Growing bottlenecks pose challenge to AI frontrunners

The race to develop and monetize artificial intelligence is starting to be marred by bottlenecks, says Kathrin Forrest, equity investment director with Capital Group.

Speaking on the latest episode of the Soundbites podcast, Forrest said competitors in the crowded field will have to solve some big supply chain and infrastructure problems before they can harness the revenue potential of AI.

“Some of the key challenges are managing bottlenecks,” she said. “Certainly leading-edge chips have been a bottleneck. Land, data centre architecture, all of those have turned into bottlenecks.”

Forrest said the problem is especially acute in the build-out of data centres — a critical component in the AI field.

“[There’s] electricity as a bottleneck, certainly something that many of us have caught up on. And just the strength in expected power demand growth in the U.S., in particular, coming from data centres,” she said, noting that Microsoft recently disclosed that it would be slowing some data-centre projects, “triggering worries about the ongoing momentum in the AI infrastructure build-out.”

Among other constraints, building new sites involves a lot of regulatory and bureaucratic hurdles, the facilities create enormous power demands, and require sophisticated cooling mechanisms.

“We need cooling mechanisms, different types of cooling mechanisms, just because the rack density in those data centres is so much higher. So liquid cooling is one of the ways to go about that. But that then comes with other bottlenecks, including water, for example,” she said.

And at every level of the AI stack, one of the biggest bottlenecks is the need for highly specialized human capital, she said — everything from welders, electricians, plumbers, all the way to AI researchers.”

Other significant challenges — less tangible but no less important — include the growing concern about the economics of the endeavour.

“Clearly, there’s a lot of money being spent,” she said. “Can we validate the value of that investment?”

The leading players must also consider their position in the competitive landscape, and determine how to invest so they neither overspend nor get left behind.

“Innovation is blurring competitive lines, and it’s about weighing the risk of spending too much versus the risk of spending too little,” she said.

Monetization

The monetization of the AI theme is proving elusive at the moment, with broad adoption of AIA tools limited by hurdles that include the existence of vast quantities of data that are difficult to access, process or analyze (dark data), challenges in how data and components interact (architecture), and regulation.

A potential path forward would be the broadening of existing IT applications with the integration of AI tools. Examples include Microsoft integrating Copilot into some of its existing productivity applications, Meta using AI tools to enhance user engagement and ad targeting, other IT companies building autonomous AI agents that integrate into their existing software.

“The direction from here I would view as a platform technology,” she said, “broadening applications beyond just the core IT, into all other sectors of the economy, with new market opportunities for companies that are thoughtful about the competitive edge, the data [and] the potential for partnerships.”

Collaborations

Forrest said collaborations between AI companies will not only uncover new market opportunities, but could address some of the bottlenecks.

Examples include:

• A collaboration between one of the leading large-language-model developers and a large e-commerce platform to improve checkout flows by allowing users to complete purchases directly in a chat interface.
• Collaborations in the wearable space, where some of the larger players work together to provide integrated healthcare, consumer and technology features. Examples includes the launch of smart glasses that can make calls, send texts and live stream.
• Collaborations that address electricity bottlenecks across geographic boundaries.

“As AI capabilities become more powerful, there are clearly opportunities to see benefits across many sectors,” she said. “Over time, we might see those benefits move from tangential to more core.”

For example, within pharmaceuticals, there are opportunities to accelerate drug discovery through predictive modeling, identify promising drug candidates faster, and streamlining clinical trials. Within energy, there is the potential for powerful predictive maintenance tools that would improve operational efficiency and decision making. Within transportation, AI would enhance flight operations and delay management. Within grocery, there is the potential to reduce food waste. Within retail, AI could increase customer personalization, reduce wait times and improve inventory management. And within financials, facilitate fraud detection and risk management, as well as process automation.

“This is a foundational technology, a platform technology,” she said. “And if you take a long-term view and analyze companies and their opportunities over the long term, you can find some really interesting investments that are not constrained by sector or geography.”

**

This article is part of the Soundbites program, sponsored by Canada Life. The article was written without sponsor input.