- A trend toward private AI - that is, on-prem AI deployments - is emerging among enterprises
- Teradata and others like Dell and HPE are pushing to make AI factory capabilities available on-prem
- While AI is giving on-prem deployments a boost, Synergy noted hyperscale capacity is still growing faster than the market overall
First there was private wireless. Then there was the private cloud. So, it comes as no surprise that the newest critical tech – AI – is now getting the private treatment as well.
Teradata this week became the latest company to hop aboard the private AI bandwagon with the launch of Teradata AI Factory.
The term “AI factory” refers to specialized compute infrastructure (both hardware and software) designed specifically for AI development, training and workload management. While hyperscalers, telcos and companies like Nvidia are working to build large-scale, high-profile AI factories like Stargate, Teradata and others are essentially looking to bring the same capabilities to a wider audience which prefers to run things on-prem.
Why? Because while public cloud environments are great for experimentation, there’s an increasing recognition that fine tuning of AI models is required for specific verticals. Combine that with concerns about data security and privacy and you have a market eager for on-prem AI factory capabilities.
That’s especially true for heavily regulated verticals like financial services, healthcare, telecoms and governments, Teradata SVP of Product Management Dan Spurling told Fierce.
“We are seeing an interesting phenomenon where customers are looking to drive experimentation and innovation in the cloud, but then they’re actually looking to financially get the benefit of on-prem … for running the large AI-scale inferencing and storing,” Spurling said.
Teradata AI Factory comprises the company’s AI platform for structured and unstructured data access and workload configuration; tools for data pipeline management and RAG setups; and a software workspace with access to key libraries, analytics and other functions. All of this runs on the customer’s existing on-premises hardware, Spurling said.
Of course, Teradata isn’t the only one who has picked up on this. Dell, HPE, others have similar AI factory offerings.
Public vs. private
Teradata’s release came as Synergy Research Group released a new report showing that while on-premise deployments still account for around a third of overall data center capacity, the scales continue to tilt in hyperscalers’ favor.
Already, on-premise deployments have dropped from 56% of overall data center capacity in 2018 to 34% in 2024. By 2030, that number is expected to fall further to roughly 22%.
Meanwhile, hyperscalers’ share of overall data center capacity is expected to jump from 44% today to 61%.
Synergy’s Chief Analyst John Dinsdale noted the shifting percentages are due to the fact that hyperscalers are rapidly expanding the footprints and are expected to grow their capacity “threefold over the next six years” – faster than the overall market’s expansion. So, while the amount of on-premise data center capacity will grow in absolute numbers, it will continue to fall as a percentage of the overall market.
“After a sustained period of essentially no growth, on-premise data center capacity is receiving something of a boost thanks to GenAI applications and GPU infrastructure,” Dinsdale wrote. “Nonetheless, on-premise share of the total will drop by around two percentage points per year over the forecast period.”