Part 2 of a 3 Part Research Series
What If Artificial Intelligence Works, But The Economics Don’t?
In the first article of this series, I explored what I believe is the most underappreciated risk facing the artificial intelligence industry. The debate surrounding AI has become almost entirely focused on capability. Every new model is compared to its predecessor. Benchmarks are scrutinised. Reasoning abilities are debated. Predictions about superintelligence dominate headlines. Yet capability and economic value are not the same thing.
The central argument of Part 1 was not that artificial intelligence will fail. Quite the opposite. AI is already proving itself to be an extraordinarily useful technology. The more important question is whether reliability constraints prevent AI from fully replacing human knowledge workers. If hallucinations remain a persistent feature rather than a temporary bug, humans may remain embedded in the decision-making process far longer than investors currently expect. That conclusion leads naturally to a second question.
What if artificial intelligence succeeds, but not to the extent currently assumed by financial markets?
This possibility deserves serious consideration because much of the capital currently flowing into the sector appears to be based on assumptions that extend well beyond the technology itself. Investors are not merely funding artificial intelligence. They are funding a particular economic outcome. They are funding a future in which AI dramatically reduces labour costs, transforms productivity and generates returns large enough to justify one of the greatest infrastructure build-outs in modern history.
The challenge is that the economics become far less certain if AI remains an assistant rather than a replacement. The distinction may seem academic, but it sits at the centre of the entire investment case.
Imagine a lawyer who becomes thirty percent more productive through the use of AI. That is clearly valuable. The law firm benefits. The client benefits. The lawyer benefits. Yet the economic value created is fundamentally different from a world in which the lawyer is no longer required at all. The first scenario creates efficiency. The second creates labour displacement. Current market valuations increasingly appear to assume the second outcome while the evidence increasingly points toward the first.
The consequence is that the industry’s future revenue opportunity may be considerably smaller than investors currently expect. This becomes particularly important when examining how AI is priced today. Most users of artificial intelligence have become accustomed to subscription models that appear remarkably inexpensive relative to the capability on offer. For a modest monthly fee, users gain access to systems capable of performing tasks that would have required hours of human effort only a few years ago. The value proposition feels extraordinary. The question is whether these prices reflect economic reality.
Few users stop to consider what their usage actually costs. A heavy user may generate thousands of interactions each month, consume significant computing resources and benefit from infrastructure costing billions of dollars to build and maintain. If providers attempted to charge users the true economic cost of their consumption, behaviour would change dramatically. The reason is simple. Human beings naturally evaluate value relative to price.
A business owner who pays twenty dollars per month for AI uses it freely. The same business owner faced with a two-hundred-dollar monthly bill begins asking different questions. Is this task worth the cost? Will this query generate a measurable return? Can this process be completed more cheaply elsewhere? The psychology changes immediately.
Demand that appears almost limitless at subsidised prices often becomes far more selective when exposed to real economic costs. This issue is particularly relevant in enterprise environments, where spending decisions are ultimately determined by return on investment.
One of the most revealing developments of the past year came from reports that large corporations were beginning to scrutinise AI spending more closely. Early adoption was driven largely by fear of missing out. Executives did not want to be perceived as ignoring a transformative technology. Budgets were approved. Pilot projects were launched. Tokens were consumed at astonishing rates. Eventually, however, businesses ask the question that every business asks.
What are we getting in return? The answer is not always obvious.
Many AI deployments produce genuine productivity improvements, but measuring those improvements can be surprisingly difficult. If employees save time but continue performing the same jobs, where exactly is the financial return? If AI assists decision-making but requires human verification, how much value has actually been created? If a company spends millions of dollars on AI tools yet struggles to identify a measurable impact on profitability, how long will that spending continue?
These questions are becoming increasingly important because the industry’s cost structure remains enormous.
Artificial intelligence differs from many previous software revolutions in a critical respect. Traditional software becomes more profitable as it scales. AI requires continuous investment in computing infrastructure. Every query consumes resources. Every model upgrade requires additional training. Every increase in demand requires additional capacity.
This creates a dynamic that should be familiar to students of financial history. The greatest investment bubbles rarely emerge because demand does not exist. They emerge because capital investment grows faster than economically sustainable demand.
The railway boom of the nineteenth century was built upon a genuine technological revolution. Railways transformed commerce and transportation. Investors correctly identified the importance of the technology. What they failed to appreciate was the difference between a useful technology and a profitable investment.
The same pattern emerged during the telecommunications boom of the late 1990s. Internet traffic grew exactly as anticipated. The problem was that too much capital was deployed chasing the opportunity. Infrastructure expanded more rapidly than sustainable profits.
The internet changed the world. Many investors still lost money. Artificial intelligence may ultimately face a similar challenge.
One of the most concerning aspects of the current cycle is the increasing evidence of circular capital flows throughout the industry. Technology companies invest in AI startups. Those startups purchase computing infrastructure from the same technology companies. Cloud providers invest in AI firms that subsequently become major customers of cloud services. GPU manufacturers invest in businesses that purchase GPUs.
None of this implies wrongdoing. However, it does raise important questions regarding the true source of demand. How much of the current expansion is being driven by sustainable economic activity and how much is being driven by capital chasing growth?
History suggests that these distinctions become painfully important when financing conditions tighten. The situation becomes even more complicated when viewed through the lens of private markets.
Private equity, private credit and infrastructure funds have become increasingly involved in financing the AI build-out. Data centres require capital. Power generation requires capital. Networking infrastructure requires capital. Much of this financing has occurred under assumptions of continued growth, abundant liquidity and favourable refinancing conditions.
Those assumptions may prove correct. But if they do not, the consequences extend far beyond the technology sector. An industry can survive a period of disappointing growth. A heavily leveraged capital structure often cannot.
This is why the current debate should not be framed as a question of whether AI succeeds or fails. The more relevant question is whether the economic value ultimately created by artificial intelligence justifies the extraordinary amount of capital currently being committed to the sector. Those are very different questions.
AI may improve productivity. It may become embedded throughout the global economy. It may remain one of the most important technological developments of our lifetime. Yet it is entirely possible for all of those things to be true while investors discover that they dramatically overestimated the economic returns available from the technology.
History suggests this outcome is not unusual. In fact, it may be the most common outcome of all.
The next article in this series will explore what happens if this process unfolds. If AI adoption continues but returns on investment disappoint, what are the implications for financial markets, monetary policy and, ultimately, Bitcoin?
