Australia | 10:00 AM
Despite growing investor scepticism, analysts believe the next phase of the AI investment cycle will be driven by productivity gains, data centre expansion and a widening group of ASX-listed beneficiaries.
- AI infrastructure investment still has further to run
- Australia emerges as a data centre investment hub
- Identifying the ASX's AI beneficiaries
- AI to feature heavily throughout August results
By Danielle Ecuyer
AI fatigue and concerns create opportunities
US commentator Ed Yardeni may have had his finger on the trigger when he wrote,
“Investors are suffering from AI fatigue. They've concluded that there is no way to estimate whether all the capital spending on AI infrastructure will generate good ROIs in the coming years. What they do know is that hundreds of billions of dollars are being spent on AI capex in the here and now.”
Morgan Stanley anticipates AI capex spending will continue to rise and forecasts expenditure to reach US$1.2trn-US$1.4trn across the major AI infrastructure companies, with Google, Microsoft, Amazon and Meta remaining the dominant hyperscalers, Oracle increasingly considered a potential fifth hyperscaler and Nvidia continuing as the leading supplier of AI compute.
Computing capacity is expected to grow to nearly 120GW by 2028 from around 30GW.
Morgan Stanley compares AI infrastructure with an "Intelligence Superhighway", arguing the more useful AI becomes, the more valuable the infrastructure supporting it becomes.
Pushing back on some of the negative narratives around AI usage regarding the cost of tokens, the analysts believe AI usage will increase as it starts to generate meaningful productivity savings at a relatively low cost.
For example, it is suggested a business could save US$55 while costing only US$2-US$5 in token usage.
The positive stance also centres around the Jevons paradox, which proposes that as AI models and chips become more efficient, the cost of using AI falls.
AI is more likely to boost usage, with more frequent use across more complicated tasks and broader adoption by businesses.
While Chinese open-weight AI models are evolving to become more cost competitive and capable, Morgan Stanley sees them as a potential threat to US large language models rather than to demand for compute.
The greatest concern is AI compute demand will continue to exceed supply due to a combination of potential hurdles from a shortage of skilled labour, access to power and political pushback on the growth in data centres.
These constraints may result in a shortfall of US data centre capacity that could exceed -10% by 2028 relative to the amount required.
Ned Davis Research also leaned into AI capex and thematic investing in a recent presentation, postulating AI inference workloads could exceed model training in 2027.
A trend towards more inference would expand the list of AI beneficiaries beyond GPUs into software, storage, networking, security and enterprise applications.
Productivity gains from major new technology investments have traditionally lagged by around five years, Ned Davis suggests, meaning the productivity gains and inference phase may develop more towards the end of the current decade rather than in the near term.
Similar to Morgan Stanley's views, the suggestion made to investors is that bottlenecks and the AI build-out might prove to be greater beneficiaries and investment opportunities than the capacity owners.
Ned Davis likes data centre electrification for an Overweight sector allocation in the investment portfolio, including the likes of GE Vernova, Siemens Energy, Vertiv, Schneider Electric and Caterpillar.
Infrastructure spending alone cannot maintain the cycle, the analysts highlight. For the AI boom to continue, it must generate productivity gains for businesses to justify its use.
Morgan Stanley remains convinced such will be the ultimate outcome.
Australia is a prime candidate for data centre investment
Turning closer to Australia, Oxford Economics has observed a "step-change" in data centre investment underpinned by a transition to AI-related compute.
The estimated data centre investment pipeline is over $300bn, which includes $37bn of projects under construction and $48bn of projects expected to break ground in the near term.
A further $195bn of projects has been proposed, with the pipeline expected to rise in coming months.
Data centre capacity, including the proposed pipeline, is forecast to reach around 25GW from current capacity nearing 2.2GW.
Construction spending is anticipated to peak near $30bn by FY30, which is three times the current level. Including IT equipment, total investment is projected to rise above $60bn by the end of the current decade.
Oxford Economics labels Australia a very attractive investment location.
Its time zone fills a gap in global computing demand between US and European operating hours while overlapping with Asian markets.
Australia offers a "low-latency gateway for cross-border operators".
Grid reliability is another factor, with the country's renewable energy resources attractive to hyperscalers seeking sustainability targets.
Land availability, the Commonwealth's National AI Plan, attractive sovereign risk, Five Eyes membership and Australia's status as a US security ally offer hyperscalers greater confidence in the country's geopolitical stability.
On the other side of the coin, as equally outlined by Morgan Stanley and Ned Davis, investment in data centres faces a set of potential constraints and challenges, including grid connections, rising demand for power and energy, increasing scrutiny over water for cooling, as well as labour constraints.
Community pushback is also rising, while the regulatory environment continues to evolve depending on state jurisdictions and changes to federal policy.
The full story is for FNArena subscribers only. To read the full story plus enjoy a free two-week trial to our service SIGN UP HERE
If you already had your free trial, why not join as a paying subscriber? CLICK HERE