ECONOMY & TECHNOLOGY
Hundreds of billions of dollars are flowing into data centres, chips and power plants every year. The International Monetary Fund now says the world economy is leaning heavily on that spending, and that a sudden loss of faith could hurt. Here is what is really going on.
Last Tuesday in Bangkok, at the annual meetings of the IMF and the World Bank, Kristalina Georgieva said something that sounded almost casual but landed like a warning. The global economy, the IMF's managing director told an audience, is now standing on one leg, and that leg is artificial intelligence.
It is a striking image, and it is also uncomfortably accurate. The Fund's latest World Economic Outlook, published on 13 October, puts global growth at 3.1 percent for 2026 and 3.4 percent for 2027. It credits strong AI infrastructure investment with offsetting the damage from higher oil prices caused by the war in the Middle East. In other words, one of the few forces holding up the world's output at the moment is a construction boom in server halls, silicon and electricity supply, financed largely by a handful of American technology giants.
That boom may turn out to be the foundation of a new industrial era, as significant as electrification or the internet. It may also turn out to be one of the most expensive misjudgements in the history of capital markets. Honestly, the evidence available today supports both stories, and anyone who tells you with total confidence that they know the ending is selling something.
The short version
- The four largest US cloud companies plan to spend roughly $725 billion on capital investment in 2026, up about 77 percent from around $410 billion in 2025. Much of it is for AI.
- The IMF's October 2026 reports name an AI market correction as one of three "salient risks" to financial stability. Its downside scenario cuts global GDP by 1.1 percent by 2028 and US output by 1.8 percent.
- Revenue from AI is growing fast but remains small next to the spending. The productivity gains that would justify the investment are real in some tasks and still unproven across whole economies.
- Chips, power and financing are the three bottlenecks, and each carries its own risk.
- The most likely outcome is not "revolution" or "bubble" but a messy mixture: valuable technology, overbuilt capacity in places, and painful losses for whoever paid the wrong price.
The scale of the spending is hard to overstate
Start with the numbers, because they are the reason this debate exists at all.
In 2025, Microsoft, Alphabet, Amazon and Meta together spent about $410 billion on capital expenditure, a record at the time. After their first-quarter 2026 earnings reports in the spring, their combined plans for this year point to roughly $725 billion. Microsoft has said its calendar-year 2026 figure will reach about $190 billion, including around $25 billion that it attributes simply to higher component prices. Amazon has talked of nearly $200 billion. Meta lifted its range to between $125 billion and $145 billion. Alphabet's guidance has sat somewhere around the $180 billion to $200 billion mark, depending on which quarter's update you read, and the company has said spending will rise "significantly" in 2027.
| Company | Approximate 2026 capital spending plan | Notes |
|---|---|---|
| Amazon | Close to $200 billion | Includes its satellite business as well as cloud capacity |
| Microsoft | About $190 billion (calendar 2026) | Around $25 billion attributed to higher component prices |
| Alphabet | Roughly $180 to $200 billion | Guidance raised in 2026; 2027 expected to rise "significantly" |
| Meta | $125 to $145 billion | Guidance raised in April on component costs and data-centre capacity |
| Big four combined | About $725 billion | Roughly 77 percent above 2025 |
Figures are company guidance as reported after first-quarter 2026 earnings and are rounded. Guidance changes frequently. Not all of this spending is for AI, though one analyst estimate puts the AI share of hyperscaler capex at around three-quarters.
To give that scale some texture: Apollo's chief economist Torsten Slok estimated earlier this year that capital spending by Amazon, Microsoft, Alphabet, Meta and Oracle would amount to about 2 percent of US GDP. By one comparison that circulated widely, it exceeds the combined military budgets of Germany, France, Britain, Japan, Italy and Canada. Goldman Sachs puts the three-year total from 2025 to 2027 at around $1.15 trillion, and some Wall Street banks expect annual spending to pass $1 trillion in 2027.
The companies themselves are not shy about their reasoning. Alphabet's chief executive Sundar Pichai has argued that the risk of underinvesting is dramatically greater than the risk of overinvesting. That is a rational position for a firm that fears being left behind in a winner-take-most race. But it is worth remembering that a rational choice for each company individually can still add up to a collective overshoot. Every railway company in the 1840s had a perfectly good reason to lay more track.
Where does the money come from?
For most of the last decade, the great technology firms funded their ambitions from enormous cash flows. That is changing, and the change is at the heart of the financial stability concern.
Look at free cash flow, the money left after investment. By the first quarter of 2026, several of the giants were reporting sharp declines. Amazon's trailing twelve-month free cash flow had shrunk to around $1 billion, a drop of about 95 percent on the year, as the data-centre build accelerated. Meta was generating a few billion dollars of free cash flow a quarter against planned annual capital spending above $125 billion. Microsoft and Alphabet saw free cash flow fall as well, even while their profits were strong.
When profits no longer cover the build, borrowing fills the gap. The IMF's Global Financial Stability Report this month includes a background analysis titled "Rising Leverage in the AI Ecosystem". According to figures cited from Morgan Asset Management, US hyperscaler bonds made up about 2 percent of US-dollar investment-grade issuance between 2022 and 2024. For 2026, the expected share is around 9 percent. That is a big shift in a market where pension funds, insurers and banks hold much of the paper.
The circularity problem
Then there is the strange geometry of the deals between AI companies themselves. In September 2025, Nvidia announced a letter of intent to invest up to $100 billion in OpenAI to help build data centres filled with Nvidia chips. OpenAI, in turn, signed huge agreements with Oracle, AMD, CoreWeave and others, with reported compute commitments climbing past $1 trillion. Nvidia has invested in CoreWeave and in other customers. A dollar can travel from a chipmaker to an AI lab to a cloud provider and back to the chipmaker as revenue.
Defenders say this is ordinary business. Vendor financing and strategic investment exist in every capital-intensive industry, and Nvidia has stated that its OpenAI investment would not be used for direct purchases of its products. As of late last year, the Nvidia deal was still described as under negotiation, not finalised. Critics see something more troubling. They point to the telecom bust of 2000 to 2002, when equipment makers such as Lucent and Nortel lent money to customers who used it to buy their equipment. When those customers failed, the suppliers were left holding both the bad debts and the collapsed demand. Global Crossing went bankrupt, and executives at several firms faced legal consequences over round-tripped revenue.
It would be unfair to say today's situation is identical. Nvidia's customers are, for the most part, some of the richest companies on earth, and Nvidia's profits are real. But the dependence on a small number of heavily indebted or loss-making AI labs for a meaningful part of future demand is a legitimate concern. The leading AI developers, including OpenAI, have reported large losses relative to their revenue, and their commitments far exceed what they currently earn. Their ability to keep paying depends on continued access to fresh capital from investors, which in turn depends on confidence. Confidence is the one input no one can manufacture on demand.
The productivity question: where is the payoff?
Every boom rests on a story about future profits. The AI story goes like this: models will get capable enough to automate or accelerate a large share of cognitive work, productivity will jump, and the companies that supply and deploy the technology will capture a meaningful part of the gain. The infrastructure being built today is the down payment.
The honest state of the evidence is that revenue is growing quickly from a small base, while economy-wide productivity gains remain hard to see in official data.
On the revenue side, there are real signs of traction. Microsoft reported that its AI business reached an annual revenue run rate of about $37 billion in early 2026, more than double the year before. Google Cloud and Amazon's cloud arm have posted strong growth, with Google Cloud expanding by well over 50 percent in the first quarter. Meta credits AI-driven ad targeting with a jump of roughly a third in quarterly revenue. These are not imaginary numbers.
But set the Microsoft figure beside its capital spending plan. A run rate of $37 billion against $190 billion of investment is a ratio of more than five to one. Some analysts have calculated that the industry would need on the order of a trillion dollars in annual AI revenue to earn a decent return on the capital being deployed, against a base today of perhaps a tenth of that. The gap does not prove failure. Young technologies often spend ahead of revenue, and the spending may precede the demand by years. It does mean that the investment case depends heavily on growth continuing at extraordinary rates for a long time.
What the studies actually say
At the level of individual tasks, the research is encouraging. Studies of customer-service staff, software developers, consultants and writers have found meaningful speed and quality improvements when workers use AI assistants, with the largest gains often going to less experienced workers. Programmers can finish certain tasks faster. Analysts can produce first drafts in minutes. Translators, designers and researchers report real time savings.
At the level of firms and economies, the picture is murkier. Surveys of businesses repeatedly find that many pilot projects never reach full deployment, that integrating AI into messy real-world workflows is harder than demos suggest, and that the measured return is uneven. Some careful studies have even found that experienced developers working on complex codebases were slower with AI tools than without them, even though they believed they were faster. Output that looks polished can still need expensive checking.
Economists have seen this movie before. In 1987 the economist Robert Solow quipped that you can see the computer age everywhere except in the productivity statistics. Eventually the gains did show up, but only after about a decade of reorganising how businesses worked, and in the United States they arrived in a burst in the late 1990s and then faded. Electricity followed a similar path: factories had electric motors for decades before they redesigned their floor plans to take full advantage. General-purpose technologies tend to pay off slowly, then suddenly, and the delay between investment and return can be long enough to bankrupt the investors who financed the early stage.
"We now have an economy that is standing on one leg, the AI leg."Kristalina Georgieva, IMF Managing Director, Bretton Woods Committee event, October 2026
This is the crux of the matter. The IMF is not saying AI will fail. Its own analysis acknowledges that AI investment has provided a significant demand boost. What it is saying is that markets and corporate budgets are priced for a particular future, and that if the productivity and profit gains disappoint, the adjustment will not be gentle.
Jobs: disruption, uncertainty and unequal exposure
No part of the AI debate generates more anxiety than employment, and none is harder to forecast with confidence.
The optimistic view says technology has always destroyed some jobs and created more, that AI will take over tedious tasks and free people for better work, and that the economy will adapt as it has done after every previous wave. The pessimistic view says this time is different because AI targets cognitive tasks, the kind that educated workers in rich and middle-income countries have relied on for security, and because it can improve quickly enough to outrun retraining.
What can we say with reasonable confidence? First, exposure is broad. IMF research from earlier in the boom estimated that around 40 percent of jobs worldwide have some exposure to AI, and closer to 60 percent in advanced economies, where white-collar work dominates. Exposure is not the same as elimination. For many roles, AI will change tasks without removing the job. But for others, especially routine writing, coding, customer support, translation, basic legal and financial analysis, and administrative work, it can substitute for human labour.
Second, the early signs are showing up at the entry level. Graduates trying to land their first jobs in software, marketing, customer service and professional services report a tougher market, and some large companies have said openly that they will grow headcount more slowly because of AI. In the spring, reports emerged that at least two of the largest technology companies were preparing to trim their workforces even as they raised investment. Whether that is cause and effect or coincidence is still debated, since interest rates, post-pandemic overhiring and tariffs have all been moving too. But the direction of travel worries labour economists, because entry-level jobs are how people build the skills that make them valuable later.
There is a subtler link between jobs and the bubble question. The strongest financial justification for today's spending is that AI will replace expensive labour. But the same logic means that if AI succeeds spectacularly at replacing workers, consumer incomes could weaken, and with them the demand that companies hope to serve. The boom needs AI to be powerful enough to transform business, yet gentle enough not to undermine the customers. That is a narrow path.
The silicon bottleneck and the geopolitics of chips
Underneath every AI model is a physical supply chain of astonishing fragility. The most advanced AI chips are designed by a small number of firms, led by Nvidia, and manufactured almost entirely by one company, Taiwan Semiconductor Manufacturing Company, using machines supplied by one Dutch firm, ASML, with high-bandwidth memory coming from a handful of Korean and American producers. Every link is concentrated, and every link sits within reach of political shocks.
Nvidia's rise captures the dynamic. The company became the world's most valuable firm and reached a market capitalisation of $5 trillion in late 2025. It has said it had around $500 billion in chip orders booked through 2026, a number that excludes some of the potential OpenAI business. Its largest customers, Microsoft, Meta, Amazon, Alphabet and a few others, account for a large share of its revenue. That concentration is a strength while those customers keep buying and a vulnerability the moment any of them pauses.
This produces three kinds of risk for the global economy. The first is supply risk: any disruption in Taiwan, whether from earthquake, blockade or conflict, would halt the production of advanced chips and ripple through every industry that uses them. The second is policy risk: export controls, tariffs and retaliation can reshape who buys what with little warning. The third is price risk, which we are already seeing. Memory and component costs have surged, and Microsoft and Meta both blamed higher component prices for part of their raised capital plans. Some of the headline growth in "AI investment" is simply the same equipment costing more, which is inflation, not additional capacity.
There is a quieter issue about the lifespan of the hardware. A large share of the spending goes to graphics processors and servers that become outdated quickly. One analysis noted that around two-thirds of Microsoft's capital spending in a recent quarter went to short-lived assets such as GPUs and CPUs. If a chip loses most of its economic value in three to five years, but is depreciated over a longer period, current profits may be flattered, and the true cost of the buildout will arrive later in the form of write-downs or replacement spending. Executives involved in the OpenAI-Nvidia talks have themselves acknowledged that predicting chip depreciation is difficult. A railway track can last a century. A data-centre accelerator may not last five years.
Electricity: the constraint nobody can code around
If chips are the first bottleneck, power is the second, and it may be the hardest to solve quickly.
The International Energy Agency's recent work on energy and AI sets out the picture. Electricity demand from data centres grew by 17 percent in 2025, more than five times faster than overall global electricity demand, which rose by about 3 percent. The agency projects that data-centre consumption will roughly double from around 485 terawatt-hours in 2025 to about 945 terawatt-hours by 2030, slightly more than Japan uses today, with AI-focused demand growing even faster. In the United States, data centres are expected to account for almost half of the growth in electricity demand through the end of the decade, and by 2030 the country is projected to use more electricity processing data than manufacturing aluminium, steel, cement and chemicals combined.
Yet the build is hitting walls. The IEA lists shortages of gas turbines, transformers and advanced chips, delays in connecting projects to the grid, and complex permitting as slowing data-centre development. Some projects face waits of years for power. Households in several regions are already worried that the cost of upgrading the grid will show up in their electricity bills, which makes this a political issue as much as a technical one. Communities have begun resisting new facilities over water use, noise and rising rates.
The electricity story cuts both ways in the bubble debate. On one hand, the sheer difficulty of getting power is a natural brake on overbuilding. Projects that cannot get connected cannot be filled with chips, which limits how fast excess capacity can accumulate. On the other hand, power infrastructure is expensive, slow and long-lived. If AI demand disappoints after utilities have committed capital to new plants and grid upgrades, someone is left holding stranded assets, and in regulated markets that someone is often the ordinary consumer.
Valuations: what the market is pricing in
All of this physical and financial activity is sitting on top of stock markets that have been rewarded handsomely for AI exposure. A small group of companies, with Nvidia, Microsoft, Alphabet, Amazon, Meta and a few others at the centre, now represents an unusually large share of the value of US equity indexes. The performance of the broader market, and therefore the pension funds, retirement accounts and sovereign wealth funds invested in it, depends on how those companies fare.
A heavy market weighting is not itself proof of a bubble. Some of these firms earn enormous profits, and some of the price rises have been backed by real earnings growth. This is a key difference from 1999, when many of the highest-flying companies had little or no revenue. Today's leaders are mostly profitable and cash-rich, even if their free cash flow is under pressure.
Still, the market is pricing in a very good outcome. Look at companies one step removed from the giants. Suppliers of electrical equipment, cooling systems, construction and power generation have seen their share prices soar, with some trading at more than a hundred times earnings. That kind of multiple only makes sense if spending continues growing for years. If hyperscalers pause, the suppliers' valuations can fall sharply, even if their businesses remain healthy.
Lessons from earlier booms
History does not repeat, but it does offer patterns, and two stand out.
The first is the railway mania of the 1840s in Britain. The technology was truly transformative, and railways did reshape the economy. Yet many of the companies that built the lines went bankrupt, because competition and overbuilding drove returns below the cost of capital. The country got the railways, and investors lost fortunes.
The second is the dot-com and telecom boom of the late 1990s. Telecoms firms laid vast amounts of fibre-optic cable on the assumption that internet traffic would keep doubling every few months. Traffic did grow enormously, just not fast enough to fill the capacity before the money ran out. Much of that fibre sat unused ("dark") for years and was later bought for pennies on the dollar, becoming the infrastructure that powered the next generation of internet companies. Again, the technology won while the early financiers lost.
This is why the question of who is borrowing and who is lending matters more than the question of whether valuations are high. And it is why the IMF's attention has turned from the technology to the financing.
What the IMF is actually saying
It helps to be precise about the IMF's position, because it has been paraphrased loosely in headlines.
In the October 2026 Global Financial Stability Report, the Fund identifies an AI-driven market correction as one of three "salient risks" to financial stability, alongside a sharp rise in sovereign bond yields and a major cyberattack. It describes the trio as plausible near-term catalysts for a sharp tightening of financial conditions. It warns policymakers against complacency and says the consequences of an AI correction could be global. The report notes that markets have remained resilient since April, supported by strong corporate profits and solid growth, and that large holdings in AI-related companies have helped offset other pressures.
The companion World Economic Outlook models a scenario in which the promised productivity gains fail to materialise. In the Fund's analysis, a steep drop in AI investment prompted by disappointment with productivity or profitability would trigger a sharp fall in asset prices. The direct effect of unwinding the investment would reduce global GDP by up to 0.4 percent and investment by up to 1.3 percent by 2028. Add the wealth effects of falling share prices and tighter financial conditions, and global output falls by about 0.6 percent. Further spillovers to government bond markets and reduced demand for US assets bring the total hit to around 1.1 percent of global GDP by 2028, including a drop of 1.8 percent in US output.
Two points about this deserve emphasis. First, the IMF does not treat this as its central forecast. It is a downside scenario, one of several, and the baseline remains reasonably optimistic. Second, even the worst case is not catastrophic by the standards of past crises. The Fund's deputy research director, Petya Koeva Brooks, said the steep drop in output in the worst-case AI scenario was not likely to push growth down to levels consistent with a global recession. That is a long way from 2008.
Still, a loss of more than one percent of global output is not small. It would represent a trillion dollars or more in forgone production, concentrated in the places most exposed to technology investment. And the Fund's own caution is that scenarios like this tend to understate the real world, where fear, forced selling and policy mistakes can make things worse than models suggest. Chief economist Silvana Tenreyro also noted that an AI downturn would reduce trade, adding a global transmission channel.
There is a deeper policy point as well. The Fund's earlier warnings, including those in its January update, were framed as a risk of a market correction if expectations about AI productivity and profitability are not met. Nine months later the warning is more prominent, the amounts invested are larger, and the economy depends on the investment more. That trajectory, rather than any single number, is what regulators are reacting to.
What would a correction actually look like?
It is worth sketching how an unwinding might unfold, not to predict it but to understand the mechanisms.
It could begin with a small disappointment: a large customer trimming orders, an AI lab failing to raise its next funding round, a quarter of cloud growth that comes in below expectations. Markets that have been priced for perfection react violently to modest misses. Share prices of chip and infrastructure suppliers drop sharply.
Falling prices damage the financing machine. Companies that were borrowing against rising equity values or against long-term contracts find lenders growing cautious. Some of the more leveraged players, such as specialised data-centre operators and neo-cloud providers that took on heavy debt to buy chips, face covenant problems. Contracts signed with weaker AI labs turn out to be worth less than expected. Banks and private credit funds, which have been lending against data-centre projects, begin to mark down assets.
Spending then slows. If capital expenditure by the big four were to fall even modestly from a base of $725 billion, the effect on suppliers, construction firms, utilities and the regions that host the facilities would be considerable. Because AI investment has been contributing materially to US growth, the slowdown would register in national accounts, not just in stock prices.
Finally, the wealth effect arrives. American households own a lot of equities, and consumer spending in the United States is closely tied to stock market wealth among higher earners. A significant fall in share prices cuts spending, which hits earnings elsewhere, which pushes prices lower. This is the feedback loop that the IMF's Athanasios Vamvakidis described when he said a correction would drag down broader asset prices and overall demand.
The failure of a bubble does not mean AI disappears. After the dot-com crash, the internet kept growing. The wreckage cleared, stronger firms bought cheap assets, and the technology continued to spread. A similar pattern is quite plausible for AI: a painful financial reset followed by continued adoption at lower prices, with the cheaper capacity becoming a gift to the next generation of users.
Why the revolution case is not foolish
Having spent most of this article on the risks, fairness requires making the case for the other side properly. The people who believe AI is a true economic revolution are not simply naive.
They point out that adoption is faster than for almost any previous technology. Hundreds of millions of people use AI tools weekly. Businesses across sectors are experimenting. Revenue at leading AI companies has grown at rates rarely seen before, and cloud providers say they cannot add capacity fast enough to meet demand. Microsoft has told investors it expects to stay constrained on capacity through 2026 even after spending huge sums. If demand is outrunning supply, the argument goes, then this is not like fibre lying dark in the ground.
They point to the underlying technical progress. Models have improved remarkably in reasoning, coding, scientific analysis and multi-step task completion. AI is being used in drug discovery, materials science, weather forecasting, protein design and mathematics in ways that would have sounded like science fiction a few years ago. If these systems continue improving, and especially if they can reliably carry out long chains of work autonomously, the economic implications could dwarf the current spending.
These arguments are serious. They may well prove right about the technology. But notice that none of them answers the financial question. A technology can be revolutionary and still be overfunded. Whether investors earn a good return depends on the price paid, not only on the importance of the invention. The relevant question is not "will AI matter?" but "will AI generate enough profit, soon enough, to justify what is being spent now?"
Who else is exposed, including the developing world
The conversation tends to focus on Silicon Valley, Wall Street and Beijing, but the effects of an AI boom or bust spread much wider.
For countries like Pakistan, the effects are mostly indirect but still real. The IMF's baseline growth forecasts depend on the world economy holding up, and the Fund is simultaneously worried about a prolonged conflict in the Middle East and the possible closure of the Strait of Hormuz. Higher oil prices already squeeze importing countries. If an AI correction were to coincide with an energy shock, with tighter global financial conditions and weaker trade, emerging economies that rely on external financing, remittances and export demand would feel the strain through several channels at once: more expensive borrowing, a stronger dollar, weaker demand for exports and possibly lower remittance inflows from Gulf and Western economies.
There is a more hopeful side. Countries with young, skilled workforces and strong IT sectors could benefit if AI tools lower the cost of building software and delivering services, allowing small firms to compete globally. Pakistan's freelancers and IT exporters are already using AI tools to raise output. But the same tools threaten the lower-end outsourcing work, such as basic coding, data entry and content production, on which many developing-country workers depend. The net effect will depend heavily on how quickly local education, infrastructure and policy adapt. Reliable electricity, affordable connectivity and investment in digital skills are not luxuries in this picture. They are the entry ticket.
What to watch in the coming months
Instead of trying to predict a date, it is more useful to track a few signals that would tell us which way things are heading.
- Capital spending guidance. The next round of earnings in late October and early November will show whether the biggest spenders keep raising their plans or start to hesitate. A pause from even one of them would be significant.
- AI revenue against spending. Watch whether cloud and AI revenue growth keeps pace, and whether companies start reporting AI profit in more detail rather than only run rates.
- Debt issuance and credit spreads. Rising borrowing costs for hyperscalers, data-centre operators and AI labs would signal that lenders are getting nervous.
- The financing health of the AI labs. Whether the leading developers can keep raising enormous sums, and whether their commitments to chip and cloud suppliers are honoured or renegotiated.
- Chip pricing and inventories. Falling prices for memory and accelerators, or growing inventories, would hint at oversupply.
- Power connections. How many announced data centres actually get energised, and on what timeline.
- Labour market data. Evidence on graduate hiring and sector employment will show whether AI is already changing work at scale.
- Macroeconomic context. Oil prices, the Strait of Hormuz, government bond yields and central bank policy all affect how well the economy could absorb an AI shock.
Revolution and bubble: probably both
So which is it, economic revolution or financial bubble?
The most defensible answer is that the question presents a false choice. Transformative technologies and speculative bubbles have gone hand in hand throughout history, because the same excitement that funds genuine innovation also overshoots. Railways, electricity, automobiles, radio, the internet: each changed the world, and each saw waves of investment destroy capital along the way. The pattern is so consistent that it deserves to be treated as the default expectation, not as a surprise.
If that pattern holds, the AI era will probably leave us with powerful and widely used tools, a lot of useful physical infrastructure, and a number of spectacular financial failures. Some companies will earn extraordinary profits. Others, perhaps including famous names, will find that they paid too much for capacity they could not monetise. The timing is the hardest part. Bubbles can inflate for far longer than sceptics expect, and anyone who tried to call the top in 1996 or 2021 learned that being right too early looks a lot like being wrong.
For policymakers, the lesson is that the time to prepare for a downturn is before it arrives. The IMF's warning is not a prediction of disaster. It is a reminder that an economy standing on one leg needs a plan for what happens if that leg buckles.
Perhaps the most honest summary is that we are running a very large, very fast experiment, with other people's money and a good deal of the world's economic momentum riding on the result. If the technology delivers the productivity gains its backers promise, the spending will look visionary. If it takes longer, costs more or earns less than hoped, it will look like folly. Most likely, it will be remembered as both, depending on whom you ask and when you ask them.
🌍 Browse All Science/Technology News →
- The AI Laboratory That Can Talk to Itself How Intelligent Machines Are Changing Scientific Discovery
Sources and further reading
- International Monetary Fund, World Economic Outlook and Global Financial Stability Report, October 2026, as reported by Reuters and other outlets covering the Bangkok annual meetings.
- International Energy Agency, "Energy and AI" and "Key Questions on Energy and AI" reports, 2025 and 2026.
- Company earnings reports and guidance from Microsoft, Alphabet, Amazon and Meta, first quarter 2026, as summarised by financial press.
- Analyses of hyperscaler capital spending, free cash flow and off-balance-sheet commitments published in spring 2026 by independent analysts and research houses, including CreditSights and Apollo.
- Reporting on circular financing among Nvidia, OpenAI, Oracle, AMD and CoreWeave, from Axios, Fortune, Bloomberg and others, 2025.
Figures for corporate spending and market values change rapidly and are rounded; readers should check the latest company filings. This article is for general information and analysis and does not constitute investment, financial or legal advice.
0 Comments