International | Nov 20 2025
Artificial intelligence is the most significant change to humanity in the 21st Century but the complexity, scale and needs for the technology create both opportunities and risks.
- Bubbles past and present, how the media handles the narrative
- Gen AI collides with national security and corporate imperatives
- How the adoption of AI is progressing and what are the fiscal benefits
- Funding the biggest tech revolution in history
- Constraints around infrastructure for potential winners
- Challenges for investors, juggling the narrative versus the facts
By Danielle Ecuyer
Framing the central tension
“Two things can be true at the same time: a) the data centers to power AI could be as economically worthwhile an investment as railroads, and b) we could still experience at least one stock market crash along the way to its general adoption.”
[historian and economic commentator Niall Ferguson]
One’s memory often delivers flawed historical reality, so gleaning from personal experience is potentially fraught with erroneous signals.
Maybe it is investing muscle memory around the turn of the century Dotcom crash that brings forth a plethora of concerns, anxieties and ‘bubble’ narratives around AI and the associated infrastructure spend in 2025?
As noted by US advisory firm Carson Group, there has been so much talk about the “AI Bubble” that there are now articles on the “bubble in articles about the AI bubble”.
Bubblicious, if it wasn’t such an important topic for both investors, governments and companies.
Lessons from the Dot-Com Era according to ChatGPT
In the years leading up to the dot-com crash, mainstream media outlets were largely swept up in the euphoria of the internet age, celebrating innovation while downplaying risk.
Between 1994 and 1998, publications such as Wired, Forbes and BusinessWeek painted a techno-utopian vision of the “New Economy”, where traditional valuation metrics no longer applied. Analyses of New York Times and Wall Street Journal archives from that period show more than 80% of coverage on internet stocks carried a positive or neutral tone.
It wasn’t until 1998, following the Asian Financial Crisis and the collapse of Long-Term Capital Management, that scepticism began to creep into the financial press. Barron’s was one of the first to warn of “The Internet Bubble”, highlighting start-ups with no profits and inflated market capitalisations.
Yet, these early warnings were drowned out by an avalanche of bullish commentary, as cable business networks and glossy magazines continued to glorify tech founders and IPO millionaires.
Meanwhile, the Federal Reserve’s policy stance added fuel to the fire. After cutting rates aggressively in late 1998 to stabilise markets during the crisis caused by Long Term Capital Management’s demise , the Fed’s easy-money environment encouraged risk-taking and speculative inflows into technology stocks.
The liquidity surge helped push the Nasdaq up more than 80% in 1999 alone. When the Fed reversed course, hiking rates six times between June 1999 and May 2000 to contain inflation and curb exuberance, the tide quickly turned. The same cheap capital that had inflated valuations began to dry up, exposing the fragility of profitless tech ventures.
By 1999, valuations had reached dizzying heights, but even then the media’s cautionary notes remained rare.
A 2002 Journal of Communication study found fewer than 10% of technology-related stories adopted a negative or cautionary tone that year. It wasn’t until early 2000 —just as the Nasdaq reached its all-time high— that major outlets like The Economist, The Wall Street Journal and The New York Times began openly questioning whether the frenzy had gone too far.
Headlines such as “How High Can Tech Go?” and “When Will the Bubble Burst?” reflected a shift from hype to hesitation, but the warnings arrived only months before the crash.
Subsequent research from Columbia Journalism Review and Harvard’s Shorenstein Center found that by early 2000, nearly half of technology-market coverage had turned negative, a dramatic reversal that coincided with, rather than anticipated, the bursting of the bubble.
The lesson was clear: far from acting as an early warning system, the media largely mirrored investor sentiment, amplifying optimism until it collapsed under its own weight. And while journalists were slow to sound the alarm, the Fed’s pivot from accommodative to tightening policy provided the spark that pricked the bubble.
Today, as headlines once again oscillate between the promise and peril of artificial intelligence, the dot-com era remains a cautionary reminder of how liquidity, sentiment and media narratives can converge to inflate, and then unravel a market mania.
Today’s media cycle and bubble narratives
In contrast, in this present cycle the media, US and notably Australian flagship publications are positively tripping over themselves to quote the bears and the naysayers.
The latest have transcended everything from Michael Burry’s apparent shorts on Nvidia and Palantir (apparent because the data he tweeted on options activity was from the rear-view mirror) and Bank of America’s Michael Hartnett who is pointing to Big Tech corporate bond issuance in the last week as signs capex spending for Big Tech has transcended past free cash flows.
A concern around all bubbles is that spend is being debt funded that cannot be financed, i.e. insufficient cash flow is generated and the returns on AI-related spend do not come forth from revenue generation with sufficient margins and a failure for earnings growth to materialise.
All the while the opportunity cost of not doing enough hangs potentially like the Sword of Damocles over Big Tech if they do not invest.
Mark Zuckerberg was quoted as saying in September “If we end up misspending a couple of hundred billion dollars, I think that is going to be very unfortunate, obviously … But what I’d say is I actually think the risk is higher on the other side”.
Larry Page, co-founder of Google, stated, “I’m willing to go bankrupt rather than lose this race”.
It is not a stretch that America’s Big Tech companies see AI as an existential risk to their existence.
As a result, they are all investing heavily and for some, increasingly, beyond their potential payback cash flow generation to be part of the AI ‘revolution’.
AI as a geopolitical imperative
Arguably the US government sees the AI race as one of the most significant geo-political risks to its supremacy, national security, and the global order.
“If China reaches advanced AI first, it will surge ahead in autonomous weaponry, cyber warfare, and intelligence analysis. Economically, Chinese firms with the world’s best AI would dominate critical industries from biotech to finance.
“Beijing could set global rules for AI, exporting authoritarian standards and undermining freedom worldwide. Chinese deepfakes and propaganda could compromise the very integrity of information itself. Losing this race would irreparably harm America’s security and prosperity.”
Robert O’Brien, the former 27th United States National Security Advisor from 2019-2021 and Chairman of American Global Strategies LLC.
The race for AI dominance for both Big Tech and the US does not obviate and/or extinguish the risks for investors around earnings, valuations and share prices, it does however imply at some level AI-related large-scale companies could well be backstopped by the US Administration in an event of a ‘too big to fail’ crisis.
One such example which has been reported in the media concerns OpenAI, given its ginormous capex spend and unprofitable business model.
As we have written about previously, the AI investment cycle is heavily reliant on OpenAI and the increasingly cross-sectional investments and shareholders.
For more details see, https://fnarena.com/index.php/2025/10/09/australias-data-centre-boom-marches-on/
Adoption, productivity & compute cycles
Morgan Stanley continues to stand in contrast to the bearish narratives around AI. Looking at the potential impacts of AI, the broker conducted a deep dive into the S&P500 in terms of AI adoption benefits.
The broker’s latest proprietary research exercise revealed an ongoing rise in share of companies with “quantifiable” benefits from AI adoption.
In 3Q2025 24% of US companies identified as “adopters”, up from 21% in 2Q2025 and 15% a year earlier. Most of the quantifiable benefits related to “productivity gain”.
Inside the S&P500, 15% of member companies cited measurable benefits, up from 14% in 2Q2025 and 11% a year earlier. The gains are across all sectors; from financials to consumer discretionary.
The net benefits where the technology is now would be a little over US$900bn and that can translate to well over 20% increased earnings power that could generate over US$13trn of market cap upon adoption.
The broker estimates, as a base case, AI-driven efficiency will generate incremental net margin expansion for the S&P500 of 30bps in 2026 and 50bps in 2027.
The Global Head of Thematic Research stresses that is where the technology is at now and it is evolving “very, very quickly”.
Is the rate of AI moving in a non-linear fashion?
A topic that Morgan Stanley continues to pursue with several big American labs developing large language models that are gathering about ten times the computational power to train their next model, which would result in models that are twice as capable as they are now if the scaling laws hold true.
Like JP Morgan, Morgan Stanley is embracing AI internally and the Head of AI stated, “I’ve often made the analogy that we own a Ferrari, and we are driving around in circles in the parking lot.”
Inferring AI technology has already advanced so far beyond most people’s capacity to leverage it. That makes humans’ own capacity and awareness and education a limiting factor.
The Head of Firmwide AI explains what keeps him up at night is educating people at Morgan Stanley how to prompt and teaching people how to speak to the machine. Until people know how to do that, they do not understand the art of the possible.
The Global Head of Research, who was a former IT analyst, outlines whether an AI bubble is forming as is being reported in the media.
The AI cycle is viewed as developing along a similar path to previous compute cycles including the minicomputer, the PC, internet, mobile and cloud. Each compute cycle is around ten times larger in terms of installed compute.
The pattern is the same even if the numbers are bigger by transcending from millions to billions and now trillions.
There are trillions of installed CPU compute. With the historical pattern the same, some US$10trn of installed GPU compute will be required, which suggests to Morgan Stanley the cycle is in its “early innings”.
Note: the first refers to the general-purpose ‘brain’ of a computer (CPU), the Graphics Processing Unit, or GPU, is designed to render graphics by performing the same operation on many pieces of data at once.
Hyperscaler capex and the AI investment boom
As highlighted by ClearBridge Investments, four major hyperscalers —Microsoft, Amazon, Alphabet and Meta— are forecast to spend a combined US$378bn in 2025, a rise of 65% on 2024.
Current estimates are for capex spend to rise again in 2026.
Post the recent quarterly results, UBS notes hyperscalers are spending US$500bn, and nearly US$600bn by 2026.
Clearview’s spending estimate does not include Oracle, or the neo-clouds like CoreWeave or Nebius.
Carson Group detailed how AI-related capex spend is running at around 1.3% of US GDP in 2025, more than double the 0.5% in 2023, and over four times the size it was in 2019 at 0.3%. It’s expected to rise to circa 1.6% of GDP in 2026.
While much talk focuses around how AI is going to impact business and the economy, arguably AI is already having a considerable impact.
In real, inflation-adjusted terms AI-related hardware and software spending contributed on average 10.5%-points (105bps) across the first two quarters of 2025 to GDP — this is as much as consumer spending.
The latter makes up 70% of the economy versus circa 4% for AI-related investment.
Compared to 1995-2009, investment in processing equipment and software contributed an additional 68bps average per quarter; for the 24 years after that from 2001-2024 an additional 27bps average per quarter.
Investment spending begets revenue generation for other segments of the economy with the world’s largest listed company, Nvidia, in top spot.
Hence, the significance of the major tech companies investing in AI needing to generate returns on their capex.
JP Morgan forecasts US$6-US$7trn of AI capex by 2030 which will require US$650bn in perpetual annual revenue generation to produce a 10% return.
McKinsey estimates the technology (GenAI) has the potential to unlock US$2.6trn to US$4trn in additional value above the estimated value potential of traditional AI between US$11trn to US$18trn.
When looking back at Morgan Stanley’s historical work and forecasts, the broker observes the potential is always underestimated. This would infer upward earnings revisions for the AI enablers –and soon the AI adopters– are likely to continue.
Looking at the nearly 4,000 companies analysed globally by Morgan Stanley, one third covered by the analysts are saying AI has an impact on the investment case.
Funding the AI revolution, bond issuance
Increasingly, the issuance of corporate bonds by US Big Tech is fraying on the nerves of some strategists. Over the US summer, Morgan Stanley estimated credit markets will fund over US$1trn of global data centres spending through to 2028.
Comparing to past cycles, the broker points to several factors which differentiate the current cycle:
-Credit quality, for example Microsoft holds a AAA rating, higher than the US government.
-The cash generative characteristics of Big Tech mean they are underrepresented in credit relative to equity markets. Five of the large firms represent around 3% of the investment grade corporate bond market versus 19% of the equity market.
-Compared to previous capex booms, the credit profile is considerably stronger, whether it was the shale investment cycle or the late 1990s telecom boom.
-Credit markets are notably deeper which facilitates a healthier distribution of credit risk.
Mike Wilson, head of Global Strategy at Morgan Stanley, continues to see momentum gaining around AI adoption with sizable return on investment and efficiency gain opportunities for companies.
An estimated additional 40bps of net margin expansion in 2026 is flagged, up from 30bps previously and 60bps in 2027, up from 50bps previously.
Fielding investor questions on the parallels to the late 1990s, some distinctions remain. Current equity indices participants are viewed as higher quality than the late 1990s including free cash flow yield for median large cap stocks approaching three times above the level it was in 2000.
Valuations also appear less stretched. When the S&P500 forward price-to-earnings are adjusted for profit margins; the index is trading around a -35% discount to the 2000-2001 tech bubble.
Morgan Stanley emphasises operational efficiency, strong profitability and free cash flow generation are all important aspects of a higher quality index than the late 1990s.
UBS noted the hyperscalers are now impacting US debt markets with recent record capital raises, including US$30bn for Meta, US$24bn for Alphabet, and US$18bn for Oracle, with Amazon the latest with a US$12bn announcement.
Ex the latest from Amazon, UBS points to a rise of US investment grade debt of 115% y/y to US$211bn year-to-date of which the hyperscalers represent US$80bn.
If current capex forecasts remain, UBS anticipates a further US$140-US$175bn in new issuance in public investment grade credit markets and some US$100-US$125bn in private market financing.
The broker does not consider there to be any meaningful impact on the investment grade index and technology sector spreads.
UBS estimates 90bps impact for US investment grade and 325bps for US high yield by year end.
Depreciation schedules, obsolescence & the value cascade
Tech-industry analyst Beth Kindig noted recently “Big Tech’s capital spending, the core metric for the AI cycle, continues to impress”.
The latest quarter capex was up 19% q/q and 75% y/y; the strongest growth we’ve seen this year.
Amazon’s Andy Jassy captured the sentiment on his latest earnings call: “The faster we grow, the more capex we end up spending… We don’t procure it unless we see significant signals of demand”.
One of the new concerns as stressed by GFC-famed Michael Burry relates to the depreciation schedules for GPUs in data centres and concerns the cadence of new higher compute GPUs from Nvidia (annual iteration) is leading to swifter obsolescence of older chips.
If depreciation schedules as adopted by the hyperscalers do not align with the real life of the chips, then earnings are potentially being overstated.
This narrative is countered by Yardeni Research as follows: hyperscalers substantiate longer chip depreciation cycles by using a “value cascade model”.
Simply put, older GPUs replaced by newer GPUs are cascaded down to other functions using less power or compute intensity but are still able to run models and tasks which can generate economic value for years.
In reality data centres have been in use much longer than AI spend accelerated, with Yardeni highlighting there were as many as 4000 data centres operating in the US as a result of rising cloud computing demand.
Case Example: ABB Ltd
With so much noise around AI, it’s worth going back to basics and ABB Ltd in their recent Capital Markets Day presentation outlined some interesting takeaways around AI infrastructure.
ABB is a Swiss-based global industrial technology company specialising in electrification, automation, robotics and motion solutions.
CEO Morten Wierod stated in a CNBC interview when asked about AI and the build-out:
“We see very strong activity in the data centre segment, and the AI build-out is happening all over the world, of course, especially in the United States. We do not see any slowdown at the moment. It is more getting ready for what’s to come in the next 2-3 years. And then we are talking about new technology.
“What we see is a massive trend towards electrification and automation overall. Electricity as an energy source is growing at more than double the pace of any other energy source. That goes across the board in the Americas, in Europe and also in Asia.
“Data centres is the fastest growing market for ABB but represents only 7% of the overall business.”
ABB estimates data centres are expected to have an annual need of one million metric tons of copper by 2030 and are anticipated to represent around 2% of global copper demand by 2030.
Data centres’ consumption of electricity in the US is expected to rise to a 12% share by 2030 from 4% in 2023 and globally to 3% by 2030 from 1.5% in 2023. Major grid investments and innovative technologies will be required for power availability and fast connections.
AI growth implies the need for over 30% additional wafer manufacturing per year (12.8m AI chips from 1.1m wafers are expected in 2026) which is equal to four times current capacity by 2030.
Morgan Stanley believes the US will face up to a -20% shortfall of power available for data centres or -44GWs through to 2028.
This broker believes there may be limiting factors or a “set of cascading constraints facing the AI industry” with power the greatest limiter.
On balance two factors —the speed of AI capabilities and “cascading shortages” across the AI infrastructure value chain— mean those players which can secure the highest “relief” for the most challenged bottlenecks, such as power, will deliver better value.
Mustafa Suleyman, head of Microsoft AI, recently wrote:
“Climbing the exponential slope. The rate of progress has been eye-watering. This year it feels like everyone in AI is talking about the dawn of superintelligence.
“Such a system will have an open-ended ability of ‘learning to learn’, the ultimate meta skill. It would therefore likely continue improving, going far beyond human-level performance across all conceivable activities.
“It will be more valuable than anything we’ve ever known.”
Conclusion, according to ChatGPT
AI’s breakneck speed, immense computational demands and multi-trillion-dollar capital cycle make it one of the most consequential technological revolutions in modern economic history.
The scale of investment is unlike anything seen in previous compute cycles and is already reshaping productivity, infrastructure, capital markets, and geopolitics.
Yet the rapid build-out also brings risks: rising debt issuance, bottlenecks in power supply and chip manufacturing, potential misalignment between depreciation schedules and real chip life, and the ever-present possibility of market corrections reminiscent of earlier technology booms.
For investors, the challenge lays in distinguishing between structural, long-duration value creation and periods of speculative excess.
The evidence suggests AI’s long-term trajectory remains powerful and economically transformative, but the journey is unlikely to be smooth.
A balanced approach requires acknowledging both the extraordinary potential of AI and the practical constraints and volatility that may punctuate its development.
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