FEATURE: Hyped up

Financial services firms stand at a pivotal juncture. Those in one camp, with their heads buried in the sand, can no longer ignore the fact that artificial intelligence (AI), particularly its sidekick generative AI (GenAI), is everywhere.

The other camp is either past the 'dabbling stage' or seriously contemplating how AI, with the right safeguards and budget in place, can be a natural fit in their business.

Take Perth-based financial planner at Guided Investor Brad Buters, who sees GenAI as a beneficial productivity tool that boosts efficiency and enables accuracy.

Where it is adding big value for many advisers is in file-noting thanks to a proliferation of AI-powered apps.

"Anytime we talk with a client, we have to make a detailed file note of that conversation and in that file note we need to capture not only the specific information, but what's changed in terms of dollar figures in their financial situation, and capture what the general conversation was like, what's in their best interest and so forth," he says.

He sees adopting AI in the same way as a new employee being hired and trained up.

"You have to spend a lot of time and effort getting the employee up to speed, and in the short term, you might not get immediate benefit from it," he says.

"But the idea is you invest in it, and you'll eventually become more streamlined because you can outsource part of what you do to that person. But in this case, you can outsource it to AI, which is awesome."

An April study by KPMG and the University of Melbourne found that nearly 65% of employees report that their workplace currently uses AI. Half of them lauded it for improving efficiency and quality of work and boosting innovation.

Across countries, including Australia, almost three in five employees intentionally use AI at work on a regular basis, with almost a third using it weekly or more.

General-purpose GenAI tools are by far the most widely used, the survey found, with most employees using free versions of tools like
ChatGPT rather than ones provided by their employer. Three in four report that their organisation uses AI, with almost half stating it is used in a broad range of tasks and functions.

"The use of AI at work is clearly delivering a range of positive performance benefits. Most employees report increased work efficiency, access to accurate information, innovation, higher quality of work and decisions, and better use and development of skills and abilities. Almost half report that AI use has increased revenue-generating activity," the study found.

In the asset management arena, CREATE-Research chief executive Amin Rajan predicts that AI and GenAI, for their "revolutionary potential," have yet to reach the level of adoption that could visibly disrupt operating models covering core processes in their value chain, as well as the information systems and talent pools that support them (see Figure 1).

"But the wheels of change have started to turn in earnest," he said in a recent report studying the existential issues affecting global asset managers.

Rajan writes that at the heart of these technologies are conversational machine-learning algorithms.

This includes "ultra-sophisticated neural networks" that are transforming how asset managers make effective investment decisions, manage risks and communicate with clients, automating specific tasks in client onboarding, compliance checks, portfolio rebalancing, performance reporting and error corrections.

Yet, what often hinders asset managers going full steam ahead is "legacy thinking on the part of senior management".

This is based on the "innovator's dilemma" - while profit margins are high, pressure to change the operating model is limited, thus reduces the urgency to create a governance process to identify and prioritise the use cases that range "from quick wins to moonshots". Top executives also tend to be incentivised by short-term results at the detriment of long-term planning.

"Historically, they typically dictated the pace in the early phase of implementation. Anything with no track record of success has invoked fear of the unknown," Rajan writes.

"However, over time, accelerators gained ground as early adopters had an edge and reshaped competitive industry dynamics. This, in turn, created the fear of missing out, as price makers turned into price takers."

University of Technology Sydney industry professor of emerging technology Nicholas Davis sees the hype surrounding AI as the reason many organisations have chosen to adopt it headfirst.

"This is often without clearly thinking through a structured return on investment (ROI) model for the business and being thoughtful about the expected benefits and understanding the costs," Davis says.

"Using a classic ROI model, I think organisations are somewhat underestimating the investment-cost side, partly because in any technology investment you tend to need at least a one-to-one match between the tech investment side and what are called 'complementary investments' in the economic literature."

These are things like skilling employees, monitoring and reporting, and governance, along with costs around maintenance and expert advice.

On the benefit side, though, Davis sees most state-of-the-art GenAI models in an enterprise system delivering accuracy rates of about 80%.

Among financial services institutions, he also finds many using AI to strategically transition and upgrade legacy systems to newer ones, leading to faster and cheaper operations compared to five years ago.

Generally, Davis observes financial services firms across their different business units use AI in idiosyncratic ways.

"The greatest benefit either comes from a small but significant, say 2-5% uplift, across an entire background process, like cyber systems etc.," he says.

This includes AI potentially "radically changing" specific processes like claims adjusting or responding to a particular customer complaint.

"If businesses are able to halve the time it takes to process, it unlocks so much value," Davis says.

Cheaper, faster, stronger

On January 21, Microsoft's OpenAI, the maker of ChatGPT, announced it would spend US$500 billion over the next four years to build new AI infrastructure.

About one week later, a little-known startup from China, DeepSeek, claimed it does not need the laggard infrastructure to function and only spent US$6 million to train its R1 model with the help of just 2000 Nvidia H800 graphics processing units (GPUs).

This is a sliver of Meta's Llama, which according to Nvidia, uses 16,000 of its H100 GPUs.

J.P. Morgan Asset Management portfolio manager of US equity Eric Ghernati says companies like AWS, Microsoft, Google, Oracle and Meta are spending big on GenAI - but with minimal ROI - and it's reminiscent of the Dotcom era.

Collectively, they have spent US$500 million between 2023 and 2025, Ghernati calculates, but their revenue has remained somewhat static.

"The main takeaway with DeepSeek is the fact that it brought up a lot of innovation - just on the software and not on hardware. And that made DeepSeek's model perform just as good, if not as good as the best models that US AI labs spend billions of dollars training. That's a powerful thing to happen," Ghernati told the J.P. MAM Media Summit in May.

For Davis, what is interesting about DeepSeek is its mathematical breakthrough and unmatched efficiency compared to other models.

"And the fact that it came from an unexpected direction. It is genuinely proving to be much more efficient, particularly on the inference side when it calculates an answer for you as a user," he says.

Since DeepSeek is from China, many express concerns about security and privacy issues, predicting it might suffer the same fate as TikTok, which has been banned or restricted or is in the process of being banned in certain countries.

Davis points out that it doesn't make much difference whether he and his team are on the open web sending a sensitive query to DeepSeek or Claude or OpenAI.

"The reason being is it's not the fact that it's in China that's the worrying thing. It's the fact that you are breaching your duties just by sending that information outside of the company," he says.

"In that sense, it would be a mistake to say that it's just DeepSeek that is the problem. The problem is actually from a professional duty, risk management perspective."

While DeepSeek proves that it can do more for less, big investors such as superannuation funds and fund managers are waging serious money on critical AI infrastructure that eats up endless energy and vital to powering its American counterparts like ChatGPT.

Billions of dollars have been injected into data centres by Australian investors, with Macquarie Asset Management splashing $17 billion in 2025 alone on providers such as Applied Digital, and last year acquiring South Korea's Hanam Data Centre.

Australians' retirement money also has large exposures to data centres via AustralianSuper's $2.2 billion investment in Texas-based DataBank, while the Future Fund recently gained a 35% stake in Canberra Data Centres.

Among the even bigger players, KKR and Global Infrastructure Partners now own CyrusOne, outlaying US$15 billion, while Blackstone paid US$10 billion for QTS.

As Ghernati points out, the parallels between the GenAI boom and Dotcom bubble of the early 2000s are difficult to ignore.

The demand for bandwidth at the time was so huge that providers went overboard with building fibre optic cables, leaving a massive overcapacity once the bubble burst and companies were bankrupted.

Fibre optic cable expert Jim Hayes described it as "irrational exuberance" fuelling the Dotcom boom, a phrase coined by former Federal Reserve chair Alan Greenspan.

"During the boom years (between 1995 and 2000), perhaps as much as US$2 trillion was spent building 80-90 million miles of fiber optic networks. But in 2001, after the bust, it was estimated that 95% of the fiber was dark fiber; there simply was not enough internet traffic or revenue being produced to justify the investment," he recently wrote in ISE ICT Solutions & Education.

"The internet boom was exposed as a bubble and subsequently burst. Hundreds of companies rode the boom up and then fell back to Earth; many did not survive."

While many are drawing parallels between the Dotcom era of yesterday and the AI era of today, Ghernati says "at this time we do not believe there is an oversupply of data centres."

"While DeepSeek's advancements have prompted a re-evaluation of data centre demand, at this time we still believe the need for data centres will persist.

"Physical infrastructure currently remains vital for AI, and lower costs could lead to wider AI adoption, thereby boosting demand for data centres. However, it could in the future be potentially at a different scale or configuration," he says.

Davis senses equal and opposing issues with the parallels. On one hand there are high levels of "follow the investor money, with uncertain ways of catching returns".

On the flip side, there appears to be something promising with the latest boom.

"The Dotcom model saw a lot of selling literal web space and business and online website ideas that ended up not having much of a basis. But the internet, the underlying core idea and network, has proved to be incredibly valuable in many ways, and, of course, incredibly frustrating and risky in things like social media [such as online bullying] and the spread of misinformation and so forth," Davis says.

Now is the time, he says, to ensure that the current AI boom doesn't end up as a bubble bursting and that it is not misused or turned into a 'tech monster'.

"This is the moment where organisations can invest in good governance and be thoughtful about safety and engage with consumers and employees about where the value of AI is," Davis says.

Down to earth

ECP Asset Management portfolio manager Annabelle Miller says tariffs and DeepSeek are two poignant examples that have a real chance at challenging US 'exceptionalism.'

Firstly, the tariffs saw US President Donald Trump come to the table suddenly because the country did not have the rare earth minerals required to run its basic industry, Miller says.

"The second portion is DeepSeek. We all applaud the Magnificent Seven, or most of them, for their breakthroughs in AI, but suddenly a random company out of China that no one's ever heard of come out with some very advanced breakthrough at a much lower cost than what the US companies are providing."

It begs the question of just how far ahead is the US in terms of technology?

"For so long, the US has been at the mountain of technology breakthroughs - everything from Cloud to ChatGPT. Now, China is hot on their heels," she says.

"There are reasons to be concerned. But I don't think there's any disputing the fact that when you look at the top 10 companies globally by market cap, nine out of 10 of them are in the US."

In the same way mega-tech firms hope to achieve rosy ROIs, they also want a healthy return on their capital.

The Big Five US tech companies, for example, saw their capex go from US$150 billion in 2023 to US$300 billion anticipated this year, she says.

"We've seen companies like Microsoft turn around and say, 'This capacity that we're building is agnostic to whether you use it for AI or just general-purpose workloads'," Miller says, adding that many companies still have not migrated to the Cloud and will need to.

"Whilst there are going to be periods of digestion when building out a hardware and infrastructure footprint, it doesn't mean that there's any dent in the long-term structural trend of companies shifting to the Cloud. But there definitely are some physical constraints around how companies are going to manage that whole energy piece."

Each decade is coloured by an overarching investment theme. In the 90s, for example, Japan dominated, while the 70s were defined by booming demand in oil and energy stocks.

The last 10 years saw a torrent of investor capital flood the US market, bolstering its 'exceptionalism' status even further.

"We've seen European investor allocations rise from mid-teens to nearly 45% as of last year, into US equities. There is a hypothesis that at some point [the investment theme] must come to an end at the end of the decade. Yet, the last 15 years, it's been the same seven US companies dominating," Miller notes.

"So, the question remains: Is there a group of companies external to the US that are rising? The answer from our end is: not really."

Ghernati predicts that AI, whether it's from the US or China, is only going to get better.

"[AI] will be injected with capabilities to enable them to be smarter, like speak Mandarin, and provide planning and reasoning, and ultimately be able to solve bigger problems," he says.

"So, AI is not only going to be a personal companion, but it is going to be somebody that you give agency to make decisions on your behalf."

For now, users harbour trust issues with AI and must pick up the pieces once the human element is shot. Users express these concerns in countless surveys, including KPMG's and University of Melbourne's findings which reveal that 64% of Australians don't trust AI.

Many employees also report mixed impacts on workload, stress, human collaboration, compliance and surveillance at work.

For example, half said they use AI rather than collaborate with peers or supervisors to get work done, and one in five said AI use has reduced communication, interaction and collaboration, raising the question of how human connectivity will be retained in AI-augmented workplaces.

Buters lauds ChatGPT as an "amazing tool" but remains sensible about its risks and being led down the garden path.

"You have to be careful about what you put into it, because at the end of the day it's open architecture," he says.

"What you can't do is rely 100% on ChatGPT because it takes away personalisation. These days, people can tell if you've copied and pasted an email from ChatGPT, as opposed to doing it yourself. Sometimes there's that element of it being too refined, so it's not human anymore."

Another weakness is that many apps pull from old data and miss up-to-date information - whether it be tax rates or the Superannuation Guarantee and the like.

"If you're 90% confident in your understanding of a topic and want a bit of reassurance then ChatGPT is great, but it's important to understand that you still have to vet the outcome," Buters says.