As AI workloads move into the cloud, companies face a deficit of processing power. At the same time, companies need to increase access to AI technology in the cloud by making data platforms fast, affordable, and on demand.
“AI is very resource-intensive, particularly when training the AI,” says Rob Enderle, President and Principal Analyst of the Enderle Group.
As companies have tried to feed data into systems like enterprise resource planning (ERP) or customer relationship management (CRM) software, they have been unable to deliver enough data to GPUs to be able to run them efficiently, according to Jonathan Martin, president of WEKA, an AI-native data-platform company.
In 2023, large enterprises focused on ingestion and training in regard to AI, but now companies are buying hundreds of thousands of GPUs, according to Martin.
To address these needs, WEKA recently announced it would provide the high-performance data platform for U.K.-based NexGen Cloud’s upcoming AI Supercloud as well as its GPUaaS platform called Hyperstack, which is a GPU on-demand platform.
The AI Supercloud will allow enterprises, research organizations, and governments to use a more affordable solution to support AI workloads.
“This approach shifts loads to where resources are underutilized and where energy costs are lower—also typically where there is energy excess as energy costs go up when that resource is constrained,” Enderle says. “So, this more efficiently uses the resources already in place.”
AI models require graphic processing units (GPUs) to be trained and run as well as reach their full processing potential.
"We always made the joke that GPUs a lot of time are like sloths that are asleep about 70% of the time because they simply cannot be served enough data, which is where WEKA comes in to help in projects like this,” Martin says.
The news comes a year after WEKA launched its Sustainable AI initiative, which was designed to raise awareness of how AI, machine learning (ML), and high-performance computing (HPC) are driving global data center energy consumption and carbon emissions.
The concept of a supercloud originated with IBM, Enderle notes. Nvidia’s GPUs power the NexGen AI Supercloud, and the graphics chipmaker originated the AI Supercloud with its DGX Cloud, according to Enderle.
A hybrid multi-cloud brings lower costs due to the built-in competition as well as greater uptime as a result of redundancy, Enderle explains.
Chris Starkey, Co-founder and CEO of NexGen Cloud, says making the Supercloud clusters available on demand allows mid-tier companies to access higher amounts of GPUs for longer periods of time. In addition, combining a data platform like WEKA along with NexGen’s cloud platform leads to more sustainable GPUs.
“What the Supercloud represents is high quantities, shorter runs, and trying to streamline across the entire process,” Starkey says. “With WEKA, we're able to do that quite significantly.”
For the AI Supercloud, NexGen needed a data platform with low latency. The low-latency WEKA data platform allows GPUs to run at peak performance and efficiency while lowering energy consumption, according to Martin.
The rise in generative AI is fueling the need for more GPU power. AI workloads require advanced GPUs, which are expensive and costly.
“GPUs are incredibly power hungry,” Martin says. “Probably with a thousand GPUs, you are consuming about a megawatt of power right now.”
AI will use more energy than the human workforce by 2025 unless the industry develops more sustainable artificial intelligence (AI) practices, Gartner forecasted.
An AI supercloud lets GPUs run more efficiently and sustainably.
“This spreads the loads across multiple cloud services and on-premise tuned to make costs more manageable since power costs are a major component of the AI cloud," Enderle says. "This has an inherent ability to push loads to where energy costs are lower, and that is increasingly data centers that are powered by low-cost sustainable energy sources like hydroelectric."
Previously, larger organizations were known to use AI, according to Martin. Incorporating the WEKA data platform into the NexGen AI Supercloud will provide access to AI workloads to smaller organizations, he says.
“The AI Supercloud provides a straightforward way to democratize access to AI for smaller organizations," Martin says. "By doing so helps them fuel the next wave of AI innovation, putting the most powerful GPUs in the world in the hands of the masses."
Companies such as NexGen scale their resources to meet the increase in demand. Data infrastructure from WEKA allows for large-scale AI model training and inference workloads.
In 2023, NexGen announced plans to invest $1 billion toward its AI SuperCloud in Europe. It began deployment in October 2023.
“There's a ton of research going into it, and I just don't think we can manufacture quickly enough to meet that demand,” Starkey says.
As GPUs get more expensive, companies will turn to GPU cloud platforms so they can use AI, according to John Abbott, principal research analyst at S&P Global Market Intelligence.
"Full-stack as-a-service business models for generative AI will gain traction," Abbott said in a statement. "The ever-rising price of AI-enabled infrastructure is dampening enthusiasm for on-premises deployments, favoring the cloud.”
The AI Supercloud holds promise in industries such as healthcare and e-commerce, according to Starkey.
“I see a huge amount of work going into building more sophisticated customer service models for e-commerce sites,” Starkey says. “We're seeing that quite commonly lead to some kind of low-hanging fruit territory we're seeing at the moment.”
Starkey says superclouds are being used to help with AI model training in cancer research and powering chatbots for e-commerce sites.
The efforts by WEKA and NexGen to enable more sustainable AI workloads in the cloud are a sign that Europe is ahead of the U.S. in sustainability, Enderle explains.
“Europe is generally more focused on sustainability than the U.S.,” Enderle says. “That extra focus on sustainability tends to drive related programs more aggressively there.”
Starkey says the AI Supercloud is currently deployed in Sweden and Norway and will go live in France soon. This year, it will also be adopted in Germany and Canada. NexGen is in talks to bring the solution to U.S. data center operators by the end of the year.
“Our strategic road map is to build smaller locations in each region and then scale out from there,” Starkey says.
Related articles:
Modernization of the data domain to include new practices, tools, and technologies is the third phase of digital transformation.
Think of Hypershield as an AI-enabled security fabric where every network port becomes a high-performance enforcement point.
With the rise of AI and its massive impact on networking, Cisco has pivoted swiftly to enable customers to utilize its converged networks to run taxing AI apps.
