Mizuho director predicts surge in enterprise-specific AI implementations will reshape data center market.
The next major driver of data center demand won’t come from generic AI models like ChatGPT, but from companies building their own private artificial intelligence systems, according to Vikram Malhotra, Managing Director and Co-Head of Mizuho Americas.
“I would imagine a Walmart, a big bank, a big manufacturer would want their own private model for their own business purposes, and not just have some generic models,” Malhotra says, pointing to an emerging trend that could reshape data center demand patterns.
Beyond Generic AI
While much attention has focused on large language models (LLMs) like ChatGPT, Malhotra argues the market is overlooking the potential impact of small language models (SLMs) designed for specific business applications.
“That speaks to a lot of private, what I’d call small language models. So there’s LLMs and there’s SLMs, and that can drive a lot more demand, both for big box warehouse data centers, but also smaller data centers,” Malhotra explains.
The iPhone Evolution Parallel
Malhotra draws a parallel between current AI capabilities and the early iPhone era: “I think we’re at the very early stages of what I’d say the iPhone analogy, which is we’re talking about iPhone one, two and three. But there’s a lot more to come for truly a large language model that may be truly machine learning and truly giving future predictions.”
Market Indicators to Watch
For investors trying to gauge this trend’s development, Malhotra identifies several key metrics to monitor:
“We’re definitely focused on the capex outlooks of the hyperscalers,” he says. “Number two, we’re focused on the cloud businesses, or the cloud demand that a lot of enterprises may take advantage of and more so marrying that with inference demand.”
The Path Forward
While computational efficiency improvements could affect near-term demand, Malhotra sees the trend toward private AI implementations as a sustained driver of growth. “We’re going to monitor how cloud evolution takes place, because in the next two to three years, they see the requirement for inference and larger boxes,” he notes.
This evolution could particularly benefit data center operators who can offer flexible solutions for both large-scale AI implementations and smaller, enterprise-specific deployments.
