.png)
Perspectives
Short, practical observations on venture capital, startups, decision-making, and the Indian ecosystem.
Is AI a Bubble? Investment, Returns and Lessons from History
Almost every bubble follows the same path
A transformative technology emerges, its benefits become visible, and early success attracts capital. Optimism turns into speculation, and spending and valuations race ahead of the technology’s ability to deliver economic value. Eventually, reality catches up, weaker players disappear, and valuations correct. The technology, however, does not disappear. The opportunity is real; the mistake is assuming that its benefits will arrive as quickly or be as large as investors expect.
Lets look at the investment and return maths for artificial intelligence
To understand the scale of the AI investment boom, I have tried to arrive at a reasonable estimate of the cumulative capital being deployed globally. There is no single, comprehensive measure of global AI investment, so I have constructed a working estimate using available industry data.
Stanford's “AI Index Report 2025” puts global corporate AI investment at $252.3 billion in 2024 alone, including private investment and mergers and acquisitions. IDC separately estimates global AI infrastructure spending at $153 billion in 2024 and $318 billion in 2025. These measures overlap and cannot simply be added, but they establish the scale of the investment already taking place. Allowing for earlier infrastructure spending, AI research and model development, and enterprise software and implementation, I am using $1.5 trillion as an indicative estimate of cumulative global AI-related expenditure through 2025. This is my own analytical estimate, not a published industry total.
For the period through 2030, I assume a further $5 trillion of investment, informed by projections such as McKinsey's “The Cost of Compute: A $7 Trillion Dollar Race to Scale Data Centers” and other industry estimates.
The resulting $6.5 trillion is the working baseline for the analysis that follows.
Can the returns from AI justify the investment?
Let us start with the size of the global economy. Global GDP is approximately $118 trillion. The economic benefits of AI will be spread across the major economies, particularly North America, Europe, China, India and the wider Asia-Pacific region.
The table below gives us a sense of the economic base across which these gains could be realised.
Source: World Bank national GDP data, rounded estimates.*
Now let us return to our assumed $6.5 trillion of cumulative AI investment through 2030.
If we spread the $6.5 trillion investment over seven years, we need approximately $930 billion of additional economic value every year to match the investment on a simple cumulative basis. It does not matter where in the world this value is created. Against the combined GDP of approximately $97 trillion across the major economies considered here, this is roughly 1% of GDP. The question is whether AI can generate this additional economic value consistently over the period. The major economies we have considered grew at approximately 3% annually in real terms over the last three years; if this accelerates to 4% over the next seven years, a simple back-of-the-envelope calculation suggests that the $6.5 trillion investment in AI may be justified, particularly given McKinsey’s estimate in The Economic Potential of Generative AI that generative AI could create $2.6–4.4 trillion in annual economic value.
But achieving this additional growth is not going to be straightforward. AI will create efficiencies, but it will also disrupt existing businesses, displace workers and make some products and services obsolete. Companies may save costs without generating equivalent new economic activity. Some industries may contract before new ones become large enough to replace them. Governments may also introduce regulations around employment, data, safety and the use of AI, affecting both the pace of adoption and the economics of deployment.
Lets look at this also from the perspective of earlier technological revolutions. Our quick research shows that incase of electricity and internet, the global American GDP (for which the data is available) did grow in this range “once the technology was sufficiently diffused”. So thats the risk - it may take time for the gains to appear in economic output. This is the central uncertainty in the AI investment debate. The technology may work, adoption may spread, and productivity may improve, yet the resulting economic gains may still fall short of what investors have already committed capital to achieve - just simply the timeline may not match. But at the moment, it is difficult to say either ways.
For a closer look at India's position, see Where India Actually Stands in the Global Artificial Intelligence Race.
Then there are warning signs on the financing side of AI investments.
1. AI Valuations may be running ahead of returns
The IMF's April 2026 Global Financial Stability Report identifies stretched equity valuations and increasing market concentration as risks, particularly given the importance of a small number of large technology companies to market performance. It warns that if these companies fail to generate the returns expected by investors, the consequences could extend beyond the AI sector.
This is a warning sign, but not conclusive evidence of a bubble. The largest AI investors have substantial existing businesses, revenues and profits. The question is whether the additional earnings expected from AI justify the valuations investors are paying today. We look at what these giant funding rounds mean for venture capital itself in Are Mega AI Raises Still Venture Capital?
2. Circular financing is becoming a concern
AI companies, chipmakers and cloud providers increasingly act as customers, investors and financiers of one another. The IMF identifies this growing interconnectedness as a potential source of financial risk. Such arrangements can support genuine commercial expansion, but they can also make it difficult to distinguish independently generated demand from demand sustained by investment within the same ecosystem.
If one major participant reduces its spending, the consequences could spread through suppliers, customers and investors. This is a risk worth monitoring, although circular financing alone does not establish that the underlying demand is artificial.
3. Is the industry building more capacity faster than it needs?
The third warning sign is the scale of planned infrastructure. Data centres, chips and power infrastructure are being built in anticipation of future demand. Yet the returns depend on how much of that capacity is used, what customers will pay for it, and how quickly the technology becomes obsolete.
The IMF notes that utilisation rates remain uneven across markets and identifies rapid technological obsolescence, tenant concentration and speculative development timing as vulnerabilities.
What can history tell us?
History shows that financial bubbles often follow a recognisable six-stage pattern. A genuine opportunity emerges, early success attracts capital, optimism becomes speculation, easy money amplifies the boom, reality catches up, and the survivors eventually create lasting value. The railway boom, the dot-com bubble and the telecom infrastructure boom all illustrate how a technology can be transformative while still producing poor returns for investors who commit too much capital at excessive valuations.
The question is how far AI has progressed along this path.
Stage 1 - A real opportunity emerges.
AI clearly qualifies. It represents a major technological shift with the potential to change how businesses operate, how knowledge work is performed and how industries create value.
Stage 2 - Early success attracts capital.
This stage is also firmly established. AI models are being adopted, companies are integrating AI into products and workflows, and early winners have attracted substantial investment. The opportunity is no longer merely theoretical.
Stage 3 - Optimism becomes speculation.
This is where the questions become more serious. Investors are extrapolating rapid adoption and future earnings, companies are committing enormous sums to infrastructure, and expectations about the scale of the eventual opportunity remain very high. The risk is that investment and valuations begin to run ahead of the economic value that can realistically be created.
Stage 4 - Easy money amplifies the boom.
This is the critical next stage to watch. Historically, abundant capital, leverage and financing arrangements helped push asset prices beyond what future earnings could justify. AI already has some warning signs, including circular financing arrangements and enormous infrastructure commitments. However, these alone do not establish that the entire sector has entered this stage.
Stage 5 - Reality catches up.
This would become evident if growth disappoints, financing tightens, customers fail to generate sufficient returns from AI, or infrastructure capacity exceeds demand. We have risks to monitor, but it would be premature to conclude that a sector-wide reckoning has arrived.
Stage 6 - The survivors create lasting value.
This is the likely long-term test regardless of whether a bubble develops. Some businesses may disappear, some investors may lose money, and others may build enduring businesses around the technology. As with railways and the internet, the ultimate value of AI will not guarantee attractive returns for every investor.
So, where does AI stand today?
If AI is a financial bubble, my assessment is that it is around Stage 3, with the possibility of moving towards Stage 4 but certainly not yet demonstrably at Stage 4.
The distinction matters. We can already see extraordinary expectations, large capital commitments and some speculative behaviour. What remains uncertain is whether these developments will translate into sufficient economic returns in the next few years, or whether easy capital and rising expectations will push investment materially beyond what the market can sustain.
The next 12 months should provide some important clues. Watch whether AI revenues and productivity gains begin to catch up with capital expenditure, whether financing becomes more dependent on debt and circular arrangements, and whether demand justifies the capacity being built. These will tell us more about whether AI is progressing towards Stage 4 or whether the investment boom can be sustained by real economic value.
Frameworks like the six-stage bubble pattern are how professional investors separate hype from value. The VC Academy program teaches that kind of structured thinking, with real case studies and deal room simulations.
Sources
Stanford Institute for Human-Centered AI, AI Index Report 2025, Economy chapter.
https://hai.stanford.edu/ai-index/2025-ai-index-report/economy
IDC, AI Infrastructure Spending Caps Historic Year at ~$90 Billion in Q4 2025; 2029 Spending to Eclipse $1 Trillion.
McKinsey & Company, The cost of compute: A $7 trillion race to scale data centers, McKinsey Quarterly, April 2025.
McKinsey & Company, The economic potential of generative AI: The next productivity frontier, June 2023.
International Monetary Fund, Global Financial Stability Report, April 2026.
World Bank, national GDP data (for the GDP table).