Artificial intelligence (AI) is reshaping our world—from autonomous cars to smarter healthcare systems—but powering this revolution is no small feat. Behind every AI breakthrough lies a semiconductor chip, a marvel of technology that’s becoming increasingly vital and, at the same time, harder to produce. At the World Economic Forum in Davos, experts gathered to discuss the future of semiconductors and the challenge of balancing AI’s growing demands with energy and resource constraints.
“It’s going to be about power,” said Rodrigo Liang, CEO of SambaNova Systems, during the panel. “It’s going to be about efficiency—actually deploying these chips across a broad range of applications, from your data centre all the way to the edge.”
But the industry is facing a bottleneck. The chips we rely on for everything from Netflix recommendations to medical imaging are energy-intensive to manufacture and use, and scaling up production is a challenge in itself.
The hidden link: Industrial gases
While the spotlight is often on chip design giants like Nvidia and AMD, few realise the essential role played by industrial gases. These gases, including nitrogen trifluoride (NF₃) and silane (SiH₄), are critical to the intricate processes that make semiconductor chips possible, from etching and deposition to cleaning.
However, the production of these gases also has its own environmental impact. The US Department of Energy’s $1.4 billion funding boost for semiconductor innovation, highlighted here, isn’t just about building factories—it’s about driving efficiency across the entire supply chain. Hydrogen, for instance, is gaining traction as a clean energy source in chip production, while also being used to cool the massive data centers where AI models run.
AI’s appetite for chips—and energy
Over the past few years, the world has seen an explosion in AI adoption, with semiconductor manufacturers struggling to keep pace. “We’ve lived through this insatiable demand,” Liang explained. “And as the world turns to inferencing—using AI not just to train models but to drive workflows—we’re going to need a lot more chips.”
But there’s a catch: energy. AI requires staggering amounts of computing power, and data centres—the beating heart of AI—consume enormous amounts of electricity. Without careful planning, the energy demands of AI could outstrip available resources.
In 2022, data centres in the Republic of Ireland, home to tech giants like Google and Meta, consumed nearly 20% of the country’s electricity.
“We’re hitting a power wall,” Liang said bluntly. “The technology is there to commoditise AI, but not everyone can scale it. People are out there building gigabyte data centres and nuclear power plants to support this.”
So, what’s the solution? Part of it lies in making chips smarter, not just faster. Liang’s company, for example, is developing chips that deliver “10x performance at 1/10th the power,” enabling AI to expand even in regions with limited energy infrastructure. Innovations like edge computing—where data is processed closer to where it’s generated, rather than in massive central servers—could also help.
The industrial gases sector has a big role to play here. Cooling data centres with liquid nitrogen or helium, for example, can significantly reduce energy use. These gases are already in high demand, but their importance will only grow as AI continues to evolve.
And then there’s sustainability. As Christina Kosmowski, CEO of LogicMonitor, pointed out, visibility into energy use is key. “You can’t just build endless data centres,” she said. “We need to balance performance with cost and sustainability. It’s about connecting the dots—how do these technologies directly support business outcomes while staying efficient?”
Geopolitics and the race for chips
There’s another factor making life harder for chipmakers: geopolitics. The semiconductor supply chain is concentrated in a few regions, making it vulnerable to disruptions. Amandeep Singh Gill, the United Nations’ tech envoy, called for greater global collaboration, saying, “Concerns about supply chains and export controls are upping the stakes. Without a diverse innovation ecosystem, we can’t unlock the full potential of this technology.”
The US, China, and the EU are already pouring billions into semiconductor R&D and manufacturing. The CHIPS Act in the US and similar initiatives aim to diversify production, but the stakes are high. Producing advanced nodes—the tiny but mighty chips at the heart of AI—is extraordinarily expensive. As Liang noted, “It’s not just about designing the chips. Manufacturing them requires investments that only a few players can afford.”
Despite the challenges, there’s optimism. Collaboration across industries—chip designers, industrial gas producers, and policymakers—will be crucial to building a sustainable AI economy. Industrial gases, though often overlooked, are set to play a starring role in ensuring the semiconductor industry can keep up with AI’s appetite for power and innovation.
As AI becomes more ubiquitous, the pressure is on to innovate smarter, greener solutions. Whether it’s through hydrogen-powered cooling systems or chips designed to sip, not gulp, electricity, the semiconductor industry must evolve. After all, powering the future shouldn’t come at the expense of it.
