AI's Energy Demands Versus Grid Realities
Artificial intelligence (AI) is advancing rapidly, but energy infrastructure is a critical bottleneck. As computational capabilities expand exponentially, their energy demands are surging beyond what traditional electrical grids can accommodate. This coming crisis threatens to constrain AI innovation and significantly increase carbon emissions at a time when technology leaders have uniformly committed to ambitious sustainability goals.
The Escalating Energy Crisis In AI Computing
The energy demands of AI workloads are growing at an unprecedented pace. Nvidia CEO Jensen Huang recently noted that future AI server racks could require up to 600kW of power, a dramatic leap from today's standards and a fivefold increase in density from what we are seeing today.
Deloitte projects that global data center electricity consumption will reach 536 terawatt-hours (TWh) in 2025, representing about 2% of worldwide electricity usage. By 2030, this figure could double to 1,065 TWh due to power-intensive AI applications like generative AI training and inference. Some estimate that a single ChatGPT query consumes the same amount of electricity as powering an LED light bulb for 20 minutes, and Goldman Sachs has estimated that it requires nearly 10 times the energy of a traditional Google search.
The DOE estimates that data centers in the U.S. currently account for 4.4% of total power demand, a figure expected to grow to 6% to 12% by 2028. McKinsey estimates that data center load may constitute 30% to 40% of all new electricity demand in the U.S. through the decade. Meeting this demand requires significant investments in grid infrastructure, which already faces permitting delays and long lead times.
Sustainability Commitments Under Pressure
The rapid increase in energy consumption directly challenges the technology sector's climate commitments. Google reported a 48% surge in total emissions from its data centers in 2023 compared to its 2019 baseline, attributing this increase to rising power demands. The company acknowledged that integrating AI into its products makes reducing emissions increasingly difficult.
This trend extends across the industry. S&P Global Ratings predicts that "U.S. data center power demand will increase at 12% per year until the end of 2030," potentially doubling the tech sector's carbon emissions. They also state that approximately 60% of this new demand could be met by natural gas generation due to constraints on renewable energy growth and the need for stable power sources. This reliance on fossil fuels conflicts with widespread decarbonization goals and underscores the urgency of finding alternative solutions.
Current Energy Strategies Are Unsustainable
Traditional approaches to addressing data center energy needs, such as diesel backup generators or overbuilding grid capacity, are proving inadequate. Diesel generators contribute significantly to carbon emissions when used and remain idle for most of their operational life. Expanding grid capacity involves long permitting timelines and high costs, making it challenging to keep pace with surging demand. These strategies fail to address both reliability and sustainability concerns, highlighting the need for paradigm shifts rather than incremental tweaks.
Assessing The Leading Energy Solutions For Data Centers
While solar, wind, geothermal, battery storage, natural gas and nuclear power all hold promise for reducing carbon footprints and enhancing resilience in digital infrastructure, each comes with its own set of limitations, whether in terms of intermittency, geographic constraints or scalability for the largest and most power-intensive facilities.
Hydrogen fuel cells and nuclear microreactors have both seized industry attention as compelling options for delivering reliable, low-carbon power at the scale and density required by modern data centers. While both technologies offer significant potential for emissions reduction and operational resilience, they differ greatly in terms of technological maturity, deployment readiness and integration pathways.
The Role Of Hydrogen Power Today
Hydrogen fuel cells are increasingly recognized as a practical and immediate solution for sustainably powering data centers. These systems emit only water as a byproduct and eliminate operational carbon emissions. This makes hydrogen an extremely appealing choice for AI data center operators working to reduce reliance on fossil fuels and maintain reliable power, even in areas where grid capacity is limited or renewable integration is challenging. Modular hydrogen solutions are already commercially available and can be scaled to deliver megawatts of power, supporting the increasingly high-density racks required by modern AI workloads.
Hydrogen-powered data centers can often be deployed more rapidly than traditional grid-connected facilities, providing a crucial advantage as AI-driven demand accelerates. Integrating hydrogen with other renewable energy sources further enhances its appeal, as does the operational efficiency gained from using byproduct water for cooling and achieving water-positive status in some installations.
The Promise Of Nuclear Microreactors
Nuclear power, and specifically the development of microreactors, is also emerging as a promising long-term solution for data center energy needs. Microreactors are compact, carbon-neutral sources capable of providing steady electricity and process heat, with the potential to operate for years without refueling. Their high reliability and ability to function independently or as part of a microgrid make them attractive for critical infrastructure and remote or grid-constrained sites.
Still, deploying nuclear microreactors for data centers is several years away, based on regulatory approvals, high capital costs and supply chain complexities. Widespread commercial implementation is unlikely before the 2030s. While several pilot projects and commercial agreements are underway, they are still in the early stages, leaving them unavailable to meet near-term AI data center demand.
Complementary Solutions For A Growing Challenge
Hydrogen and nuclear are ultimately complementary, rather than competing, solutions. While hydrogen is well-positioned to address immediate and near-term challenges, offering flexibility, rapid deployment and the ability to support peak and backup power needs, nuclear is expected to provide stable, large-scale baseload power as new projects come online in the next decade. Together, these advanced technologies can help the data center industry meet the dual goals of supporting AI-driven growth and achieving sustainability targets.
The sector's rapid evolution means that new innovations and hybrid approaches may rise to the forefront. As the digital landscape transforms, the industry must remain agile and open to emerging technologies that can meet the scale, reliability and environmental standards needed to support future innovation.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Business Insider
14 minutes ago
- Business Insider
Top strategist Vincent Deluard predicts a summer slump in stocks — but says 'recessions have been canceled'
The record-breaking stock market will tumble before September, but a recession effectively is "canceled," a leading strategist told Business Insider. "I expect US stocks to experience a sharp but brief correction in the summer," Vincent Deluard, the director of global macro strategy at StoneX, a financial services network, told BI in an interview. In a recent note, Deluard raised the prospect of a "brutal but brief" sell-off in late July or early August. He based the call on Donald Trump's tariff-negotiation period ending on August 1 and potentially spooking markets; the likelihood of further interest-rate cuts being delayed because of accelerating inflation and a tightening labor market; and the narrow breadth of the latest market rally, he said, adding that narrow rallies often lead to a correction within a month. He added in his note that foreign investors were anxious about "Trump's antics" with tariffs, the deficit, and their exposure to the US economy. He predicted selloffs would be short-lived as overseas buyers can't resist Big Tech stocks given their dominant market positions and central roles in the AI revolution. Deluard told BI he expects "several steep corrections" in stocks over the next two years due to "erratic policymaking, pressure from rising long-term bond yields, and selling from foreign investors." 'Recessions have been canceled' The macro specialist told BI there was a risk of a "brief stagflationary slowdown," but he ruled out a prolonged downturn as "recessions have been canceled by the shift to intangible assets, permanent stimulus, and demographics." Deluard spelled out his thinking in a note in May. He highlighted that the US economy has spent just 1% of the past 16 years in recession, down from around 40% of each decade between the 1860s and 1930s. He added that the main changes were a transition from an industrial economy to a less volatile services one, big rises in government spending on healthcare and social support programs for an ageing population that have acted as a "permanent economic stimulus," and policymakers adopting a "whatever it takes" approach to avoiding recessions. Deluard told BI that the Federal Reserve might have to raise its inflation target from 2% to as much as 4%, because officials will realize they can't aim lower in an era of sustained deficit spending. He said that that would be "quite positive" for stocks, as it would support higher corporate earnings and lower interest rates. Deluard, an adjunct professor of finance at Saint Mary's College of California, told BI the housing market would likely remain weak due to "poor affordability, growing supply, and high and sticky mortgage rates." He suggested house prices would "cool down but not crater" over the next two years, buoyed by higher construction costs, rising incomes, and low unemployment.
Business Insider
14 minutes ago
- Business Insider
A new type of dealmaking is unnerving startup employees. Here are the questions to ask to make sure you don't get left out.
As a new kind of dealmaking is sweeping Silicon Valley, forcing employees to be vigilant about how much trust they are willing to put in startup founders. Over the past two years, instead of acquiring AI startups outright, Big Tech companies have been licensing their technology or making deals for top talent, with startup employees sometimes getting divided into separate camps of haves and have-nots. Those with the most desirable AI skills reap a windfall while those who remain are shrouded in uncertainty. That recently happened to Windsurf employees after the AI coding company was on the verge of being acquired by OpenAI for $3 billion, but was instead split in half. Google paid billions to hire Windsurf's CEO and top talent, and the hundreds of employees who remained were bought by another startup, Cognition. Unfortunately for startup employees, many investors expect these kinds of novel transactions to continue as the velocity of developments in AI makes companies unlikely to want to wait months or years for regulatory approval. Candidates need to ask tough questions about the founder Given traditional M&A has mostly gone out the window, it is more important than ever for startup employees to do their homework, advises Steve Brotman, managing partner at Alpha Partners. "In light of what we just saw with Windsurf, it's crucial to understand the ownership dynamics," Brotman said. " You don't want to be working 100-hour weeks only to realize your options are underwater or your exit upside is capped. And remember: companies that are transparent and deliberate about governance tend to be better long-term bets, both for your career and your equity." "Ask hard questions about runway, revenue, burn, and investor syndicate quality," Brotman continued. "Who's on the board? Are they structured for long-term growth or a quick flip?" The most important thing candidates should assess is how much they trust the founder, according to Deedy Das, an investor at Menlo Ventures. "Nobody wants to talk about the fact that founders control almost everything that happens in a company, including how you get paid, when you get paid, how the equity vests, and when you can sell the equity," said Das. "It's everything, so having trust in your founder to do the right thing by the team is extremely important." Just as investors would typically research a founder before writing a check to one of their many portfolio companies, prospective employees should ask around about founders whom they could be tied to for years, said Hari Raghavan, cofounder and CEO of Autograph. "They should be doing diligence on whether this is a standup person," said Raghavan. "Do your best to suss out, 'Are these guys going to take care of me?'" Raghavan suggests that founders should sign a written pledge agreeing to treat employees well in terms of stock options and exit scenarios. "These are things that any good founder should be doing, and the vast majority of good ones do, but I think even just establishing that set of rules is a good idea," he said. Prospective employees should not be afraid to "interrogate" a founder on how they are thinking about an exit, according to Jake Saper, a general partner at Emergence Capital. "Ask founders how they would weigh staying independent, a classic acquisition, or a licensing deal that carves out key people," Saper said. "Their answer tells you a lot about the journey you're signing up for." Scrutinizing the fine print has also become more important, said Saper. "Make sure offer letters and stock agreements spell out vesting acceleration, treatment of options, and retention bonuses if only 'substantially all' of the team moves," Saper said. "Those clauses mattered at Inflection and Windsurf, and they will matter again." In 2024, Microsoft hired the founder of Inflection AI, Mustafa Suleyman, and some of the startup's staff to help lead its AI efforts. In June, Meta paid $14 billion for a 49 percent stake in the data labeling company Scale AI and hired its founder, Alexandr Wang, to run its Superintelligence group. Meta also hired some of the startup's researchers. Last week, Scale AI laid off 14% of its workforce, or 200 employees, and revealed it is unprofitable. Finally, Saper says to take a hard look at the underlying business model of a startup to make sure it can last. "Startups with unique data feeds, embedded distribution or clear recurring revenue have leverage to stay independent," Saper said. "If a company's main asset is a brilliant but portable research team, you should assume Big Tech will come knocking."
Yahoo
42 minutes ago
- Yahoo
Tradition meets AI in Nishijinori weaving style from Japan's ancient capital
KYOTO, Japan (AP) — Nishijinori, the intricate weaving technique for kimonos that dates back more than a thousand years in Japan's ancient capital of Kyoto, is getting a high-tech collaborator: artificial intelligence. The revered colorful weaving style associated with 'The Tale of Genji' of the 11th-century Heian era, has gone through its share of ups and downs. But its survival is more perilous than ever today, as demand for kimonos nose-dives among Japanese grappling with modernization. Hironori Fukuoka, the fourth-generation successor to his Nishijinori business, is determined to keep alive the art he's inherited, even if that means turning to AI. 'I want to leave to legacy what my father has left for me,' he said in his rickety shop in the Nishijin district of Kyoto, a city with statue-filled temples and sculpted gardens that never seems to change. 'I've been pondering how the art of Nishijinori can stay relevant to the needs of today,' said Fukuoka. Besides the AI project, Fukuoka is also working on using his weaving technique to make super-durable materials for fishing rods and aircraft. Where tradition and technology meet Giant looms clatter at his shop, called Fukuoka Weaving. The patterns on the gorgeous fabric, slowly turning out from the loom, are repetitive and geometric, which makes it conducive to translating into digital data. Deciding which hand-dyed color thread goes where to make the patterns is much like the on-or-off digital signals of a computer. Such similarity is what Fukuoka focuses on in exploring how AI might work for Nishijinori, with the help of Sony Computer Science Laboratories, an independent research arm of electronics and entertainment company Sony Corp. AI only makes suggestions for the designs and doesn't do any of the actual production work. But that doesn't bother Fukuoka or the researchers. 'Our research stems from the idea that human life gets truly enriched only if it has both what's newly innovated and what never changes,' said Jun Rekimoto, chief science officer at Sony CSL, which is also studying how AI can be used to document and relay the moves of a traditional Japanese tea ceremony. 'We don't believe AI can do everything. Nishijinori is a massive, complicated industry and so it starts with figuring out where AI can help out,' said Rekimoto, also a professor at the University of Tokyo. What has come of it is a startling but logical turn in thinking, fitting of the art adorning kimonos worn by Japan's imperial family. The AI was fed various Nishijinori patterns that already existed and instructed to come up with its own suggestions. One was a bold pattern of black and orange that seemed to evoke a tropical motif. Striking a balance To Fukuoka, some of AI's ideas are interesting but simply off. The difference between AI and the human effort is that the former can come up with multiple suggestions in a matter of seconds. Fukuoka immediately gravitates toward the one that uses a motif of a leaf to define the angular lines of a traditional pattern, something he says a human wouldn't have thought of. He finds that ingenious. The kimono the AI collaboration has produced is a luscious soft green, although it doesn't have a price tag and isn't in production yet. The weaving is carried out by the old-style machine under the guidance of the human artist in the traditional way. Nishijinori kimonos sell for as much as a million yen ($6,700). Many Japanese these days don't bother buying a kimono and may rent it for special occasions like weddings, if at all. Putting one on is an arduous, complicated affair, often requiring professional help, making kimonos even less accessible. A creative partnership Dr. Lana Sinapayen, associate researcher at Sony CSL, believes AI often gets assigned the creative, fun work, leaving tedious tasks to people, when it should be the other way around. 'That was my goal,' she said in an interview at Fukuoka Weaving, of her intent to use AI in assistant roles, not leadership positions. Digital technology can't automatically represent all the color gradations of Nishijinori. But AI can figure out how to best do that digitally, and it can also learn how the human artist fixes the patterns it has produced. Once that's all done, AI can tackle arduous tasks in a matter of seconds, doing a pretty good job, according to the researchers. Artificial intelligence is being used widely in factories, offices, schools and homes, because it can do tasks faster and in greater volume, and is usually quite accurate and unbiased, compared to human efforts. Its spread has been faster in the U.S. and other Western nations than in Japan, which tends to be cautious about change and prefers carefully made, consensus-based decisions. But the use of AI in arts and crafts is promising, such as text-to-image generative AI for the creation of visual images from text prompts, according to a study by Henriikka Vartiainen and Matti Tedre, who looked at the use of AI by craft educators in Finland. 'As computers have taken over many routine-like and boring tasks that were previously performed by people, the computer revolution has also been said to liberate time and offer new opportunities for human imagination and creativity,' they said. Yuri Kageyama, The Associated Press Error in retrieving data Sign in to access your portfolio Error in retrieving data Error in retrieving data Error in retrieving data Error in retrieving data
