AI's Surging Energy Demand Sparks Data Center Efficiency Crisis

AI's ballooning energy consumption puts spotlight on data center efficiency

Georgia Institute of Technology provides funding as a member of The Conversation US. Artificial intelligence is growing fast, and so are the number of computers that power it. Behind the scenes, this rapid growth is putting a huge strain on the data centers that run AI models. These facilities are using more energy than ever. AI models are getting larger and more complex. Today's most advanced systems have billions of parameters, the numerical values derived from training data, and run across thousands of computer chips. To keep up, companies have responded by adding more hardware, more chips, more memory and more powerful networks. This brute force approach has helped AI make big leaps, but it's also created a new challenge: Data centers are becoming energy-hungry giants. Some tech companies are responding by looking to power data centers on their own with fossil fuel and nuclear power plants. AI energy demand has also spurred efforts to make more efficient computer chips. I'm a computer engineer and a professor at Georgia Tech who specializes in high-performance computing. I see another path to curbing AI's energy appetite: Make data centers more resource aware and efficient. Energy and heat Modern AI data centers can use as much electricity as a small city. And it's not just the computing that eats up power. Memory and cooling systems are major contributors, too. As AI models grow, they need more storage and faster access to data, which generates more heat. Also, as the chips become more powerful, removing heat becomes a central challenge. Cooling isn't just a technical detail; it's a major part of the energy bill. Traditional cooling is done with specialized air conditioning systems that remove heat from server racks. New methods like liquid cooling are helping, but they also require careful planning and water management. Without smarter solutions, the energy requirements and costs of AI could become unsustainable. Even with all this advanced equipment, many data centers aren't running efficiently. That's because different parts of the system don't always talk to each other. For example, scheduling software might not know that a chip is overheating or that a network connection is clogged. As a result, some servers sit idle while others struggle to keep up. This lack of coordination can lead to wasted energy and underused resources. A smarter way forward Addressing this challenge requires rethinking how to design and manage the systems that support AI. That means moving away from brute-force scaling and toward smarter, more specialized infrastructure. Here are three key ideas: Address variability in hardware. Not all chips are the same. Even within the same generation, chips vary in how fast they operate and how much heat they can tolerate, leading to heterogeneity in both performance and energy efficiency. Computer systems in data centers should recognize differences among chips in performance, heat tolerance and energy use, and adjust accordingly. Adapt to changing conditions. AI workloads vary over time. For instance, thermal hotspots on chips can trigger the chips to slow down, fluctuating grid supply can cap the peak power that centers can draw, and bursts of data between chips can create congestion in the network that connects them. Systems should be designed to respond in real time to things like temperature, power availability and data traffic. Break down silos. Engineers who design chips, software and data centers should work together. When these teams collaborate, they can find new ways to save energy and improve performance. To that end, my colleagues, students and I at Georgia Tech's AI Makerspace, a high-performance AI data center, are exploring these challenges hands-on. We're working across disciplines, from hardware to software to energy systems, to build and test AI systems that are efficient, scalable and sustainable. Scaling with intelligence AI has the potential to transform science, medicine, education and more, but risks hitting limits on performance, energy and cost. The future of AI depends not only on better models, but also on better infrastructure. To keep AI growing in a way that benefits society, I believe it's important to shift from scaling by force to scaling with intelligence.

AI's infrastructure problem is bigger than we think -- here's how to solve it

Creating AI data centers without smart tech wastes crucial opportunities The world is rushing to capitalize on the commercial and societal promise of AI. In the northeast of England, the recently announced Teesworks data center project promises to be Europe's largest data center. Across the Atlantic, Amazon's sprawling facilities in Indiana underscore how enterprises and governments are sprinting to build infrastructure for the AI era. The UK government's new Compute Roadmap, for example, calls for at least 6GW of AI-ready data center capacity by 2030 -- triple the current national footprint -- to keep pace with the US and other leading markets. But beneath this breakneck growth, a quieter crisis is emerging. The computational demands of AI tools may be racing ahead, but the infrastructure required to support it -- above all, power -- is trailing behind. An urgent question must be answered: how can the grid keep up with our desire to scale AI? Projects like Teesworks and Amazon's Indiana buildout are part of a global rush to shore up data center capacity. Yet this rapid buildout is exposing a fundamental mismatch. Even as AI's hunger for computing resources grows exponentially, there are high-profile harbingers of the potential bottlenecks introduced when national grids can't keep up. In Northern Virginia - the world's densest cloud hub - new AI and cloud projects have had to be paused due to a lack of electricity. Over in Ireland, data centers now consume more than 20% of national electricity, prompting proposals that they build their own private power lines. The UK, meanwhile, is relaxing planning rules for new transmission towers to speed up grid upgrades. This isn't a problem inherent to regional infrastructures - it's a global phenomenon brought about by putting the AI cart before the energy horse. And with AI's runaway growth unlikely to slow down anytime soon, the focus must be on finding solutions to reduce energy demands as much as expanding grid capacity. The data backs up the anecdotes. According to a Deloitte survey, 72% of US energy and data center executives view power capacity as extremely challenging as a result of widespread AI adoption, and 82% see innovation - not just grid expansion - as the only viable solution. Bloomberg Intelligence reports that there's now a 12-24 month gap between when data centers need power and when the grid can deliver it, a delay that is stalling growth in key markets. The issue is both technical and systemic. Even when renewable energy is available - such as wind power from Scotland - it often cannot reach the data centers that need it most, thanks to constrained transmission infrastructure. We face a kind of energy Catch-22: the need for more energy is desperate, but the energy we generate cannot always be used where it is required. The problem is compounded by the fact that conventional data center hardware is not designed to be energy-efficient at the scale now demanded by AI workloads. The solution is not simply to build more data centers and expand the grid accordingly, but to also rethink the fundamentals of computing infrastructure. The investment gap is threefold: we need more data centers, yes, but also better grid access, accelerated renewable integration, and - critically - a new generation of energy-efficient hardware within the data centers themselves. Moore's Law, which drove decades of exponential growth in computing, is reaching its limits. AI demands something more radical. The industry must look to technologies such as analogue computing, neuromorphic chips, and especially light-based (all-optical) architectures that eschew the costly energy conversions of current electro-optical networks. These innovations promise not just marginal gains, but step changes in energy efficiency -- delivering the performance needed for AI workloads while slashing the electricity required per calculation. At the moment, we measure AI progress in benchmarks, parameters, and flops. But this is a flawed metric if we ignore the energy cost of each inference. The industry must now prioritize "watts per task" as much as it does "exaflops". This is not just about engineering, but about ethics: as AI becomes central to fields from healthcare to climate science, unchecked growth in energy demand risks both the planet and the public's trust in AI's benefits. Solving the energy challenge is not optional -- it is existential for the AI industry. The International Energy Agency (IEA) warns that electricity demand from data centers worldwide is set to more than double by 2030, with AI at the heart of the surge. Without a shift towards smarter, more efficient infrastructure, we risk both environmental harm and the slowing of AI's transformative potential. Every week, new forecasts predict exponential AI growth and the accompanying environmental strain. The answer is not to slow progress, but to accelerate investment in the technologies that can break the link between AI expansion and energy consumption. This is a call to action for the entire industry -- for tech leaders, policymakers, and researchers to collaborate on global standards for efficiency, to back breakthrough research in energy-efficient hardware, and to ensure that the infrastructure of the future is designed for the demands of the present. The question is no longer if AI will change the world, but whether we have the world to sustain AI's rise. The time to invest in smarter, not just bigger, foundations is now. We list the best IT infrastructure management services.

AI's ballooning energy consumption puts spotlight on data center efficiency

Artificial intelligence is growing fast, and so are the number of computers that power it. Behind the scenes, this rapid growth is putting a huge strain on the data centers that run AI models. These facilities are using more energy than ever. AI models are getting larger and more complex. Today's most advanced systems have billions of parameters, the numerical values derived from training data, and run across thousands of computer chips. To keep up, companies have responded by adding more hardware, more chips, more memory and more powerful networks. This brute force approach has helped AI make big leaps, but it's also created a new challenge: Data centers are becoming energy-hungry giants. Some tech companies are responding by looking to power data centers on their own with fossil fuel and nuclear power plants. AI energy demand has also spurred efforts to make more efficient computer chips. I'm a computer engineer and a professor at Georgia Tech who specializes in high-performance computing. I see another path to curbing AI's energy appetite: Make data centers more resource-aware and efficient. Energy and heat Modern AI data centers can use as much electricity as a small city. And it's not just the computing that eats up power. Memory and cooling systems are major contributors, too. As AI models grow, they need more storage and faster access to data, which generates more heat. Also, as the chips become more powerful, removing heat becomes a central challenge. Cooling isn't just a technical detail; it's a major part of the energy bill. Traditional cooling is done with specialized air conditioning systems that remove heat from server racks. New methods like liquid cooling are helping, but they also require careful planning and water management. Without smarter solutions, the energy requirements and costs of AI could become unsustainable. Even with all this advanced equipment, many data centers aren't running efficiently. That's because different parts of the system don't always talk to each other. For example, scheduling software might not know that a chip is overheating or that a network connection is clogged. As a result, some servers sit idle while others struggle to keep up. This lack of coordination can lead to wasted energy and underused resources. A smarter way forward Addressing this challenge requires rethinking how to design and manage the systems that support AI. That means moving away from brute-force scaling and toward smarter, more specialized infrastructure. Here are three key ideas: Address variability in hardware. Not all chips are the same. Even within the same generation, chips vary in how fast they operate and how much heat they can tolerate, leading to heterogeneity in both performance and energy efficiency. Computer systems in data centers should recognize differences among chips in performance, heat tolerance and energy use, and adjust accordingly. Adapt to changing conditions. AI workloads vary over time. For instance, thermal hotspots on chips can trigger the chips to slow down, fluctuating grid supply can cap the peak power that centers can draw, and bursts of data between chips can create congestion in the network that connects them. Systems should be designed to respond in real time to things like temperature, power availability and data traffic. Break down silos. Engineers who design chips, software and data centers should work together. When these teams collaborate, they can find new ways to save energy and improve performance. To that end, my colleagues, students and I at Georgia Tech's AI Makerspace, a high-performance AI data center, are exploring these challenges hands-on. We're working across disciplines, from hardware to software to energy systems, to build and test AI systems that are efficient, scalable and sustainable. Scaling with intelligence AI has the potential to transform science, medicine, education and more, but risks hitting limits on performance, energy and cost. The future of AI depends not only on better models, but also on better infrastructure. To keep AI growing in a way that benefits society, I believe it's important to shift from scaling by force to scaling with intelligence. This article is republished from The Conversation under a Creative Commons license. Read the original article.

The World Faces a $2T Data Center Challenge in the Age of AI

The capacity demand is such that the world would halt if we did not find a global consensus and effective means to power up data centers. | Credit: Bruna Santos, Pexels. * AI's rapid growth is driving an unprecedented surge in global data center demand, requiring trillions in new infrastructure investment. * Meeting this demand poses major challenges for energy, sustainability, and resource allocation worldwide. * Governments, industry leaders, and chipmakers must collaborate on sustainable solutions to power AI while minimizing emissions. The emergence of AI has led to an immense increase in data center capacity needs. The capacity demand is such that the world would halt if we did not find a global consensus and effective means to power up data centers. Projections from analysts, governments and major data center developers all point to a rapid expansion of the current global data center capacity over the next several years. How much would this cost, and how would it be used? Let's break it down. The Modern-Day Data Center Current estimates call for $9-15 million per megawatt to develop a modern data center. This cost includes land acquisition, building design, building construction, structured cabling, power and cooling systems, and securing the facility. The actual number will vary depending on actual materials costs, local labor costs, local regulatory compliance, and unforeseen delays. Using this range, we see that a new 10MW facility would cost between $90 and $150 million. Scaling up to 100MW - something that usually requires multiple buildings - brings the price tag to almost $1 to $1.5 billion. The gigawatt-and-beyond facilities now being discussed by many developers bring the cost to almost $10 to $15 billion and beyond. A global construction boom from 55GW to 300GW thus caps out at around $2-2.5 trillion. These latter amounts are similar to the entire economy of Brazil, Canada, or Italy. They would constitute about 2% of the world's GDP. Clearly, it would take several years to achieve construction at this level and put hundreds of massive new data centers into operation. Is there enough opportunity out there to justify such a heavy lift? The AI Challenge Continues To Grow AI proponents will tell you "yes." The large models being used to develop today's generative AI (or GenAI) platforms require computing resources costing in the hundreds of millions, even billions, of dollars. As these platforms continuously move into the mainstream of consumer search and corporate/government research, we can expect the demands they make to grow exponentially. Additionally, the next generation of "agentic AI" featuring more proactive AI platforms that act as agents for automated tasks and custom requests can be expected to make further substantial demands. A third area is government and scientific research. Here AI is put to use to work on complex modeling and simulation problems found in climatology, other earth sciences, epidemiology, medical diagnostics, military research, and cryptography across all consumer, business, and government functions. A fourth area is, of course, cryptocurrency development. Although Bitcoin's energy-intensive proof-of-work architecture is giving way to less intensive proof-of-stake protocols in blockchain and stablecoin development, the near-future demands of the blockchain/cryptocurrency world should not be underestimated. Demands on Electricity Grids But the challenge of simply building enough new data centers to meet all of these needs does not end with the data centers themselves. Enough power must be available for them, and as much of that power as possible must be sustainable. In 2024, the International Energy Agency (IEA) estimated that around 1.5% of the world's electricity served data centers. It's likely this is now higher at around 1.7-1.9%. The IEA estimates that only about half of that demand is met by renewables. In developing countries, data centers account for more than 20% of electricity demand growth to 2030. There are valid concerns about this level of energy use and the impact of it on efforts to curb global greenhouse gas emissions. It's critical to note that this energy usage is very unbalanced and skewed. The United States, for example, has 44% of the world's data servers , with a very large concentration within a single range of Northern Virginia in the Washington, D.C. area. Energy constraints have already emerged as an issue there, likewise in Chicago, Dallas, Phoenix, and California's Silicon Valley. Texas recently legislated new policies to cut back data center power consumption during peak periods. There are also concerns about having enough water to cool large data centers, particularly in Arizona in the US, and arid regions elsewhere in the world. So what can we do? Major Chip Producers For their part, major chip producers including NVIDIA and TSMC are working daily to make their chips more efficient - to use less electrical power even as they deliver more computing power. Most of the world's major data center operators long ago committed to Net Zero energy usage strategies, and though they often have to restate their goals and processes, it is rare to see a large new data center being planned today without a commitment to using sustainable energy. The world seems to be moving in the direction of developing more nuclear energy, which although not a renewable resource, is a sustainable source that emits steam as its byproduct. A declaration at the United Nations COP28 meeting in Baku, Azerbaijan, during December 2023 was signed by more than 20 nations (including the United States), with a mission to triple the amount of nuclear energy by 2050. Viewing all of these efforts and more, the International Data Center Authority (IDCA) provides a metric that shows a correlation between data centers and related digital infrastructure and economic efficiency. Rather than simply looking at a nation's total emissions or its per capita (per-person) emissions, IDCA examines emission levels compared to the productivity of a nation's economy. This number, expressed as the number of tons of CO2 produced by each $1 billion of nominal GDP, finds a world average of about 350 tons of CO2 produced per $1 billion of GDP. But in the United States, this number drops to 177, about half of the world average. Germany comes in at about 150, as does Brazil, France and the U.K. at 100, and Scandinavian nations under 100. China and India, however, produce around 700 tons of CO2 for each $1 billion of their economy. Much of the developing world is also at a level less efficient than the world average. The kicker is that AI, when put to proper use in making searches more efficient, in automating systems from transportation and logistics to manufacturing and customer service, and to creating a new class of high-end skilled service employment, should force a downward trend in emissions levels by developing more efficient economies. There is a path toward developing the massive infrastructure required to support AI's demands, but it will require concerted efforts by governments to take the lead in sustainability, regulatory consistency and AI-driven economic efficiency. These are highly complex, difficult issues unsuitable for slogans or truncated thinking. However, several nations are already proving that a well-functioning 21st-century economy will place data centers in a key role as we all work toward reducing emissions and reaching Net Zero.