Big technology companies are pouring vast sums into artificial intelligence, creating a boom for the businesses supplying everything from computer chips to data centres. But as spending climbs and cash flow comes under pressure, investors face a growing question: will AI ultimately generate enough demand and productivity gains to justify the investment, or could the spending boom end painfully?
In his caustic analysis of Wall Street entitled, ‘Where Are the Customers’ Yachts?’, Fred Schwed warned against the dangers posed to investors by rampant bull runs fuelled by hot narratives with the simple phrase, ‘Booms go boom'.
June’s SpaceX initial public offering, a putative stock market debut from Anthropic with an even loftier price tag, and the ongoing spending race among the so-called artificial intelligence hyperscalers all make the thought of an AI investment bust seem unlikely. But the sums involved are huge and debts are piling up, so the stakes are high, should anything unexpected go wrong.
Who is doing all this spending?
The heaviest spenders on AI include the large language model (LLM) developers and labs Alphabet, Meta, Microsoft, Amazon and SpaceX; cloud services provider and data centre operator Oracle; and the so-called ‘neoclouds’ Coreweave and Nebius.
Meta’s latest LLM is Muse, as it moves on from Llama to challenge SpaceX’s Grok, Microsoft’s Copilot and Alphabet’s Gemini, among others. Oracle is rolling out cloud infrastructure (OCI) that features data centres, including Project Jupiter in New Mexico, a site designed to support OpenAI and Softbank in their $500 billion Stargate Project.
The neoclouds, as their nickname suggests, offer a fledgling business model, whereby they buy graphics processing units (GPUs) and server computing power, often with a hefty amount of debt. They then rent out that computing power to a mixture of AI start-ups, hyperscalers and even governments, who want ready access but are worried that shortages of compute (the power, hardware and infrastructure required run AI systems) could crimp their development plans.
Fears of being left behind in the AI technology race are supporting a tidal wave of investment, as OpenAI’s ChatGPT, Anthropic’s Claude and Amazon’s Nova jostle with Muse, Grok, Copilot and Gemini, as well as Chinese rivals DeepSeek, Kimi and Qwen. That spending is in turn creating bottlenecks and shortages of labour, water supply, power and silicon chips, which are only serving to further boost demand as no-one wants to miss out.
The launch of ChatGPT by OpenAI in late 2022 sparked a step up in capital investment, but the appearance of DeepSeek R1 in early 2025 fuelled a stunning spending splurge.
The numbers are getting big
Aggregate capex across Alphabet, Amazon, Coreweave, Meta, Microsoft, Nebius, Oracle and SpaceX has shot from $32 billion in 2016, or barely 8% of combined sales, to a forecast $971 billion, or 46% of sales this year.
Analysts believe the octet will spend another $1.3 trillion in 2027 and $1.4 trillion in 2028. The latter year does see forecasts suggest that the capex-to-sales ratio will finally start to ebb as combined revenue growth starts to outstrip spending growth, based on assumptions regarding adoption of AI by the ultimate end users – companies and consumers – who start to generate the productivity gains and time and cost saving they crave from their use of the technology.
The hyperscalers may well be wishing that such a forecast is correct and they can get off the spending treadmill. September’s calls from Anthropic’s Dario Amodei, OpenAI’s Sam Altman and SpaceX’s Elon Musk that the speed of AI frontier development needed to slow down were seen by some as a desire to achieve just this, or get government investment or at least some degree of regulatory protection from overseas rivals – notably the Chinese, whose AI models are taking a different, open-source, less power-and-compute intensive path.
Either way the current capex boom will, in time, be followed by a surge in depreciation costs, as data centres go live and deploy their silicon chips, computing power and water and energy infrastructure, with the result that revenues must also rise quickly to cover the costs.
This is the core of Michael Burry’s bear thesis on AI, and also the conclusion of a paper from Columbia University, which estimates that AI revenues must hit $3.5 trillion by 2032 to provide a return on the investments and spending to which the hyperscalers and AI labs are committed.
That figure represents just under 9% of US GDP and almost matches American consumers’ annual consumption on food.
Starting to think about the outcomes
None of this means demand will disappoint or that the AI boom is going to fizzle, although it means the risks to the hyperscalers, and investors in them or lenders to them, are greater.
This is because free cash flow is withering; capital intensity, in terms of fixed assets on the balance sheet, is growing; debt is accumulating, to perhaps add financial gearing to increased operational gearing and thus greater swings in earnings relative to changes in revenues; and competition is hotting up, as the hyperscalers emerge from the oligopolistic, or even monopolistic, areas of dominance into a new arena where they are battling each other for supremacy.
There are four possible scenarios:
- The capex boom continues, to the enormous benefit of the providers of key picks and shovels, either literally or metaphorically, in the form of suppliers of everything from servers and computing power to data centres, earth-moving equipment and power and water supplies, all the way through to silicon chips and semiconductor production equipment. At the same time, hyperscaler cashflow continues to wither, so that the spenders have to raise fresh equity or more debt.
The risk here, therefore, is that capital markets rebel and either charge steep interest rates on the debt or lowly valuations for the equity, to reflect the increased risk. Or they just refuse to hand over any fresh cash, if they feel the potential returns may be inadequate relative to the risks, with the result the spending finally has to stop.
Nvidia is trying to short-circuit this by providing vendor financing of its own, but the experiences of Cisco, Alcatel, Lucent, Nortel and more in the late 1990s and early 2000s during the tech, media and telecoms boom show that looks great when all goes well but becomes a horrible mess if any customers get into trouble.
- The capex boom continues and free cash flow surges as demand for the AI labs’ large language models (LLMs) rockets, amid genuine proof that they provide huge savings or productivity benefits, or both, to governments, corporations and consumers alike. In this case, the dangers lie in competition between the labs and the LLM developers as supply surges and token prices collapse, akin to what happened with fibre and broadband demand during the tech, media and telecoms bubble of 1998-2000.
Demand was there, and ultimately exceeded all expectations, but an onrush of supply blew up a lot of the early-stage movers. Any collapse in prices for tokens, at a time when long-term compute and power and water deals look to come with fixed prices, could hammer profits.
- In the best of all worlds, demand rockets but the AI hyperscalers and LLM labs and developers are able to calibrate capex right on cue, to maximise profits. The danger here is this jackpot attracts more competition, although by then the early-mover advantage could be too great to overcome.
Regulatory intervention, or customer push back, may then be the next threats, but these challenges have yet to truly disrupt the dominant positions in their target markets that Microsoft, Alphabet, Meta and others have successfully carved out for themselves in their current incarnations.
- In the worst of all worlds, capex is cut and free cash flow stays weak as either demand disappoints or competition erodes any profits that the AI developers and labs seek, as their moats do not prove strong enough to fend off rivals. In this case, users of AI get all of the benefits, rather than the providers.
This situation is not as far-fetched as it sounds. US corporate earnings, as benchmarked by the total recorded by members of the S&P 500, have grown at a compound annual growth rate of 6.4% over the past 20 years and by 6.7% over the past 40, despite all of the benefits offered by technology to wider economic growth.
What are the other risks?
In sum, the biggest spenders on AI development and infrastructure face the same threats as any other company: competition, technological obsolescence, customer disaffection and regulation.
All of them could work to erode hoped-for profits and cash flow, whether demand meets, exceeds or disappoints expectations.
Other dangers lurk, such as short-term bottlenecks in the supply chain as buyers scramble for silicon chips, even if advances in production technology and available capital can overcome those given time, unless customer end-demand disappoints in the interim, or it turns out buyers have been double- or triple-ordering semiconductors in an effort to find supply.
The worst case is that AI simply fails to live up to expectations and corporations and consumers spend less than expected, as the productivity benefits of using AI prove more ephemeral than hoped and nowhere near enough to justify the spending that the LLM developers and hyperscalers need to cover their own investment and costs.
What to keep an eye out for
The key then is to find signs that may prove valuable when it comes to working out which of the four scenarios is developing. By the time one of them shows up in company profit and loss accounts, cash flow statements and balance sheets it will be too late, and the share prices and bond prices will have already moved – and probably by a lot.
Token and GPU prices could be helpful guides, and the more companies that talk about productivity and profit benefits from their use of AI the better. While the opposite also holds true, so any surveys on corporate return on investment will help to frame the debate.
The most obvious of all would be a change in trajectory at the AI hyperscalers’ capex budgets. Not one of them has announced a cut as yet and they have offered a succession of upgrades instead, with the upcoming quarterly results season the latest test here.
Yet even that could mean some investors end up leaving a fading party, or joining a joyous one, too late. One gauge of sentiment therefore remains the Philadelphia Semiconductor Index, or the SOX.
Silicon chip and semiconductor production equipment (SPE) firms are among the key beneficiaries of the spending boom, with Nvidia top of the tree. The GPU designer’s phenomenal profits and cash flow show that the AI spending boom is for real.
Any slowdown or stumble here, or at Micron, Advanced Micro Devices (AMD) or Broadcom, for example, could be a sign that spending is about to cool, slow down or fall. If there has been any double- or triple-ordering, or any vendor financing that has been provided to the wrong people, it may well show up here first.
The SOX was very quick to respond to the launch of ChatGPT in November 2022 and stumbled when the release of DeepSeek R1 initially challenged the narrative of more spending, more compute, and more power with its cheaper, open-source model in early 2025. The emergence of that rival only served to fuel further spending by the American hyperscalers, to the benefit of the SOX index and its 30 members.
Russ Mould: Investment Director
Russ Mould is AJ Bell's Investment Director. He has a Master's degree in Modern History from the University of Oxford and more than 30 years' experience of the capital markets.
He started out at Scottish...