1. The AI Bubble Paradox: A Financial Crash, a Technological Revolution
What if the AI bubble bursts—and AI becomes even more economically powerful afterward? At first glance, the two outcomes seem contradictory. If companies have invested hundreds of billions of dollars in AI infrastructure and fail to earn adequate returns, surely the investment has been wasted. If valuations collapse, surely the technology has disappointed. But this reasoning confuses the financial performance of an investment with the economic usefulness of the asset it creates.
The internet did not become useless when the dot-com bubble burst. Many companies disappeared, investors suffered enormous losses, and business models that had looked revolutionary proved commercially unsustainable. Yet the networks, infrastructure and accumulated capabilities of the digital economy continued to support its expansion. The technology survived the excesses of the investment cycle.
AI could follow a similar path. Some companies may fail, some valuations may prove indefensible, and some data centres may never generate the returns their owners expected. None of these outcomes would establish that AI had exhausted its economic potential. A technology can succeed economically while many of the investors financing it fail financially.
2. The $60 Trillion Question: AI and the Service Economy
Much of the public discussion about the AI boom focuses on a relatively narrow group of companies: chip manufacturers, cloud providers, data-centre operators and AI laboratories. Their revenues and valuations are visible, making them natural objects of market attention. But they represent only one part of the economic story.
AI’s potential reach extends into the service economy, where people process information, write software, analyse financial data, conduct research, teach, design, advise, administer and make decisions. These activities account for a substantial share of economic output. Many are also exposed to the capabilities that modern AI systems are developing.
Consider a software company that once needed a large team to build and maintain a product. AI may allow a smaller team to accomplish more work. A financial institution may automate parts of its research and administration. A consulting firm may deliver analysis faster. An education business may use AI to personalise explanations and reduce the cost of producing learning materials.
The economic effect does not remain confined to the companies selling AI. It spreads through the businesses using it, the customers receiving cheaper services and the new activities that become commercially viable when production costs fall. AI’s economic footprint could be far larger than the AI industry’s own revenue.
The frequently discussed figure of roughly $60 trillion should be treated as a hypothesis about the economic activity potentially exposed to substantial AI-driven change, not as an established estimate of AI revenue or additional GDP. The precise number depends on what is being measured. The underlying question, however, remains important: how much of the global service economy could AI transform?
3. The 25% Hypothesis: How Much of World GDP Could AI Transform?
Suppose AI eventually influences economic activity equivalent to 25% of world GDP. That is a provocative proposition, but what would it actually mean? It would not necessarily mean that AI companies collect revenue equal to one-quarter of global output. Nor would it mean that AI creates an equivalent amount of additional GDP. A technology can transform existing economic activity without capturing all the value it helps produce.
A business might use AI to reduce costs, allowing it to lower prices and sell more. Another might maintain its prices and retain the savings as profit. A third might develop a new service that could not previously be delivered economically. Workers might become more productive, while some existing jobs and business models lose value. These outcomes can occur simultaneously.
The 25% hypothesis is therefore best understood as a proposition about the possible economic reach of AI. It asks whether AI could become deeply embedded in a large share of production, services and decision-making—not whether the companies building the technology will capture an equivalent share of global income.
General-purpose technologies often create value far beyond the industries that produce them. Electricity transformed factories, homes and transport without the electricity industry capturing the value of everything it enabled. Computing and the internet followed a similar pattern. AI could do the same across information-intensive work. Whether the eventual footprint reaches 25% is an open question: a hypothesis to investigate, not a settled forecast.
4. The Infrastructure Paradox: What Survives an AI Bust?
The financial risk is real. Companies may be building capacity ahead of demand, paying high prices for specialised hardware or assuming that today’s growth rates will continue indefinitely. Competition could compress margins, models could become cheaper to operate, and customers might prove unwilling to pay enough to justify the investment.
If expectations change, the correction could be severe. Shares could fall, funding could become more expensive, and companies could cancel or delay projects. Investors who financed infrastructure on optimistic assumptions could lose substantial amounts of money. But a financial loss does not automatically make the underlying infrastructure worthless.
A data centre consists of more than an AI business plan. It includes buildings, power connections, cooling systems, networking, security and computing equipment. Some of these assets can support other workloads; others are more specialised and may lose value rapidly as technology changes. Their future usefulness depends on technical compatibility, operating costs, electricity availability and actual customer demand.
The key question is not whether every AI investment will survive. It is whether enough of the resulting infrastructure remains useful after the investment assumptions that justified it have been revised. That distinction separates a technological dead end from an investment cycle that has simply overestimated the speed or profitability of adoption.
5. When Computing Becomes Cheap: The Second Life of AI Infrastructure
Imagine that AI companies build more computing capacity than customers are initially willing to pay for. Owners still face electricity bills, maintenance costs, financing obligations and hardware depreciation. They may decide that earning a smaller return is better than leaving usable capacity idle.
Surplus computing resources could then be offered to businesses, developers and researchers at lower prices. Depending on the hardware and facility, potential uses include cloud computing, AI inference, enterprise hosting, data processing, simulations and scientific workloads. Not every AI accelerator is equally suitable for every task, but computing infrastructure is not inherently limited to the original application that motivated its construction.
Lower prices could change the economics of software development. A small company might be able to run an application that was previously too expensive. Researchers could conduct more experiments. Businesses could automate tasks whose benefits did not justify the earlier cost of computing. New applications might emerge precisely because the infrastructure has become cheaper.
This produces an intriguing feedback loop. The investment boom creates excess capacity; the correction forces prices down; cheaper computing makes more applications viable; and those applications generate new demand. The financial bust would still hurt investors, lenders and workers caught in the downturn. But the economy could inherit an asset base that supports a wider phase of adoption.
The paradox is that an AI investment boom could fail on its original financial terms while helping make AI more accessible.
6. The Deflationary Engine: AI Makes Services Cheaper
AI’s economic significance may ultimately depend less on the size of its own industry than on how much it changes the cost of producing everything else. Suppose a company can deliver the same service with fewer hours of work, or produce twice as much output with the same team. Its cost per unit may fall. If competition forces the company to pass those savings to customers, prices can decline. If the company retains the savings, profits may rise instead. If lower prices stimulate demand, the industry may expand even while each unit becomes cheaper.
These are different routes through which productivity gains affect the economy. They do not all produce the same result, but each can change the relationship between output, costs, employment and prices. AI could exert these pressures across software, administration, research, design, customer support, financial analysis and other knowledge-intensive activities.
Yet productivity growth is not the same thing as economy-wide deflation. Prices also depend on wages, energy, housing, credit, demand and the degree of competition. AI may reduce the cost of producing one service while data-centre construction raises demand for electricity and specialised equipment. A more productive economy can still experience inflation in scarce resources.
The central possibility is that AI creates a persistent downward pressure on the cost of many digital and cognitive services. If those gains spread widely enough, they could influence the overall inflation environment and the decisions of central banks.
7. The Real-Estate Disruption: Less Space, More Productivity
Real estate is another part of the economy where AI could change the relationship between output and physical assets. Remote and hybrid work have already challenged the assumption that every employee needs to occupy a dedicated desk in a central office every working day. If AI allows smaller teams to perform more work, some businesses may require less office space for a given level of output. The effect could be particularly strong in markets where office supply is abundant and demand is already weakening.
Residential property could experience a different adjustment. Remote work may allow some households to live farther from major employment centres, shifting demand towards more affordable locations. At the same time, people working from home may prefer larger living spaces or a dedicated office. The result need not be a universal fall in housing prices; it could be a redistribution of demand between locations and property types.
AI may also improve productivity within the real-estate sector itself. Better design tools, construction planning, materials management, valuation, maintenance and property operations could reduce costs or improve the use of existing buildings. If these improvements expand supply or lower operating costs faster than demand grows, prices and rents could come under pressure in some markets.
However, land is not software. Planning restrictions, infrastructure, financing costs, location and scarcity continue to matter. Cheaper construction does not guarantee cheaper housing when land is limited or demand is strong. The larger hypothesis is that AI could reduce the amount of conventional office space required to generate economic output while improving the productivity of the property industry.
8. The Great Capital Reallocation: From Real Estate to Technology and Commodities
Capital is not static. Investors compare expected returns, risk, liquidity and the opportunities available across different assets. If AI changes the economics of real estate, it could also change where investors and businesses prefer to commit their resources.
Suppose a company discovers that it can operate effectively with fewer offices. It may reduce its property commitments and redirect some of the money towards computing infrastructure, automation, research or product development. An investor who expects weaker returns from a particular property market may look for opportunities elsewhere. New investment may flow towards technologies that promise productivity gains or industries supplying the infrastructure those technologies require.
This does not mean that every rupee or dollar leaving real estate will automatically enter technology or commodities. Property losses can destroy wealth, credit conditions can tighten, and investors may prefer cash or government bonds during a crisis. Capital reallocation is a possible outcome, not a mechanical law.
Still, AI could alter the relative attractiveness of different investments. Semiconductors, robotics, computing infrastructure and AI applications may benefit from the expansion of automation. Copper, electricity grids, energy resources and critical minerals may benefit from the physical requirements of that expansion. The technology that makes digital services cheaper can simultaneously increase demand for selected physical inputs.
The broader idea is not simply that capital moves from one asset class to another. It is that AI could change which forms of capital the economy needs, and how much of each form is required to produce a given level of output.
9. The Monetary Paradox: Can AI Give Central Banks More Room to Ease?
If AI reduces costs across a substantial part of the economy, it could change the environment in which central banks make decisions. Central banks generally respond to inflation, employment, economic activity and financial conditions—not to productivity gains in isolation. But if AI allows businesses to produce more at lower cost, it may reduce inflationary pressure relative to what would otherwise have occurred. If demand is weak at the same time, central banks may have more room to lower interest rates without pushing inflation above their objectives.
A financial downturn could strengthen the pressure to ease policy. If an AI-market correction causes businesses to cut investment, lenders to tighten credit or households to reduce spending, policymakers may respond to the resulting weakness. Lower interest rates or other support measures could help stabilise demand and financial conditions.
This is the connection between AI-driven productivity and monetary flexibility. A technology that expands productive capacity can, under the right conditions, make it easier for the economy to grow without generating the same inflationary pressure.
But the relationship is not automatic. AI infrastructure requires electricity, chips, construction and other resources that can become expensive. Productivity gains may increase profits without lowering consumer prices. A central bank facing persistent inflation or unstable inflation expectations may have little room to ease, even during a market crash. The more defensible proposition is that AI could alter the balance between productive capacity and nominal demand, changing the conditions under which monetary easing becomes appropriate.
10. When Deflation Meets Monetary Easing
There is a common assumption that falling prices and monetary expansion cannot coexist. In reality, the outcome depends on how quickly productive capacity grows relative to spending, credit and the money supply.
If AI makes services cheaper and increases the economy’s capacity to produce, prices may face downward pressure even while nominal demand continues to expand. In that situation, monetary easing could support spending without immediately generating the same inflation that would occur in an economy whose productive capacity was constrained.
But the opposite outcome is possible. If credit creation and demand grow faster than productive capacity, inflation can return. If AI investment pushes up energy and infrastructure costs, some prices may rise even as digital services become cheaper. If automation weakens employment and household incomes, demand may fall more sharply than production costs.
The economy could therefore experience different price movements across sectors rather than one uniform deflationary trend. The important question is not whether AI automatically permits money printing. It is whether AI-driven productivity changes the relationship between economic growth, prices and monetary policy.
11. The Commodity Counterforce: Digital Abundance, Physical Scarcity
AI may make certain forms of intelligence and digital production more abundant, but it does not eliminate the physical constraints of the economy. Data centres require land, electricity, cooling, networking equipment and computing hardware. Expanding power systems requires generation capacity, transmission lines and materials. Robotics and automation require machines, motors, electronics and industrial inputs. Even if AI reduces the amount of human labour required for a task, the infrastructure needed to perform that task still has a physical cost.
This creates a tension between digital abundance and physical scarcity. Software, analysis and some services may become cheaper to produce, while demand for selected resources increases. The resulting price movements will depend on whether supply can expand quickly enough to meet the new demand.
Commodities are not a single trade, either. A productivity boom may increase demand for copper and electricity infrastructure while reducing demand for other inputs. Technological improvements may also reduce the material required per unit of output or substitute one resource for another.
The broader hypothesis is that AI could reshape the composition of demand. Capital may become less dependent on some conventional assets while becoming more dependent on the physical systems that support computing, automation and industrial production. Digital abundance does not mean the end of scarcity. It may change where scarcity matters most.
12. The New Economic Cycle: A Bubble Bursts, but the Transformation Continues
Put these mechanisms together and a different picture of the AI boom emerges. AI could transform a substantial share of service-sector activity, increasing productivity and reducing the cost of many services. Excessive investment could nevertheless lead to a financial correction. Some investors could lose money while usable data-centre capacity survives. If that capacity becomes cheaper, new applications may become viable and adoption may spread further.
At the same time, remote work and AI-driven productivity could change demand for office space and alter the economics of real estate. Capital may be redirected towards technology, automation, commodities and productive infrastructure, although financial losses and risk aversion could interrupt that process. If productivity gains reduce inflationary pressure, central banks may have more room to support demand—but physical bottlenecks or excessive monetary expansion could produce the opposite result.
These outcomes are connected, but they are not guaranteed to occur together. The timing matters, the distribution of gains matters, and the response of businesses, investors and policymakers matters. A financial crash could delay adoption as well as make infrastructure cheaper. Deflationary pressure could coexist with inflation in particular resources. Capital could leave one asset class without flowing into the next.
Even with those qualifications, the central idea remains worth considering: the financial cycle and the technological cycle are not the same cycle. An investment boom can end because returns fail to meet expectations. A technology can continue advancing because its capabilities remain useful. A crash can destroy wealth while leaving infrastructure, knowledge and cheaper production methods behind. The economic transformation may then continue in a different form, led by businesses and applications that were not the original focus of the boom.
The most important question may therefore not be whether the AI bubble bursts. It may be what kind of economy emerges from the assets, capabilities and price changes that remain afterward. The next phase of AI could be less about a handful of spectacularly valued companies and more about the widespread use of cheap computing to change how the rest of the economy works.
A bubble can burst without ending a revolution. Sometimes the crash changes who can afford to participate in it.
This question also connects to the broader abundance paradox: when technology makes a capability cheaper and more accessible, the economic value of the surrounding system can change.
Related reading
The Abundance Paradox: When More Makes Us Value Less