Artificial general intelligence (AGI) startup SingularityNet is investing in data center supercomputing modules.
The company, which is developing decentralized and blockchain-based AI infrastructure, this week announced a $53 million investment to further the development of AGI.
The first phase, which will utilize half of the investment, will be allocated to the creation of what the company claims will be the world’s first modular supercomputer dedicated to decentralized AGI research, while also building High-Performance Computing (HPC) and AI data centers. Precise details of the deployment were not shared.
SingularityNet said it will be using Ecoblox’s ExaContainer modular data center solutions incorporating GPUs and CPUs from Nvidia, AMD, and Tenstorrent, as well as servers from Asus and Gigabyte.
The supercomputer will be used to optimize the training of deep neural networks and large language models.
Resources will also be used for the construction of the first in a series of modular compute containers able to be positioned around the world and relocated as needed.
“The dramatic progress the AI field has seen recently is the result of convergence of multiple aspects including sophisticated learning algorithms and cognitive architectures, and massive amounts of data, processing infrastructure, and energy,” said SingularityNet CEO Dr. Ben Goertzel.
“The novel neural-symbolic AI approaches developed by the SingularityNet AI team decrease the need for data, processing, and energy somewhat relative to standard deep neural nets, however, even given these advances, the need for significant supercomputing facilities remains.”
Dubai-based Ecoblox has developed a 20 ft container it says can support 60kW per rack in an N+1 configuration and offer 380 petaflops of FP16/Bfloat16 performance (11.52 petaflops of double precision FP64 performance) via up to 96 Nvidia H100NVL cards or 192 H100 cards for customers whose models don't require NVLink.
SingularityNet said some of its recent hardware purchases include a range of GPU compute systems, including a modular data center equipped with Nvidia L40S GPUs, AMD Instinct and Genoa GPUs, Tenstorrent Wormhole server racks, and servers featuring Nvidia H200 and GB200 (Blackwell) systems.
“The work that Dr. Goertzel and his team are doing to bring AGI into both their supercomputers and into end products is great,” said Jim Keller, CEO of Tenstorrent. “Tenstorrent’s heterogeneous compute featuring our CPU, our RISC-V, and our AI accelerator technology are the perfect fit to help them accomplish this goal. Combine that with our open-source software stacks, and I am confident that SingularityNet will have what they need to accomplish their mission.”
Switzerland-based SingularityNet aims to create a “decentralized, democratic, inclusive and beneficial” artificial general intelligence that is not dependent on any central entity. It is aiming to use blockchain-based technologies.