And the demand is only accelerating. The rise of AI, the Internet of Things, and cloud services is driving the number of transistor operations – and with it, data traffic and storage needs – through the roof. \u201cIt\u2019s not just the switching,\u201d Hilgenkamp explained in a recent talk. \u201cMoving data, whether on a chip or across to a data center, consumes similar amounts of energy. Data centers themselves require cooling, which results in even higher electricity and water consumption. The system as a whole is becoming a major driver of energy consumption.\u201d<\/p>\n
AI\u2019s nuclear appetite<\/p>\n
The implications are not abstract. Across Europe and beyond, new data centers are straining local grids, forcing governments to rethink energy infrastructure. In some cases, dormant nuclear plants are being reconsidered to meet demand. But is this the solution? \u201cYou can restart one plant today,\u201d Hilgenkamp noted, \u201cbut in two years you\u2019ll need two more. The growth is exponential.\u201d<\/p>\n
Analysts estimate that information and communication technologies (ICT) already consume around 5 percent of global electricity, a share that could double within a decade if AI adoption continues unchecked. The semiconductor industry is approaching fundamental physical limits: switching transistors much more efficiently than a projected 10\u207b\u00b9\u2078 Joule threshold encounters barriers of thermal noise and astronomical equipment costs.<\/p>\n
That leaves a sobering reality: unless computing itself changes, AI risks becoming an energy crisis of its own.<\/p>\n
<\/p>\n
Series<\/p>\n
Watt Matters in AI<\/p>\n
Watt Matters in AI is a conference that aims to explore the potential of AI with significantly improved energy efficiency. In the run-up to the conference, IO+ publishes a series of articles that describe the current situation and potential solutions. Tickets to the conference can be found at wattmattersinai.eu.<\/p>\n
Looking to the brain for answers<\/p>\n
This is where neuromorphic computing comes into play. Instead of relying on the decades-old \u201cvon Neumann architecture\u201d, where memory and processing are kept separate, shuffling data back and forth at high energy cost, neuromorphic approaches take inspiration from the brain.<\/p>\n
\u201cOur brains are the most energy-efficient processors we know,\u201d Hilgenkamp said. \u201cThey run for a day on the equivalent of a peanut butter sandwich, about 20 watts, yet can outperform supercomputers in pattern recognition and learning.\u201d<\/p>\n
The key difference is that in the brain, memory and computation are intertwined. Neurons and synapses both store and process information, allowing massively parallel, low-power operations. By mimicking this structure, neuromorphic chips could cut the energy cost of AI tasks by orders of magnitude.<\/p>\n
From synapses to silicon<\/p>\n
Some of this is already visible in today\u2019s AI hardware. Neural networks, used in applications ranging from ChatGPT to medical imaging, are themselves a simplified representation of how biological neurons connect and learn. Chipmakers like Nvidia have thrived because their graphics processors excel at the core operation these networks rely on: vector-matrix multiplication.<\/p>\n
But Hilgenkamp stresses that the next leap requires new materials and designs. One promising route involves memristors, resistive elements that can \u201cremember\u201d their state and change conductivity in response to voltage pulses, much like synapses adjust their strength when we learn. Researchers at IBM Zurich and in the Netherlands are experimenting with phase-change materials and disordered semiconductors to build these adaptive components. Early prototypes have already demonstrated impressive gains in tasks such as speech recognition.<\/p>\n
Other concepts, such as spiking neural networks, mimic the brain\u2019s method of transmitting information through voltage spikes rather than steady digital signals. Intel, IBM, and several academic groups have demonstrated chips that can run such networks far more efficiently than conventional processors.<\/p>\n
Urgency meets opportunity<\/p>\n
The stakes are clear: without breakthroughs in architectures like neuromorphic computing, AI\u2019s energy demand may outpace what grids and societies can support. With them, we may gain not only efficiency, but also new forms of computing better suited to the messy, adaptive challenges of real-world intelligence.<\/p>\n
Hilgenkamp points to growing momentum in this field. Dutch researchers, startups, and global industry leaders are all exploring paths \u201cbeyond von Neumann.\u201d National and European programs are beginning to channel funding into neuromorphic R&D. On November 26, Eindhoven will host the conference “Watt Matters in AI?<\/a>“, dedicated to exactly these questions: how big is the problem, what technologies might help, and what policies are needed to keep AI\u2019s footprint in check. Hilgenkamp chairs the program committee for this conference.<\/p>\n \u201cAI is not just a technological revolution,\u201d Hilgenkamp concluded. \u201cIt is also an energy revolution. To make it sustainable, we need to look beyond the chips we know, and learn from the most efficient computer we already have: the human brain.\u201d<\/p>\n","protected":false},"excerpt":{"rendered":"Artificial intelligence is rewriting the rules of technology, but it is also rewriting the world\u2019s energy balance sheet.…\n","protected":false},"author":2,"featured_media":84788,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[21],"tags":[371,111,139,69,145],"class_list":{"0":"post-84787","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-computing","8":"tag-computing","9":"tag-new-zealand","10":"tag-newzealand","11":"tag-nz","12":"tag-technology"},"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/84787","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/comments?post=84787"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/84787\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media\/84788"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media?parent=84787"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/categories?post=84787"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/tags?post=84787"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}