Artificial intelligence (AI) is becoming part of daily life for many people around the world. At the individual level, people are increasingly using AI models for search queries. While Google still dominates the search market, ChatGPT has posed the most significant threat to its dominance.
On a business level, no industry is left out, from agriculture to healthcare, from finance to entertainment, organizations around the world are integrating AI into their daily operations.
The world's demand for and use of AI is expected to grow exponentially in the coming years, so technology companies are responding to this development by building massive data centers. But this growth comes at a cost: energy consumption, economic costs and environmental impact. Traditional computing simply cannot keep up with growing computing and energy demands. To sustain the AI revolution, we must rethink the physics of modern computing.
Energy issues
Even without considering AI, electronic computing is at a critical juncture. Moore's Law is failing, Dennard scaling has broken down, and the result is the proliferation of "dark silicon," the portions of transistors on athat must remain unpowered or idle to avoid overheating.
Training a large AI model is no easy task. Large language models (LLMs) are trained on massive amounts of data and have trillions of parameters. They predict, measure, adjust, and repeat the process billions of times. It is estimated that the computing power required to train AI models will double every six months.
Processing and moving such large amounts of data requires massive parallelism and power. In traditional computing, higher power requires higher density systems. Higher density means more resistance, and more resistance means more heat. This forces data centers to shift a lot of energy from computing to cooling, with up to 40% of total data center energy consumption used to prevent server meltdowns.
The infrastructure that supports AI is already struggling, and it's clear that traditional computing can no longer support future development.
Economic issues
Data center operators are facing a financial conundrum: either limit compute density to what their current cooling facilities can handle, hampering their business capabilities, or push thermal limits, causing accelerated aging of hardware and components, increasing operating expenses and waste.
In addition, the cost of building new data centers is also very high - McKinsey predicts that US$5.2 trillion in investment will be required by 2030. If data centers continue to rely on traditional computing, investing in inefficient infrastructure will be a huge financial risk. Ordinary consumers are also affected by poor economic conditions; as AI puts unprecedented pressure on the grid and data center power demand rises, electricity prices are rising. These costs are passed on to surrounding households in the form of rapidly rising electricity bills.
