Many technology sector analysts believe that the stock market price declines within the tech sector (and the overall market), that occurred in the aftermath of DeepSeek’s recent product releases represented an “over-reaction”. The most common argument made in favor of this “bullish” narrative is that computing efficiencies (in software and hardware) and associated cost reductions made possible by DeepSeek innovations will increase the demand for AI applications, and therefore increase the demand for the same set of AI inputs (e.g. computer chips, data centers, and cloud computing software) produced by the same companies.

Those pursuing this line of argumentation claim that an economic concept called the “Jevons Paradox” supports their bullish thesis. The Jevons Paradox refers to a microeconomic phenomenon whereby efficiency-enhancing technological innovations that lower the number of resource inputs required to produce a unit of output, “paradoxically” leading to an increase in the total demand for that resource that rises above and beyond the level that existed prior to the introduction of the efficiency-enhancing innovations. According to this line of argument promoted by bullish pundits, the more economical use of AI inputs enabled by DeepSeek will actually increase demand for those same inputs.

In this article, I am going to analyze whether this bullish conjecture is supported by the Jevons Paradox when analyzed in its proper historical context. My thesis is that Jevon Paradox and associated historical experience do not support a bullish thesis for AI-oriented US tech stocks and that it actually suggests very bearish implications.

The Jevons Paradox in Proper Historical Context

In 1865, William Stanley Jevons published The Coal Question: An Inquiry Concerning the Progress of the Nation and the Probable Exhaustion of Our Coal Mines . Jevons, who was one of the most important economists of the 19th century, wrote this book because he was deeply concerned about the potential depletion of Britain's coal reserves and the impact that this would have on the nation’s economic and geopolitical future. At the time, many in Britain were optimistic regarding the long-term sustainability of the nation’s coal supplies, largely because of technological advancements—such as the Watt steam engine—that had significantly reduced the amount of coal that was needed to produce a given amount of economic output.

The Jevons Effect: A Paradox of Efficiency

In Chapter VII, titled Of the Economy of Fuel , Jevons warned against complacency regarding technological improvements that reduced coal consumption per unit of economic output. He famously stated:

"It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth."

Jevons explained what has become known as the Jevons Paradox . Jevons argued that technological innovations that enabled less coal to be consumed per unit of output would increase the gross consumption of coal. Jevons explained that this somewhat counter-intuitive outcome will tend to occur because,

“The reduction of the consumption of coal, per unit of work, will enable us to do more work for the same amount of coal. This is the key to the paradox that the more economical the use of coal becomes, the more its consumption increases.”

Jevons summarized the phenomenon thusly:

“Whatever, therefore, conduces to greater efficiency in fuel consumption will accelerate rather than retard the exhaustion of coal mines.”

Historical Evidence Cited in Support of the Jevons Paradox

Several examples of the operation of the “efficiency paradox,” have been offered in support of the existence of the Jevons Paradox.

1. Steam engines. Newly designed Watt steam engines required roughly 10 pounds of coal per horsepower-hour compared to about 45 pounds per horsepower-hour for older Newcomen engines. Despite this enormous increase in efficiency, coal consumption in Great Britain increased from about 15 million tons in 1800 to about 100 million tons in 1865.
2. Iron production. Improvements in smelting technology, such as the use of coke instead of charcoal and the development of the hot blast furnace, made iron production cheaper and more efficient. Whereas in 1780, producing one ton of pig iron required 8 tons of coal, in 1830, the same amount of production required only 3 tons of coal. Despite using less coal per unit of production, the use of coal in the production of iron and steel production skyrocketed such that by 1865, iron and steel production was consuming roughly 30% of Britain’s coal output.
3. Railway transport. In the 1830s locomotives consumed roughly 80 pounds of coal per mile. By the mid-19th century, this had improved to roughly 35 pounds of coal per mile. Despite this fact, the use of coal for railway transportation increased by a factor of more than 100 during this time.
4. Steamships. In the 1830s, steamships consumed roughly 10 pounds of coal per mile. By 1860 this had been reduced to about 2.5 pounds of coal per mile. Despite this fourfold increase in efficiency, consumption of coal by steam-powered ships in Britain went from 500,00 tons to over 10,000,000 tons by 1865.

Jevons Paradox: A Microeconomic Law or A Myth?

While the Jevons Paradox presents an intriguing argument, and statistics such as those cited above are quite alluring, it is not at all clear whether and to what extent the Jevons Paradox is actually a real microeconomic phenomenon. It is certainly not a universally applicable law of microeconomics, nor it is a hypothesis that can be scientifically verified.

1. Contradicting Empirical Evidence: There are many observed instances in which greater efficiency does, in fact, lead to a decline in the overall consumption of a resource. The transition from incandescent bulbs to LED lighting led to diminished electricity consumption; efficiencies in refrigeration technology led to less demand for electricity consumption; automobile fuel efficiency has led to a major deacceleration of the growth of oil consumption. These are just a few examples where greater efficiencies in the use of a resource due to technological advances has resulted in lower amounts of resource consumption despite the increased production of the products that employ these resources as inputs. This directly contradicts the expected outcome of the Jevons Effect.
2. The fallacy of inferring causation from correlation: It is not possible to isolate how much (if any) of the increased consumption of coal during the 19th century was attributable to efficiency improvements. Economic growth, population expansion, and societal transformations all factors that contributed to increased resource consumption – likely far more so than the Jevons Effect.
3. Counterfactual Inference: It is impossible to know what the consumption of coal would have been if efficiency-enhancing innovations in the use of coal hadn’t been developed. One thing is for sure:  Due to population growth, economic development, societal changes and other factors, railway transport was going to grow regardless of whether energy efficiencies had been discovered. Indeed, when analyzing history, we can never know “what would have happened.”  It is actually possible that if the innovations that improved efficiencies in the use of coal had not been developed, other even more efficient fuels (i.e. petroleum-based) might have developed even sooner and economic history might have been completely different. For example, the use of coal as a fuel might have collapsed much sooner than actually occurred historically and the entire economic history of the world may have been completely different as different industries would have emerged at that time and geopolitical dynamics (due to sourcing of petroleum resources) would have been vastly different.

The Jevons Paradox in Contemporary Context

Notwithstanding these empirical and conceptual shortcomings, since it was created, the Jevons Paradox has been repeatedly employed as a foil to argue that technological developments that enable lesser quantities of inputs to be used in the production of a given unit of output, may actually lead to an increase in the total consumption of that input.

Historically, the Jevons Paradox has been most frequently employed in discussions about fuel consumption. For example, in recent times, some climate change activists have argued that measures aimed at improving fuel efficiency will not cause a decline in the consumption of fossil fuels nor help to reduce carbon-dioxide emissions, due to Jevons Paradox.

More recently, in the aftermath of recently announced efficiencies in computational resource usage and associated declines in the market values of several high-tech companies in the US  -- e.g. NVIDIA (NASDAQ: NVDA ), Microsoft (NASDAQ: MSFT ), Google (GOOG) (NASDAQ: GOOGL ) – several financial markets commentators have sought to employ the Jevons Paradox to argue that market participants were “over-reacting.”. They argue that despite the revolutionary computational efficiencies enabled by innovations introduced by DeepSeek, the consumption of inputs used in the production of AI applications will actually increase. In other words, even though AI applications using the DeepSeek LLM are expected to utilize 90%+ less computational resources (software and hardware), it is argued based on the Jevons Paradox that the consumption of computational resources (e.g. computer chips, data centers and cloud software) will increase.

Is the Jevons Paradox Relevant to AI Technology?: A Historical Perspective

In my next article, I am going to perform an in-depth analysis of whether the application of the Jevons Paradox to arguments about the profitability and valuations of certain US tech companies is even logically coherent. However, for the remainder of this article, I will only focus on the validity of the implicit historical analogy between coal as an energy input and the sorts of inputs that are utilized in the development of AI applications – e.g. computer chips, data centers and cloud computing software.

The key question is:  Do computer chips, data centers, and cloud computing services play a similar role in the value creation chain for AI applications that coal did for locomotives and steam ships in the 19th century?  If not, then the analogy breaks down and the Jevons Paradox must be considered to be of questionable relevance in the debate regarding the demand for products and services provided by companies in the US tech sector.

Superficial-minded tech analysts recently enamored with the Jevons Paradox, tend to misleadingly speak about the inputs consumed in the production of AI applications as if they were a singular resource and an undifferentiated commodity that can be analogously compared to coal that was used as a fuel in the 19th century. For example, in discussing the Jevons Paradox they carelessly use terms such as “GPUs” and “compute” as if they were a singular and undifferentiated commodity. This is a fundamental error. The inputs that generate AI (e.g. computer chips, data centers, and cloud computing software) are multiple and highly differentiated.

Furthermore, simple-minded tech analysts have failed to recognize the fact that the technological innovations introduced by DeepSeek are not merely enabling efficiencies in the use of a singular resource or a set of resources – it is enabling total and/or partial substitution of one set of inputs (and configurations of inputs) for another new set of inputs (and configurations).

This isa  critical distinction, because the historical technological innovations in engines (e.g. from Watt to Newcomen steam engines) merely enabled more efficient consumption of coal; they did not prompt the substitution of coal for another source of fuel.

The significance of this erroneous historical analogy being made by tech industry commentators can be illustrated with a historical hypothetical counterfactual example. Imagine that in 1865, technological innovations had caused a shift from coal-powered engines to more energy-efficient diesel-powered engines. Now imagine a stock market analyst at that time claiming that because of the fuel efficiencies made possible by diesel engines, the demand for coal was going to increase and coal mining companies were going to increase their profits. This would be absurd!  The companies that produced coal in the 19th century were (and still are) fundamentally different from the ones that produced and refined petroleum products. The switch from coal to diesel would have helped the new producers of crude oil and refined petroleum products and would have devasted the producers of coal.

This serves to illustrate the intellectual poverty of the argument that stock market analysts are presently making when they say that the profits and valuations of incumbent producers of inputs -- e.g. NVIDIA, Microsoft, Google and Oracle (NYSE: ORCL ) -- used in the production of AI applications (e.g. computer chips, data centers, and cloud software) will benefit from the efficiencies enabled by DeepSeek. The innovations enabled by DeepSeek will change the types and mix of inputs used in the development of AI applications. As will be discussed in my next article, the producers of the computer chips, data centers, and cloud software of today will be different from the producers of the key inputs in the post-DeepSeek world of AI applications development. As such the profits and valuations of many tech companies will be devasted.

Indeed, history has shown, time and time again, that major technological innovations rarely help the profitability or market valuations of incumbent firms. The forces of “creative destruction,” famously described by Joseph Schumpeter, tend to destroy the competitive position of incumbent firms and lead to the emergence of new leaders. Furthermore, history has shown that the “first movers” in a technological transition are rarely the ones that ultimately emerge as winners. For example, the first manufacturers of automobiles were not ultimately the winners in the automotive industry and the first manufacturers of airplanes were not ultimately the winners in the aviation industry.

Concluding Thoughts

In this article, I have demonstrated that the bullish narrative for US tech stocks that is based on the Jevons Paradox is premised on a false historical analogy. When this historical analogy subjected to careful scrutiny, it completely breaks down. In fact, the historical analogy between coal producers of the 19th century and today’s tech companies that produce AI inputs suggests quite the opposite conclusion:  Innovations enabled by DeepSeek (and soon others) will be extremely bearish for the profitability of many incumbent US AI tech companies.

Nobody should get the impression that I am bearish on AI, nor “pessimistic” about future economic developments just because the Jevons Paradox cannot be used to support conjectures about the profitability or valuations of US AI tech companies. To the contrary, I believe that the sorts of innovations introduced by DeepSeek (which will be exponentially enhanced by many others) will be extremely bullish for consumers and the economy as a whole. The decimation of the business models of many incumbent tech companies that I have described in this essay are simply classic examples of Schumpeterian “creative destruction”. I fully expect that the overall impacts of AI innovations on the economy will be very positive, but the effects on many specific companies will be bearish.

We are extremely bullish on the transformational power of AI in the global economy. Indeed, we are highly focused on investing in companies – most of which are not in the tech sector – that we believe will greatly benefit from the AI revolution.

Furthermore, we believe that developments in AI at the microeconomic level will soon have massive impacts on a macroeconomic level, and our portfolios will be positioned for the associated macroeconomic and geopolitical shifts.