Kathy Gibson reports from Pinnacle TechScape – As artificial intelligence takes off, organisations are focusing on how quickly they can build and deploy the relevant infrastructure.
Globally, as much as $1-trillion is being invested in building and deploying AI-first data centres.
These AI factories need to be bigger as well as smarter, flexible and able to scale.
Wynand Moller, regional manager at Supermicro, says most consumers view AI as a tokenised transaction.
But the organisations running their own data centres infrastructures are having to rethink these developments.
Because the traditional data centres are not at all suitable for AI workloads, Moller explains.
Enterprises have read the writing on the wall: 92% of them are investing in AI, and 50% are already deploying agents.
“But there is a big AI deployment gap in that only 1% of organisations have mature AI infrastructures.”
Complexity is the top barrier, but the massive investment required is also putting the brakes on.
This creates an opportunity for AI factory organisations to bridge the gap with purpose-built infrastructure and to industrialise AI to scale value.
“You can continue to provide traditional servers and have linear growth,” says Moller. “Or you can look to building AI factories and have exponential growth.”
The bottom line, says Moller, is that companies need AI factories if they want to stay competitive.
He describes an AI factory as a turnkey data centre, a complete integrated solution built for AI scale that includes compute, networking, storage and management software – ready to deploy AI workloads.
The factory analogy is used because, whereas a physical facility turns raw material into products, an AI factory turns data, power and models into AI tokens.
Moller explains that SuperMicro AI factories are built to scale, using Nvidia, AMD and Intel reference architectures and designs to build the best infrastructure for each use case.
“Traditional data centres fall short because they are CPU-centric by design,” he adds. “This means there is a big performance gap – and organisations pay with an ROI (return on investment) penalty.”
Existing data centres were designed to run applications and transaction. They typically weren’t designed for liquid cooling, and they use a lot of power to run AI workloads.
AI factories, on the other hand, are purpose-built so they cost less to run.
Moller encourages partners to work with SuperMicro to help their customers deploy their own AI factories.
“Our recommendation is to start small: do a sandbox area, test applications, and gauge the value. Then scale in phases as demand grows by adding GPU racks, storage and fabric capacity.”
They can ensure investment protection by using the same core technologies for early racks and later expansions, avoiding the need for forklift upgrades and new designs.
“SuperMicro is one of the world’s market leaders for AI factories,” Moller adds. “We supply compute, network, storage and software – and we manufacture everything ourselves.”
On top of this, SuperMicro offers an array of cooling solutions, including CDUs, air to liquid and liquid to air.