What Happened
The Guardian published a detailed data analysis on 7 June 2026 using six charts to map the current state of the global AI boom - from infrastructure spending and stock market concentration through to real-world productivity evidence and environmental costs. [1]
The piece lands at a pointed moment. Elon Musk's SpaceX, which develops AI models alongside its rocket business, announced it is seeking a US$1.77 trillion valuation on the US stock market. Anthropic, the startup behind the Claude chatbot, filed for an initial public offering. OpenAI, the developer of ChatGPT, is expected to follow.
Those IPO moves sit against a backdrop of extraordinary capital deployment. Spending on AI - from datacentres to chips - is forecast to race from US$765 billion this year to US$1.6 trillion in 2031, according to Goldman Sachs. Meanwhile, ChatGPT has reached 1 billion monthly active users according to Sensor Tower data, a record for any app. Enterprise adoption has climbed from 33 per cent of companies in 2023 to nearly 80 per cent today, per McKinsey.
The spending is not confined to the hyperscalers. News site Axios recently reported on an unnamed company that spent US$500 million in a single month on licences for Claude Code - a figure that illustrates just how fast AI costs can compound inside a large organisation.
Why It Matters
The gap between what is being spent and what is being earned is the central tension in this story, and it matters for anyone making AI budget decisions right now.
Jim Bianco of Bianco Research found that 41 AI-related stocks now account for nearly half the S&P 500's market value. That level of concentration is not normal. When a single theme drives that much index weight, a sentiment shift does not need to be catastrophic to cause serious damage across a portfolio.
Neil Wilson, an analyst at investment platform Saxo UK, is direct about the risk. "The entire market has become one giant AI edifice," he said. "The danger is a repeat of the dotcom bubble - a huge crash, and years of lost returns. By some measures valuations aren't as stretched as then but this looks like an incredibly dangerous market." [1]
For Australian businesses, the concern is not abstract. Superannuation funds, listed investment companies, and technology-heavy growth portfolios all carry significant exposure to the US tech sector. A sharp correction in AI valuations would ripple through Australian retirement savings quickly. At the same time, local companies are being pitched AI contracts at prices that assume the productivity gains will materialise on schedule - an assumption that deserves scrutiny before the ink dries.
The environmental dimension adds another layer of complexity. One voice in the Guardian analysis asked plainly: "Has the government calculated whether such an expansion is feasible? Do they have the money to do it? Have they taken into account the associated environmental damage?" [1] That question applies equally to Australian state and federal governments currently committing to AI infrastructure programmes.
Key Details
The productivity evidence is real but uneven. Anthropic's Claude Mythos model is calculated to reach a 50 per cent success rate on tasks that would take a human expert between eight hours and two days. That is a meaningful capability threshold, not a marketing claim. The problem is that capability benchmarks and deployed business value are different things, and the distance between them is where most AI projects currently stall.
The macroeconomic picture is genuinely strange. Despite significant reductions in US government employment under the Trump administration and mass layoffs across multiple industries, US GDP kept growing - 2.1 per cent in 2025 and 1.6 per cent in Q1 2026, according to the US Bureau of Economic Analysis. A Harvard economist's analysis, however, suggests that without the datacentre construction boom, those figures could look very different. Specifically, "investment in information processing equipment & software" accounted for 92 per cent of US GDP growth in the first half of 2025. [1]
That is a fragile foundation. An economy growing primarily because one sector is spending heavily on infrastructure that has not yet proven its return is exposed in ways that standard GDP readings do not capture.
Goldman Sachs itself acknowledges the execution risk. One analyst quoted in the Guardian piece put it plainly: "At the scale of capital being committed, even modest delays in execution invite real scrutiny around the demand assumptions used to underwrite these investments." [1] The costs, meanwhile, are not coming down as fast as the industry hoped. As one source in the analysis noted: "The costs are getting completely out of control." [1]
Background and Context
The current AI cycle has structural similarities to the late 1990s dotcom boom, but also genuine differences. The technology actually works - ChatGPT's 1 billion monthly active users is not a vanity metric, and enterprise adoption at 80 per cent of companies is a real behavioural shift. The dotcom era produced companies with no revenue and no product. This era has both.
What it shares with the dotcom period is the valuation logic: buy now, profit later, and trust that the total addressable market will eventually justify the price. That logic held for Amazon and did not hold for Pets.com. The question is which AI companies are which.
The optimist case is straightforward. As one voice in the Guardian analysis put it: "We are very much at the early stages of the AI revolution still. There are many people doing tasks that could be done by an AI. The amount of change we are going to see will be huge." [1] The Wayne Gretzky framing - "You miss 100% of the shots you don't take" - also appeared in the piece as a summary of the prevailing corporate attitude to AI investment. [1]
The sceptic case is equally coherent. If 92 per cent of US GDP growth in early 2025 came from datacentre investment, and that investment is predicated on demand projections that have not yet been validated by actual revenue, then the growth is circular. Companies are spending on AI infrastructure because they expect AI to generate returns; the spending itself is generating the GDP growth that makes the investment look justified; and the cycle continues until it does not.
For Australian operators, the ACCC's ongoing scrutiny of digital markets and the OAIC's AI governance guidance both signal that regulatory costs will be part of the total cost of AI ownership - a factor that is rarely included in the vendor ROI models being pitched to local businesses right now.
What Comes Next
The IPO pipeline is the near-term test. If Anthropic and OpenAI achieve the valuations they are seeking, it validates the current pricing of AI assets and gives the boom another leg. If either IPO disappoints - or if the private credit market tightens, as Wilson flagged as a risk - the repricing could be fast.
For Australian businesses, the practical question is how to participate in genuine AI productivity gains without overcommitting to infrastructure or licences that assume a pace of capability improvement that may not arrive on schedule. That is a question of AI strategy first, AI automation second, and AI training for the people who will actually use the tools third.
The Goldman Sachs forecast of US$1.6 trillion in annual AI spending by 2031 will either look prescient or reckless depending on whether the productivity gains show up in the next two to three years. Right now, the honest answer is that nobody knows - and any vendor or consultant who tells you otherwise is selling something.
Frequently Asked Questions
Is the AI investment boom comparable to the dotcom bubble?
There are real similarities and real differences. The dotcom era featured companies with no revenue and no working product trading at extraordinary multiples. Today's AI leaders - OpenAI, Anthropic, Microsoft's AI division - have genuine products with massive user bases. ChatGPT reaching 1 billion monthly active users is a concrete data point, not a projection. The similarity lies in the valuation logic: current prices assume future returns that have not yet materialised at scale. Neil Wilson of Saxo UK described it as "an incredibly dangerous market," and the concentration of 41 AI stocks representing nearly half the S&P 500's value is historically unusual. Whether that ends in a crash or a gradual normalisation depends heavily on whether productivity gains show up in corporate earnings over the next two to three years.
Why does it matter that datacentre investment accounted for 92 per cent of US GDP growth?
A Harvard economist's analysis found that "investment in information processing equipment & software" accounted for 92 per cent of US GDP growth in the first half of 2025. That figure matters because it reveals how narrow the foundation of recent US economic growth actually is. When a single category of capital expenditure is doing almost all the work of keeping an economy growing, the broader economy becomes unusually sensitive to any slowdown in that category. If AI demand projections turn out to be overstated, or if datacentre construction slows due to financing constraints or energy supply issues, the knock-on effect on GDP could be significant - and that would affect Australian export markets and superannuation fund returns, not just US tech stocks.
What should Australian businesses do differently given these risks?
The core discipline is separating genuine productivity use cases from speculative infrastructure commitments. Enterprise AI adoption has climbed to nearly 80 per cent of companies according to McKinsey, but adoption rate and return on investment are different metrics. Australian businesses should be asking vendors for evidence of actual productivity gains in comparable deployments, not just capability benchmarks. They should also factor in the full cost of ownership - including licensing (one unnamed company reportedly spent US$500 million in a single month on Claude Code licences), integration, training, and the regulatory compliance costs that the ACCC and OAIC are increasingly making explicit. A structured AI strategy process that stress-tests demand assumptions before capital is committed is the most practical risk management tool available right now.
How does Anthropic's Claude Mythos model performance relate to real business value?
Anthropic's Claude Mythos model is calculated to reach a 50 per cent success rate on tasks that would take a human expert between eight hours and two days. That is a meaningful capability threshold - it means the model can genuinely substitute for expert human effort on a defined class of complex tasks, roughly half the time. The gap between that benchmark and deployed business value, however, depends on whether those tasks are actually bottlenecks in your organisation, whether the 50 per cent success rate is acceptable given the cost of failures, and whether your team has the skills to supervise and correct AI outputs reliably. Capability benchmarks are a starting point for evaluation, not a substitute for it.