates data-center electricity use in 2024 at about 415 terawatt-hours – roughly 1.5 percent of global electricity demand. In the base case of the IEA “Energy and AI” analysis, that figure roughly doubles to about 945 TWh by 2030, just under 3 percent of global electricity use. Absolute pressure in grids and AI-dense locations remains real: accelerated servers grow much faster than conventional servers, according to the IEA.
Training large models keeps hundreds of GPU accelerators under high continuous load in the high three-digit watt range per card – around the clock, often for weeks. Quantum computers also need energy, mainly to cool helium near absolute zero (0 Kelvin or -273 °C). In operation, many superconducting systems still sit in the low tens of kilowatts and thus far below classical supercomputers. Industry reports said that running China’s then top-tier Tianhe-2 supercomputer in 2019 needed roughly the output of a small power-plant block. Annual energy costs for such machines sit in the millions of euros. A quantum computer with about 25 kW continuous load lands – depending on the electricity price – in a very different order of magnitude, often low five-digit euros per year. Important: that is an operating comparison, not proof that quantum hardware can replace AI training today.
Diamonds instead of helium could be the rescue
Real energy savings in production are still limited. As Dr. Mark Mattingley-Scott, then IBM Quantum Ambassador Leader EMEA & AP at IBM Germany, told Digitale Welt, helium-cooled systems also draw a lot of power – diamond-based quantum computers less so. Made synthetically, they could become a cheaper alternative to complex helium cooling and need less infrastructure. A diamond-based quantum accelerator in an autonomous vehicle would then use only a few hundred watts, “and those are only prototype figures,” Mattingley-Scott said. He calls green quantum computing essential and argues for an industry-wide standard.
In any case, quantum computing offers potential – mainly as a hybrid building block, not as a blanket AI power substitute. Fraunhofer FOKUS in Berlin worked with partners in the PlanQK project on quantum-supported AI; a classic use case remains fraud detection in banking, where classification and prediction meet. Christopher Savoie, CEO of Zapata AI, pointed in Computerwoche to new cancer drugs at the University of Toronto that would not have been possible without a quantum-based model. At the same time, the energy hunger of classical AI infrastructure remains the operational problem CIOs must manage today: capacity, PUE, inference cost and site power. Quantum computing is a strategic research and portfolio lever – the 2026 data-center bill is still dominated by GPUs and cooling.
Frequently asked questions
What is a qubit?
A qubit (quantum bit) is the smallest information unit of a quantum computer. Unlike a classical bit, which stores only 0 or 1, a qubit can carry multiple state combinations at once through superposition and entanglement. That property enables massive parallelism for suitable algorithms – and also makes qubits extremely error-prone and cooling-intensive.
Can quantum computing really cut AI’s energy hunger?
Only for tightly defined problem classes and only once error-corrected systems scale in production. Today’s AI workloads run on GPU clusters; quantum hardware does not blanket-replace that training and inference. The energy edge appears mainly versus supercomputers on selected optimization and simulation tasks, not in everyday chatbot use.
Why do many quantum computers need helium cooling?
Superconducting qubits operate near absolute zero. Cryogenics (helium, dilution refrigerators) dominate power use – often around 10 to 25 kW per system. Alternative modalities such as ion traps, neutral atoms or diamond-based approaches aim to cut or avoid that cooling load.
What do diamond-based quantum computers mean in practice?
They use defects in synthetic diamonds (such as nitrogen-vacancy centers) as qubits and are designed to need far less cryogenics. IBM representatives see a path to more compact, lower-energy accelerators. As of 2026 this remains largely prototype and research territory rather than a ready replacement for GPU farms.
What should CIOs prioritize alongside the quantum story?
Three tracks: (1) make data-center and inference costs transparent, (2) pick hybrid HPC use cases with clear ROI, (3) inventory and migrate post-quantum cryptography. Waiting only for the quantum leap steers neither power nor crypto risk.
Read more on Digital Chiefs
Digital ChiefsLower Costs, Higher Productivity, More Environmental Protection – German Agriculture Relies on AIDigital ChiefsSmart City: Revealing the Reality Behind the Paper GlitzDigital ChiefsAchieving Greater Energy Efficiency in Data Centers Through Lights-Out ManagementMore from the MBF Media network
cloudmagazin
mybusinessfuture
Image source: AI-generated (July 2026)
Translated from the German original using artificial intelligence. The German version is authoritative.