Nvidia partner says it can cut data center energy use by 50% as AI boom strains power grid

The artificial intelligence boom is ramping up demand for more powerful processors as well as the energy needed to keep data centers cool.

That’s an opportunity for data center company Sustainable Metal Cloud, which operates “sustainable AI factories” made up of its HyperCubes in Singapore and Australia.

The HyperCubes contain servers fitted with Nvidia processors which are submerged in a synthetic oil called polyalphaolefin that draws heat away more efficiently than air. The company said its platform reduces energy consumption by up to 50%, as compared to traditional air cooling technology typically used in data centers.

“It enables high density hosting for GPUs. It enables the sort of hosting that we need to see for platforms like [Nvidia’s] Grace Blackwell,” said Tim Rosenfield, co-founder and co-CEO of Sustainable Metal Cloud, referring to the new generation of AI graphics processors Nvidia announced in March.

The Singapore-based firm also said its immersion cooling technology is 28% cheaper to install than other liquid-based solutions. The HyperCubes are designed to go into any data center and can be deployed in unused spaces within existing data centers.

“Our solution being containerized means we can go anywhere very quickly. And we can open up new availability zones in response to demand from customers …,” said Rosenfield.

He said SMC is expanding into other markets like Thailand and India.

The firm already counts Nvidia and Deloitte among its major enterprise partners. SMC is a preferred cloud partner of Nvidia for compute and AI and offers GPU clusters designed by the chip giant. In July, SMC announced a partnership with Deloitte in which it will provide access to Nvidia’s GPU computing infrastructure for the consultancy’s clients to build AI applications.

Governments and businesses have rushed to capture the transformative impact of AI and data center demand has boomed as a result. 

Countries like Singapore, where SMC is headquartered, are also looking to mitigate the hefty energy consumption by pushing for “green” data centers to support its AI ambitions where the country has committed more than 500 million Singapore dollars ($379.7 million).

Sustainable Metal Cloud has received funding from Singapore state investor Temasek-backed ST Telemedia Global Data Centres, one of Asia’s largest data center operators.

SMC is currently raising $400 million in equity and $550 million in debt, with the funds going toward its data center expansion beyond Singapore, Bloomberg reported, citing people familiar with the matter.

Technology companies are increasingly looking into liquid cooling for its data centers as increasing performance needs also ramp up overheating risks.

Traditionally, data centers have large aisles to allow cooled air to circulate as overheating can lead to equipment failure and downtime. In theory, liquid cooling should enable these facilities to pack more servers into their existing footprint.

Giordano Albertazzi, CEO of digital infrastructure provider Vertiv, told CNBC in June that liquid-cooling adoption could accelerate in 2024. Vertiv’s thermal management offerings include hybrid air- and liquid-cooling, as well as fully liquid-cooled data centers.

During Computex Taipei in June, Nvidia CEO Jensen Huang championed the benefits of Supermicro’s direct liquid-cooling technology, which he said saves energy and lowers costs in AI data centers. 

Supermicro CEO Charles Liang told CNBC in June that liquid cooling has greater power efficiency leading to better performance, less pollution and lower energy costs.

Despite the fanfare, challenges remain in deploying liquid cooling, according to SMC’s Rosenfield. 

“Most data centers are not ready for liquid of any type, whether it is immersion or direct chip cooling. The market is figuring out the best way to employ this and I think there’ll be multiple ways,” said Rosenfield. 

Vertiv’s Albertazzi said, “There is still a lot of air cooling that still happens in the data center and will continue to happen even in the full high-density AI data center.”