Data centers have become the new cell towers: no one wants to live near one, not even if they offer money to build one. There are compelling reasons, such as pollution or what electricity bill prices skyrocket. In this context, a San Francisco startup has had an idea that can be crazy or genius: convert people’s homes into data centers.
The idea. SPAN is a San Francisco startup that sells home electrical solutions, such as smart electrical panels or vehicle chargers. His new idea is called XFRAa “distributed data center solution” that installs an NVIDIA GPU node directly into homes. That is, instead of having a giant building full of GPUs, they distribute them in small panels throughout entire neighborhoods.
These nodes, which have a design reminiscent of a larger air conditioner, contain 16 NVIDIA RTX PRO 600 Blackwell GPUs, 3TB of memory, and liquid cooling. In statements to Ars Technicathe company’s vice president said that “Data centers are noisy, unsightly and often drive up local electricity bills. This is quiet, discreet and makes energy more affordable for the provider and the community.”
Image: SPAN
Everyone wins. According to SPAN, their solution is a win-win for both hyperscalers and owners. AI companies are able to increase their capacity more immediately, avoiding the long lead times of large data centers. Meanwhile, owners who install one of these nodes will have discounts and even not pay anything for their electricity and internet bill. Electricity increases are one of the arguments of those who reject data centers, so this may increase their attractiveness for owners.
Qlanes of the future. SPAN already has a pilot program underway to test its invention in 100 homes this year. The intention is to begin installing XFRA nodes in newly built homes, although they also consider the option of installing them in existing homes, in addition to offering more powerful nodes for commercial clients.
The company’s plans are to install 80,000 XFRA nodes throughout the United States, which would achieve 1 gigawatt of distributed power at a much lower construction cost. Of course, it would not serve to replace traditional data centers used to train AI models, but this computing capacity would be oriented towards other uses such as inference, cloud gaming or content streaming.
Context. The AI boom requires a lot of computing power and companies started to build mega data centers like there is no tomorrow. They soon realized the problem: there is no power for so many chips. What has followed is an electrical network that does not support so many data centersthe skyrocketing electricity billthe technologies using nuclear energy and one strong opposition from the community. All of this is making building and launching a data center a process that can take years, That is if they don’t end up being cancelled.
Maybe it’s not so far-fetched. The energy problem has led technology companies to consider take data centers to spacewhere energy is unlimited, or submerge them in the sea to cool them. The options are on the table, although there are already those who warn that They are a chimera and we still do not have the necessary technology. In this sense, the idea of SPAN is presented as a much more realistic option, although seeing the rejection that data centers are arousing among citizens, it is not clear that it will be well received.
Image | Xataka, with ChatGPT
In Xataka | Quietly, Aragón is becoming a data center “powerhouse”: now it has taken a crucial step
GIPHY App Key not set. Please check settings