
Japanese vending machine manufacturer claims new system can reduce energy consumption by up to 85%.
Fuji Electric has adapted its vending machine cooling technology — based on ejectors and derived from Fukushima — to cool AI servers, promising to reduce energy consumption by up to 85%.
AI-generated summary
Ejector-based technology developed after the 2011 Fukushima nuclear accident to reduce energy waste in vending machines.
Tokyo | Financial Times Fuji Electric, Japan's largest convenience machine maker, has adapted its refrigeration technology to cool servers in artificial intelligence data centers. The company claims the system can reduce energy consumption by up to 85%.
The technology derives from a project developed after the Fukushima nuclear accident in 2011, when vending machines became the target of criticism for wasting energy. In 2015, engineers created an ejector with CO₂ refrigerant that eliminated the need for a compressor and reduced energy expenditure by 20%.
While conventional systems use high-powered electrically powered compressors, Fuji Electric's ejector harnesses waste heat to propel fluid through a precision nozzle.
"Simply extending conventional technology would not have been enough," said Hideto Shirai, who leads Fuji Electric's efforts to sell the technology to data centers.
Although unfeasible for vending machines due to cost, the solution found space in the IT infrastructure market.
The Japanese manufacturer aims to sell more than 10 billion yen ($61 million) worth of ejector systems for data centers by 2030.
The company's role in the AI supply chain has helped its shares rise nearly 47% over the past year.
Experts point to challenges for large-scale adoption. Analysts indicate that large data center operators have already closed architectures for the coming years, making it difficult for new suppliers to enter, in addition to competition with absorption coolers from rival brands such as Kawasaki and Ebara.
Alex Cordovil, research director at the Dell’Oro group, specializing in data center infrastructure, classified the technology as "promising", but highlighted that it is still in a "genuinely embryonic" phase.
The data center cooling market is moving from traditional systems with water, glycol or hydrofluoroolefins to direct liquid cooling, which circulates refrigerant directly into heat-generating components such as CPUs and GPUs.
Fuji Electric's system can be used in liquid cooling structures. Ejector technology performs better in regions with high temperatures, opening up potential space in areas of Japan where summer heat exceeds 35°C, as well as Southeast Asia.
However, Cordovil pointed out that a rival technology powered by waste heat — absorption chillers manufactured by Japanese companies Kawasaki, Ebara and Yazaki — has an advantage in adoption by data centers.
Goldman Sachs analyst Ryo Harada expressed skepticism about Fuji Electric's ability to access the North American data center market.
"The technology may be attractive, but commercialization is equally important," he said. "Tech giants have already defined their next-generation data center architectures for the next two to three years, so it should take at least that long for new entrants to gain relevance."
AI outlook — possibilities, not facts
Fuji Electric aims to sell more than 10 billion yen worth of systems by 2030
Possible · Within months

Folha launches this Sunday (4), the day of the first round of the 2026 elections, the Ballot Box, an artificial intelligence chatbot to answer voters' questions about voting, counting and necessary documentation, based on data from the TSE and service reports from the newspaper.

The SeePets app is launched this Sunday (4) to centralize information about lost and found pets. Developed by Wanderson, the free tool uses geolocation and works as a social network focused on animal welfare.

The Brazilian electronic voting machine, developed in 1995 after a conversation between minister Carlos Velloso and Paulo Camarão, completes 30 years of official use in 2026, having revolutionized counting and inspired systems in 33 countries.

David Robinson, a former security employee at OpenAI, criticized the company's culture of accelerated development, saying it increases the risk of failures in AI systems. He advocated for more space for security experts and research before the launch of advanced models, comparing the need for security measures to those adopted in sectors such as nuclear energy and aviation. The criticism adds to a debate in the sector about the speed of AI development, following episodes in which security controls failed at companies such as OpenAI and Anthropic.

The Federal Court of Appeals for the 8th Circuit granted a preliminary injunction to xAI, temporarily halting enforcement of the Minnesota law prohibiting the creation of non-consensual sexual deepfakes through AI tools like Grok Imagine. The company claims the law is unconstitutional and that its technology includes strict protections against abuse.
The Brazilian Space Agency (AEB) instituted a new innovation policy to strengthen the national space sector, increase economic competitiveness and expand Brazil's technological autonomy through seven strategic axes.