International | Oct 06 2025
AI investment has collided with a technology stock market melt-up prompting a cri-de-coeur of 'bubble'.
-Bubble warnings clash with evidence of self-funded growth as tech giants pour cash into AI
–Hyperscaler capex surges to record levels, underpinning the global data centre build-out
–Productivity gains emerge across finance, healthcare and retail as AI adoption accelerates
–China’s AI Plus Initiative signals global competition, with trillions in digital investment ahead
By Danielle Ecuyer
Bubble warnings meet a stock market melt up
Media have been flooded with speculation and experts opining on the “AI technology bubble”, drawing comparisons to the Dot-Com internet crash as evidence investors have been drawn into a similar paradigm of 2025’s modern day boom and bust cycle driven by FOMO and irrational euphoria.
Value investor Porter Collins, one of Steve Eisman’s team who bet against the US housing market and collateralised debt obligations prior to the 2007 GFC crash, tweeted on the back of a US$100bn deal between the world’s largest company Nvidia and OpenAI,
“This is no doubt a positive for both companies in the near term, but this is close to full Ponzi”.
‘Ponzi’ because of the circularity of the deal whereby Nvidia has invested US$100bn in OpenAI who will in turn invest in 10GWs of data centre capacity using Nvidia hardware. For context, Nvidia’s market capitalisation is tickling all time highs of US$4.59trn at US$188.89 per share and has generated cash flow from operations of almost US$70bn in 1H26.
Concerns around AI and the associated investment spend coincide with US technology stocks continuing to drag major US indices higher over the last quarter, and more surprisingly for some, over the often-volatile month of September.
The S&P500 rose 3.53% last month, the strongest September performance in 15 years, for a quarterly return of 7.79%. Nasdaq was the outperformer, up 5.61% in September and 11.24% over the quarter against a backdrop of ongoing macro concerns on tariffs, the US government shutdown and rising global geo-political risks.
Chinese big tech had an even more pronounced bullish surge, led by Alibaba and Tencent. Alibaba Cloud reported 26% revenue growth in its latest quarterly report and is reported as having 33% market share in China’s cloud infrastructure market.
As previously discussed in FNArena’s ongoing GenAI series (available at https://fnarena.com/index.php/tag/gen-ai/), the financial metrics around AI spend and the associated ancillary infrastructure are eye-wateringly large and at such scale, the nominal numbers almost become meaningless.
However, the rate of growth and change as of now has shown no signs of abating. For every bear calling the doomsday bell on AI and technology stocks, there are a suite of analysts crunching the latest numbers on the outlook for data centres, a major theme for Australian investors.
Before our readers jump off the AI narrative, maybe another glance is justified according to the most recent research insights and updates on this matter.
Are historical precedents clouding judgment?
Prior to taking a deep dive into the findings, a presentation by Tom Lee of Fundstrat on his General Principles for Investing focused on a topic that was also highlighted some years back by Hamish Douglas, formerly of Magellan, on cognitive bias.
Lee framed it as “Stop carrying a ‘Lehman hammer’” to avoid cognitive bias. In the context of what Porter Collins tweeted as a value investor, Lee is a renowned growth investor.
One could do worse than contemplating the possibility of bias once share prices have risen, resulting from historical experience and events like the GFC, which for many was a traumatic experience.
Billionaire founder of Amazon, Jeff Bezos was recently interviewed and asked about AI and investment bubbles at Reuters’ Italian Tech Week.
What follows is a snap shot which sheds some of Bezos’ thoughts on what seems to be happening today.
“Benjamin Graham, the great investor, is famous for saying in the short term the stock market is a voting machine; in the long term it’s a weighing machine and so as founders and entrepreneurs and business people our job is to build a heavy company.
“We want to build a company that when it is weighed it is a very heavy company. We do not want to focus on the stock price and so that is, you know, that will be misleading because it can be disconnected from the fundamentals and when bubbles happen. So that’s one thing that happens.
“The second thing that happens when people get very excited as they are today about artificial intelligence, for example, is every experiment gets funded. Every company gets funded. The good ideas and the bad ideas. And investors have a hard time in the middle of this excitement distinguishing between the good ideas and the bad ideas.
“And so that’s also probably happening today. But it doesn’t mean that anything that’s happening isn’t real. AI is real and it is going to change every industry.
“In fact, it’s a very unusual technology in that regard in that it’s a horizontal enabling layer. Today we talk about AI first companies like OpenAI an Anthropic and Mistral and so. There are so many startup companies that are kind of AI companies of various kinds and that’s normal for this phase.
“But that is not the biggest impact that AI is going to have. The biggest impact that AI is going to have is it is going to affect every company in the world. It is going to make their quality go up and their productivity go up. I mean every company, literally every manufacturing company, every hotel, every you know consumer products company etc. And so that is hard to fathom, but it’s real.
“There is no doubt we don’t know how long it will take exactly. We don’t know how quickly that transition will occur, and it’ll probably occur at different rates in different industries. But that is very real.”
The interview is available in full: https://www.youtube.com/live/4wTSZDZ_seU?si=yJyIhqjlkPhDImt4
Hyperscaler capex is accelerating not slowing
According to I/O Fund tech analyst Beth Kindig, the “AI bubble narrative ignores the re-acceleration in big tech capex”.
Kindig touches on why the AI trade was weaker earlier in 2025, beyond the tariff impacts. Total capex in 1Q2025 came in at US$77.3bn, a rise of 62.4% on the prior year but down -2% on the previous quarter, which could account for the concerns around the durability of spend.
The broad-based decline of investment from three of the four hyperscalers suggests it was a timing issue and power constraints emerged as the main reason for the quarter-on-quarter decline rather than a secular change.
By 2Q2025, as recently reported, hyperscaler capex rose 23% on the first quarter to US$95bn and up 63% on the prior year, with multiple flagged upgrades to 3Q2025 investment spend. The Street’s 2025 capex targets moved to around US$359bn from circa US$300bn, a 30% rise in the quarterly updates from the hyperscalers.
The spend is across servers, data centre shells, interconnects, cooling and networking. Microsoft guided its capex higher, now expected to reach US$80bn, above consensus’ original expectation of US$63bn at the start of 2025.
Alphabet (Google), Amazon, and Meta all communicated upgrades, with the latter’s capex up 100% in 2Q on the previous year and up 24% on 1Q25.
The AI spend was not only confirmed as not slowing, but in fact it is accelerating.
Microsoft exemplifies the rate of investment and demand
Morgan Stanley has postured that OpenAI’s US$300bn contract with Oracle infers capacity constraints with Microsoft and the latter’s preference to serve a higher-margin, long-term value, diversified enterprise customer.
Taking a deep look into the latest Microsoft results, the analyst stresses the acceleration in Azure (hyperscaler business) to 39% growth annually in constant currency, with commercial bookings up 35% and total value of contracted future revenue that has not yet been recognised up 35%.
Looking ahead, Morgan Stanley forecasts capex-implied Azure AI revenues of US$88.2bn to US$205.8bn in FY29 versus US$10.38bn to US$24.1bn in FY25, based on a 30% to 70% gross margin range.
CFO Amy Hood noted, “our margins on the AI side of the business are better than they were at this point by far than when we went through the same transition and the server to cloud transition”.
Even though Azure AI generated US$13bn at an estimated gross margin of 45%, the analyst’s forecasts adopt a more conservative 30% future gross margin, which infers considerable revenue upside if margins come in closer to 50% down the track.
Microsoft, Amazon, Google and Meta, on a trailing twelve-month basis, have generated US$493.31bn in cash from operations. Some US$291.35bn was allocated in capex, implying as long as the hyperscalers are growing earnings and cash flow, then funding can continue.
The hyperscaler and data centre thematic also includes the ongoing transition to the cloud.
The latest CIO survey results indicated an estimated 44% of application workloads are in the public cloud, up from 40% in 2Q2024. Morgan Stanley notes CIOs expect 68% of workloads to be in the public cloud by 2027 and 32% to remain ‘on premise’ over the next three years.
Growth in demand for data centres continues
In July research from consultants EY Parthenon highlighted Boston Consulting Group’s estimation of data centre demand growth of up to 491% from 2024 to 2030.
JP Morgan sits mid-range projecting 223% growth through to 2028, while a conservative estimate by Electric Power Research Institute forecasts demand could come in above 130% on the upper end over the next six years.
Boston Consulting Group estimates hyperscalers will need to invest around US$1.8trn between 2024 and 2030 to meet both AI and cloud demand.
In other forecasts as highlighted by Kindig, Bank of America estimates global data centre spending will rise 25% in 2025 to US$506bn and maintain a 20%-plus growth rate for the next three years.
McKinsey & Co projects capacity will nearly triple to 219GW by 2030 with AI contributing 70% of that demand at 156GW.
Citi’s Global Tech team has upgraded its AI hyperscaler capex through to 2029 by around 20% to US$2trn, with 2026 capex upgraded by 17% to US$490bn, equalling 55GWs of net new compute capacity for AI.
E&P’s latest technology update shows hyperscalers are likely to spend over US$600bn on capex next year.
Productivity gains monetise AI
AlpineMacro pointed to the Chair of AI and Robotics at Singularity University: “AI is moving so fast it is now a blur”.
The research had two major takeaways: AI is “rapidly evolving from a specialised tool into a fully democratised general-purpose technology” and “it’s not just enhancing innovation – it’s accelerating the entire innovation cycle.”
From a productivity perspective, the traditional macro-economic models are failing to pick up on the full impact of AI.
The author argues impacts are more visible in knowledge-based sectors including law, software, healthcare and finance, where AI is able to enhance output, reduce latency on decision-making and facilitate more complex problem-solving.
Integration of AI into workflows is uneven, with industrial applications lagging in terms of economic metrics.
Wrapping some estimated numbers around possible impacts, Morgan Stanley calculated the potential FY26 EBIT/EBITDA uplift from operating cost efficiencies underpinned by agentic AI implementation for Softlines Retail and Brands (Gap, Macy’s and Victoria’s Secret for example) by around 20% on average, with margins up around 200bps from cost savings of circa US$6bn.
The examples were ranked highest in terms of AI “reward” and “recognition” scorecard.
JP Morgan announced its AI program called LLM Suite, an internal portal using OpenAI to employ large language models across its extensive databases and software applications.
The bank’s chief data analytics officer Derek Wauldron explained JP Morgan is being “fundamentally rewired” for the AI era. This was a major topic at a four-day executive retreat in July.
FNArena sources say the offering is basically a ChatGPT wrap product for JP Morgan which can create simple PowerPoint presentations and Excel spreadsheets.
Is it being used in-house? Definitely yes, which just lends support to the adoption use cases of this technology and the demand.
“The broad vision that we’re working towards is one where the JPMorgan Chase of the future is going to be a fully AI-connected enterprise,” Wauldron stated.
In Citi’s discussion of possible AI impacts on productivity, this broker points to academic research that around 20%-40% of production tasks could be AI-automated, leading to labour cost savings of between 30%-40%.
In turn, this suggests a productivity gain of 6% to 16%, or growth on average of 0.5ppts to 1.5ppts a year.
Citi states AI investment is supporting US economic growth through demand-side impacts, and it remains early in the evolution with only 5% of GenAI projects at full scale and creating “meaningful” value.
AI investment is approaching levels that characterised the 1990s productivity boom. A read-through suggests the US could experience a similar productivity boom in the next few years.
Citi concludes AI technologies are still in an adoption phase where capex is being invested on data centres, electricity generation and grids, and in hardware to build out the infrastructure.
While still at an early stage, the potential productivity gains from AI appear significant. ChatGPT’s parent OpenAI, post the latest share sale for around US$6.6bn, is valued at US$500bn versus US$300bn at the US$40bn SoftBank Group raising in March.
That is a 67% uplift in six months and now places OpenAI as the world’s most valuable start-up, after SpaceX at US$400bn in July.
It is a global trend?
Chinese AI-related stocks have been performing strongly.
On August 29th, China’s State Council released the “AI Plus Initiative”, which articulated the country will “seek extensive and deep integration of AI application” across Science and Technology, Industrial Development, consumption upgrade, people’s well-being, governance and global co-operation, including targets of 70% penetration by 2027 and 90% penetration by 2030.
Cit’s analysis highlights a US$1.6trn investment required to meet global digital connectivity. A large proportion of the gaps are in developing economies.
The share of greenfield investments in the digital economy between 2020 and 2024 has grown to 28% from 20%, respectively, with Asia and large emerging markets —Mexico, Brazil and Saudi Arabia— being the largest recipients.
Funding from the US has been the largest source, followed by Taiwan, China, Singapore and Korea.
European asset manager Amundi has also looked at China’s position and “race for technological leadership” and concluded the race is far from over.
Amundi posits the US continues to hold major advantages in capital markets and innovation ecosystems, in contrast to Chinese regulators which historically crack down on the financial sector.
For China to move forward technologically, it needs to focus on end-demand innovation, with Amundi cautioning the country’s policies could crowd out market forces and stifle the development of innovation.
A Different Cycle to the 1990s (according to ChatGPT itself)
All of the above suggests today’s cycle is profoundly different to the late 1990s. The DotCom boom was characterised by companies with little revenue, no profits, and business models often untested in the real world.
When capital dried up, the bubble burst. In contrast, today’s AI build-out is being financed by the most profitable companies in history, deploying their own immense cash flows into physical infrastructure such as servers, data centres, cooling systems, and interconnects.
For investors, the implications are clear. Volatility is inevitable, and valuations will be debated fiercely, but the foundations of the AI era are being laid by companies with balance sheets and income statements to match.
The risks may lay more in execution, regulation, and power supply than in a speculative collapse.
The cycle may prove disruptive, but it is unlikely to end in the same implosion that scarred the technology sector a generation ago.
Technical limitations
If you are reading this story through a third party distribution channel and you cannot see charts included, we apologise, but technical limitations are to blame.
Find out why FNArena subscribers like the service so much: “Your Feedback (Thank You)” – Warning this story contains unashamedly positive feedback on the service provided.
FNArena is proud about its track record and past achievements: Ten Years On