- AI factories are in high demand as AI expands, with tech and telecom giants partnering to build these specialized data centers
- Supply shortages and record-low data center vacancy rates are delaying the rollout of AI factories, despite growing demand
- Analysts say AI could drive over a trillion dollars in infrastructure spending in the next five years
AI factories are becoming the new prime real estate, but turning blueprints into reality is easier said than done.
To date, these facilities — specialized data centers equipped with dense computing resources and advanced cooling systems — have mostly been used to train large language models (LLMs). We’ll need way more of them as artificial intelligence (AI) continues to integrate into every facet of modern life and the demand for compute power rises.
A number of key players in the tech and telecom sectors (data center providers, chip manufacturers, equipment vendors, etc.) are all looking for a stake in the game. For example, AI giant Nvidia has announced partnerships with telecom giants including Telenor, Fastweb, Indosat, Singtel and Swisscom, all of whom are helping to build or investing in AI factories.
These telecom companies are keen on AI factories to gain a competitive edge.
SK Telecom (SKT) told Fierce it is leveraging AI factories to transform its carrier network into an Edge AI-powered network. It also will use the facilities to explore new revenue models through AI data centers and establish service infrastructure for AI-driven customer services like Personal AI Agent and Telco LLM services.
Even national governments are stepping up and saying that they want to have their own AI factories that are “completely sovereign, where the data is completely resident in region,” said Kevin Cochrane, CMO of GPU cloud provider Vultr.
Tip of the iceberg
Amid record GPU demand, Cochrane said the market hasn’t even seen the impact of AI factories yet.
“There has been this massive acceleration in GPU revenue and AI revenue in general, but all of that is focused on a very narrow subset of companies,” he added. “This is only scratching the surface of the market that's to come.”
Dell’Oro expects AI deployments will push the data center physical infrastructure market over $50 billion by 2028. The firm also predicted that liquid cooling, another key component of these facilities, will see $15 billion in revenues over the next five years.
While 2023 was an exceptionally strong year for DCPI growth, Dell’Oro analyst Lucas Beran said that bump isn’t attributable to AI factories yet.
“It remains very early regarding the impact of AI,” he said.
High in demand, low in supplies
In part, AI factories haven't hit the markets because the industry is struggling to keep up with record-low data center vacancy rates and high demand for components.
AI factories may house tens of thousands of GPUs and accelerators, interconnected by specialized networks and supported by dedicated DCPI. To sustain the rapid growth of AI and prevent what some experts fear could be an "AI bubble burst," the supply for these factories must increase substantially.
Beran said end-users would like to see new data center capacity “faster than the industry can deliver it.” Adding to that, the shift to denser computing workloads is driving a need for higher amperage power distribution and liquid cooling systems, both of which are currently constrained, he told Fierce.
“Part of the challenge here is the pace at which utilities are able to supply power to these new facilities, which is leading to delays or data center owners and operators utilizing on-premises energy storage and/or power generation to try and overcome this," Beran added.
To build these factories, design changes also need to be implemented and the power and cooling infrastructure to be ordered, manufactured and sold. Generally, that takes 18–24 months, so the second half of 2024 and 2025 is when he expects AI factories will have a more meaningful impact on market growth.