A company focused on upgrading data centres by offering graphics processing unit (GPU) cloud using direct chip cooling technology, can promise customer savings by 30%-40% compared to an existing data centre build. And, they’ve partnered with Nvidia and Deloitte to expose enterprise clients to AI in the most sustainable way.
The company, called Sustainable Metal Cloud (SMC), has deployed its facilities in two data halls at Singapore’s ST Telemedia Global Data Centre (STT GDC). The average data centre today relies on air cooling technologies to keep its servers cool, as computing is a process that releases heat.
Increasingly, immersion cooling has been growing in popularity, but Tim Rosenfield co-founder and co-CEO of SMC says that this solution has rarely been deployed on a large scale yet. Meanwhile, SMC’s hyperscale liquid cooling data centre build has already been deployed in a 20 megawatt (MW) data centre build in Australia.
Rosenfield calls this the hypercube, which he boasts has a power usage effectiveness (PUE) of 1.03 — this ratio refers to the total amount of power used by a computer data centre facility to the power delivered to computing equipment. The lower the PUE, the more efficient a data centre is.
A typical data centre today takes anywhere between US$7 million to US$10 million per MW to build. But SMC’s solution, which can be retrofitted into an old data centre, and built into a new data centre, takes anywhere between US$3.5 million to US$4 million per MW to build.
Rosenfield explains that the main cost savings comes from the fact that the hypercube has about a 50% reduction in energy over today’s air cooled data centre. Since power in Singapore is expensive, this can in turn result in substantial cost savings.
See also: Keppel’s private fund enters JV to develop 80MW data centre campus project in Taiwan
Since Nvidia announced its Blackwell chips this March, the industry has been grappling to upgrade the technologies to cater for what is dubbed as “the world’s most powerful AI chip”.
“And I’m not speaking for Nvidia, but my personal opinion is that they want their customers to be able to use [their new AI chips], and they want their technology to be used cost effectively and rapidly,” says Rosenfield. “I think our point is that we've pioneered a way to develop to deliver their platform at scale, more cost effectively and more sustainably than any other technology.”
SMC’s solutions come at a time when Singapore has just announced its ambition to expand its data centre capacity — it expects to add at least 300MW in the next few years, with another 200MW allocated only for operators who use green energy options.
See also: Keppel and Australia’s Woodside sign offtake term sheet for liquid hydrogen
This news announced by Deputy Prime Minister Heng Swee Keat at the ATxSummit in May is part of Singapore’s new green data centre roadmap. Most would remember that the city-state imposed a moratorium on data centre build back in 2019, on the back of energy and sustainability concerns.
Unsurprisingly, Rosenfield says that there has been global demand for SMC’s solutions. The SMC hypercube, which is live and functional in STT GDC, already services clients.
SMC’s clients include enterprise clients, large language model (LLM) builders, individuals who use GPUs for gaming, and research.
He reveals that a customer in Singapore has developed a high performance multi-modal LLM with Singapore native data.
