The rapid advancement of artificial intelligence is creating a silent crisis. While generative AI tools like ChatGPT and Midjourney capture the public imagination, they are powered by a voracious and unsustainable appetite for computational resources. This demand is pushing our global data centre infrastructure — and our energy grids — to a breaking point.
According to Goldman Sachs, power usage by data centres is projected to increase by 160% by 2030, driven almost entirely by AI workloads [1]. This surge is not a distant problem; it’s happening now. Each query on a platform like ChatGPT is estimated to consume at least ten times the energy of a traditional Google search, and the cost of training a single large AI model can already run into the tens of millions of dollars [2].
This exponential growth in demand is creating a three-pronged crisis: economic, environmental, and logistical.
The Unsustainable Economics of Scaling AI
The cost of building and operating the infrastructure required to power the next generation of AI is staggering. McKinsey estimates that a $7 trillion race to scale data centres is underway, a figure that highlights the immense capital investment required [3]. This isn’t just about building more servers; it’s about a fundamental shift in the economics of computing.
As Kirk Bresniker, Chief Architect at Hewlett Packard Labs, noted, we are heading for a hard ceiling. Sometime between 2029 and 2031, the cost of resources to train a single, state-of-theart AI model is projected to surpass the entire U.S. GDP [2]. This trend is not just unsustainable; it’s a barrier to innovation. If only a handful of trillion-dollar companies can afford to train foundational models, it stifles competition and concentrates technological power.
“If something is provably unsustainable, then it’s inherently inequitable. As we look at pervasive and hopefully universal access to this incredible technology, we have to be looking into what we can do. What do we have to change?”
— Kirk Bresniker, HPE Fellow & VP [2]
This economic pressure forces companies to make difficult choices, often prioritizing shortterm performance gains over long-term efficiency, further exacerbating the problem.
Environmental Impact and Energy Consumption
The environmental cost of this AI boom is equally alarming. The projected increase in data centre power consumption means that by 2030, AI alone could account for 8% of total U.S.
electricity demand [3]. This has led to unprecedented partnerships, with a third of U.S. nuclear power plants now in talks with tech companies to directly power new data centres [2].
This isn’t just about electricity. The massive amount of heat generated by densely packed GPUs requires intensive cooling, often using vast quantities of water. The entire supply chain, from manufacturing specialized silicon to constructing new facilities, carries a significant carbon footprint.
| Metric | Projection | Implication |
| Data Center Power Demand | +160% by 2030 | Strain on national energy grids |
| AI Share of U.S. Power | 8% by 2030 | Increased carbon emissions |
| ChatGPT Query Energy | 10x a Google Search | Exponential growth in operational energy use |
This trajectory puts the tech industry on a collision course with global climate goals. Without a more efficient computational paradigm, the progress of AI could be directly at odds with environmental sustainability.
The Physical and Logistical Limits
Beyond cost and energy, we are hitting physical limits. Data centres are becoming so large and power-hungry that they are difficult to site and build. They require access to massive power generation, high-speed fibre networks, and often, significant water resources for cooling. This has led to a construction boom, but even this has its limits.
Furthermore, the reliance on a specific type of hardware—GPUs from a small number of vendors—creates supply chain bottlenecks and geopolitical risks. The demand for these specialized chips far outstrips supply, leading to long wait times and inflated prices.
This is not a problem that can be solved by simply building more of the same. The current approach of scaling classical computing hardware is a brute-force method that is rapidly approaching its practical and economic limits.
A New Path Forward
The AI infrastructure crisis is not a problem of engineering; it’s a problem of physics. The limitations of classical computing are becoming apparent, and a new approach is needed. This is where hybrid quantum-classical computing offers a viable, long-term solution.
By identifying the specific, computationally-intensive parts of AI workloads that are suitable for quantum processors, we can offload these tasks to a more efficient paradigm. This doesn’t replace classical computing but augments it, allowing each type of processor to do what it does best.
Quantum Links AI was founded on this principle. Our intelligent routing platform automatically determines the optimal computational path for any given task, seamlessly integrating the power of quantum computing into existing AI workflows without requiring specialized expertise.
This hybrid approach is not just about achieving faster results; it’s about building a more sustainable, efficient, and equitable future for artificial intelligence. It’s about finding a way to continue the incredible progress of AI without bankrupting our economies or our planet.
In our next post, we will explore exactly how quantum computing can alleviate these pressures and provide a practical solution to the AI cost crisis.
References
[1] Goldman Sachs. (2024). AI to Drive 160% Increase in Data Center Power Usage. [2] VentureBeat. (2024, July 25). AI training costs are growing exponentially — IBM says quantum computing could be a solution. https://venturebeat.com/ai/ai–training–costs–aregrowing–exponentially–ibm–says–quantum–computing–could–be–a–solution/
[3] McKinsey. (2025, April 28). The cost of compute: A $7 trillion race to scale data centers.
https://www.mckinsey.com/industries/technology–media–and–telecommunications/ourinsights/the–cost–of–compute–a–7–trillion–dollar–race–to–scale–data–centers
No comment yet, add your voice below!