Technological Change Under Capitalism: Conversional Dynamics and Catalytic Spectralization

Or...Why Technological Innovation Will Not Save Us

In my new book, The Physics of Capitalism, I focus on the concept of energy conversion as the critical element for understanding the energy scale dynamics of modern capitalism. Understanding these scale dynamics is important if we want to construct a stable global civilization that operates within the ecological constraints of the planetary biosphere (see figure 1 below).

The fundamental problem we face is that modern civilization is too energy-intensive and inefficient. We consume vast amounts of energy and resources from the natural world, we convert some of that energy for useful tasks like mechanical work and electricity, and then we dump the vast majority of the energy we consumed back to the natural world in highly useless and degraded forms. We're dumping so much of this useless energy into nature that even the best natural sinks are having a hard time absorbing and assimilating all the junk we're tossing, leading to greater levels of chaos and thermodynamic instability throughout the biosphere, in the forms of rapid global warming, mass extinctions, ocean acidification, dead zones, severe droughts, epidemics, and other dangerous phenomena. If we continue down this path, we risk triggering major ecological tipping points that could cause massive, sudden, and irreversible changes throughout large parts of the biosphere, potentially threatening the survival of global civilization itself.

Figure 1: Wikipedia image showing the current status of several major planetary boundaries, as measured and assessed in this 2023 paper.

A common assumption among many people is that technological innovation, chiefly in the form of greater energy efficiency, can overcome just about any material or ecological challenge that humanity faces. That's the view that I put under pressure in my book. It's well-known in the ecological sciences that higher energy efficiency tends to produce larger energy systems, an observation known as the Jevons Paradox, often called the "backfire effect" by many economists. The basic idea is that when technologies and processes become more efficient, they also tend to become cheaper, which drives up demand for them and therefore leads to economies of scale and greater levels of consumption.

There are many examples of this process in modern times. When steam engines became much more efficient in the 19th century, they consumed more coal in total, not less. When the Bessemer process was invented for making steel, the world produced a lot more steel, not less. When the Haber-Bosch process was invented for making ammonia, the world produced much more ammonia, not less. When the Bayer process was created for converting bauxite into alumina, the world started producing much more aluminum, not less. This paradoxical trajectory in energy use happens in large part because efficiency improvements under capitalism, as I explained in my book, are heavily driven by the need to expand exergy capacity, where exergy is the maximum amount of energy available for useful work in any given process. What that means is that when many thermodynamic devices become more efficient, they also become more energy-intensive. A classic example is the transition from low-pressure steam engines to high-pressure steam engines in the Industrial Revolution, which made many steam engines both more efficient but also more powerful, capable of producing large quantities of useful work that allowed them to move big trains, steamboats, factory line shafts, and other devices and systems.

Many neoclassical economists have argued, in papers like this one, that severe macroeconomic rebound effects are unlikely to occur, much less backfire effects. But it's worth emphasizing just how fundamentally flawed and useless these studies are. First, their results are highly sensitive to time effects. If you watch a policy play out for 10 years, it may produce energy savings, but then it may backfire 20 years later, and your short observation window wouldn't be able to capture that effect. Second, these results are highly sensitive to boundary effects. A particular policy may produce energy savings for a specific city or country, but that same policy may shift more energy use to other cities and countries, a kind of geographic substitution that's been a major reason why many Western nations have become "cleaner" and more efficient. Third, these studies are also sensitive to the demand patterns they assume about the economy. If consumer or industrial demand preferences shift, as they often do, then energy savings in one period may not be predictive at all for what might happen in later periods. And fourth, many of the datasets used in energy and emissions accounting are so flawed and so thoroughly plagued by uncertainties, manipulations, and statistical gimmicks that it's practically impossible to believe the bulk of the results cited in the decoupling literature. This last one is a subject that I prominently cover in my book but won't go into any further details here. Among scientists who have seriously studied the issue, the consensus is that the backfire effect is very much real and happens often for many technologies, even if it may not happen for all of them (see The Economic Superorganism by Carey King for a prominent treatment of the subject).

Anyway, these are all issues that have been well-studied and debated in the academic literature. But in my book, I analyze the technological dynamics of modern capitalism from a completely new perspective, and in the process I tease out some new conclusions about why we should be skeptical of the notion that technological change can always save us. That's the main purpose of this article, so let's get into it.

The Conversional Network

An energy conversion happens when energy changes from one form to another. Think of a car engine burning gasoline to generate mechanical energy. The chemical energy stored in the gasoline is converted, at least partially, into the mechanical work of the tires, which is what allows the car to keep moving. Think about a photovoltaic cell receiving sunlight and converting that into electricity or a steam turbine converting the heat energy of the steam into electricity by pushing around and rotating a giant magnet. Energy conversions are the biophysical heart of everything we do. Nothing we care about would exist without the devices and technologies that execute these conversions, certainly not our civilizations. Examples of conversional devices include cars, computers, smartphones, robots, steam turbines, and photovoltaic cells. The human body is also a conversional system, although it would be a category mistake to classify it as a technology, since it evolved independently of modern technologies and technologies are ultimately built and developed by human labor itself. It’s important to emphasize that even systems or devices that don’t convert any energy themselves, like roads and railways, are still partly built and maintained by conversional devices, like excavators and bulldozers.

To see how energy conversions relate to the scale and growth of an economy, I introduce the metaphor of a network, which I call the conversional network, or coronet for short. As I write in the book, coronets are “biophysical pathways through which energy is changed or converted into different forms.” Coronets can be conceptualized at different scales of analysis. Power plants, cities, human brains, and even entire economies can be thought of as coronets, because they all process and convert energy flows through highly structured energy pathways, like power grids. At an aggregate economy-wide level, coronets constitute all conversional devices that are mutually interacting or functionally interdependent, with the conversional technologies acting, in effect, like the nodes of the network interacting together through dynamic feedback loops.

Conversional devices are embedded in their coronets and are therefore entangled with upstream and downstream nodes. This entanglement means that conversional devices are dependent on other conversional devices to continue operating or functioning as expected. For example, before natural gas can make it to a power plant to drive a turbine, it has to be routinely pressurized at compressor stations along the way. For another example, when cars run out of gasoline, they go back to the gas station to fill up, but they can only do so because tanker trucks have transported the gasoline to the stations in the first place, and the tankers can only transport the gasoline after fractional distillation at a refinery has extracted the gasoline from crude oil, and that can only happen because a bunch of other conversional devices have brought the oil to the refinery, and on it goes. When you use your computer to search something on Google and it brings up the AI Overview, your search is first routed to a nearby data center, it triggers inferencing from a collection of AI chips, the output is relayed to other parts of the data center by various devices, and finally the answer is sent back to your computer. There’s a whole spectrum of devices and technologies that are entangled and interacting together to make it all happen. Because conversional devices are entangled together in their wider coronets, changes to certain devices can precipitate changes in other devices interacting in the network. And this brings us to one of the fundamental ideas in the book: spectralization.

What is spectralization?

I define spectralization as any change to the conversional methods of a conversional device that has a notable effect on the efficiency or power output of that device. Let’s go through a few examples to make sure we understand what this means. An obvious example of a spectralization is the switch from gas-powered cars to electric vehicles, since it’s a change to the underlying conversional methods of the vehicle. In this particular case, the spectralization of the vehicle involves changing the input fuel (from burning gasoline to using electricity), changing the operational design of the conversional components (using electric motors instead of heat engines to generate mechanical work), and changing the waste outputs (from tailpipe emissions like greenhouse gases to the absence of emissions in electric vehicles). But spectralization can come in many different forms. When James Watt added his famous condenser to the steam engine, he did not change the input fuel but he changed something important about the operational design of his device, which greatly increased its thermodynamic efficiency. When Richard Trevithick transitioned the steam engine from using low-pressure steam to high-pressure steam, he likewise changed the operational design of his devices but not the input fuels. For a final example, consider the emergence of catalytic converters. These devices clean vehicle tailpipe emissions by breaking down and removing harmful gases, but they don’t affect the operational design of the internal combustion engine itself. They merely affect the waste outputs produced by the engine.

Not all changes, even those made to conversional devices, are spectralizations. If a carmaker adds leather seats to a car, that change may affect the financial value of the car, but it doesn’t involve any change to the methods of energy conversion used by the car. If a furniture maker produces a new kind of table, that change is not a spectralization because tables are not conversional devices. Likewise, expanding a highway from four lanes to six lanes is not a spectralization, as highways are not conversional devices. On a side note, the reason why I use the term spectralization is because I see conversional technologies as changing along a cumulative spectrum, with more recent parts and components getting integrated with older components to produce new devices. Think of the internal combustion engine and some of its fundamental components: pistons, crankshafts, and flywheels. All of these components individually were invented long before the internal combustion engine itself, but in the late nineteenth century they were refined and combined in a different way to produce an entirely new kind of device. That’s the basic essence of spectralization. For a more modern example, when NVIDIA unveiled the Blackwell AI chip in 2024, it basically took two GPUs and stuck them together, connecting them with a special link that allowed the two GPUs to function more or less as one unit. But it's the same basic idea as what was done with the internal combustion engine: take stuff that's already around and combine it in different ways. Take existing conversional technologies and spectralize them in new directions.

Spectralization should be distinguished from the more familiar concept of diffusion. Diffusion is generally understood to be about the spread of new or existing technologies, and these technologies or systems don’t have to be conversional in nature, whereas spectralization is specifically about the initial emergence of conversional methods and conversional technologies. In effect, spectralizations are meant to be understood as device-level inventions in the conversional methods of conversional technologies.

Catalytic Spectralization

Recall that, in our civilizational coronets, conversional devices are fundamentally entangled together: the functions and operations of any conversional device are dependent on the functions and operations of other conversional devices. Computers and refrigerators need electricity to continue working, and they ultimately get all that electricity from power plants. But power plants themselves need various energy sources to produce the electricity in the first place, and those sources have to be delivered, at least in the case of the fossil fuel industry, through vehicles, machines, and other conversional systems. Even for renewables, the input energy sources come directly from nature (like sunlight and wind) but the conversional devices themselves still need to be constructed, transported, and installed, and all of that work requires other conversional devices. This multilevel entanglement across our economic coronets implies that when we spectralize a given conversional technology, when we make changes to its conversional methods, there will be downstream and upstream impacts to other conversional devices in the broader coronet. In other words, spectralizations can induce other spectralizations.

In the history of capitalism, certain conversional devices arise that become so important for the broader society that their spectralizations can then spectralize many other devices across the economy. These devices act, in effect, like catalysts that drive major changes in other conversional devices and economic sectors. I call this process of technological change under capitalism catalytic spectralization. In the book, I talk about the steam engine as being the dominant catalyst of the Industrial Revolution, explaining how spectralizations in steam engines produced even more advanced steam engines that could be used in new ways, and how those technological developments changed other industries across the British economy. The spectralization and diffusion of steam engines throughout the British economy intensified the development and refinement of numerous major technologies and mechanical systems, such as trains, steamboats, power looms, spinning machines, line shafts, steam turbines, internal combustion engines, and so much more.

Another example of a major catalyst, more relevant to our current times, is the AI computer chip, especially NVIDIA’s GPUs. These ultra-fast chips have been used to train and implement all the major large language models we’ve heard so much about, such as ChatGPT, DeepSeek, Copilot, Grok, Llama, and others. The spectralization of NVIDIA’s AI GPUs have been the dominant catalyst for the development and modification of major data center technologies in the current AI revolution, affecting the development and diffusion of everything from memory chips and network switches to liquid-powered cooling equipment, aeroderivative gas turbines, and small modular fast neutron reactors. Here’s a high-level overview of how catalytic spectralization works in the current AI age. As NVIDIA’s chips become faster and more powerful, the on-chip memory is no longer enough, so a company like Micron now builds dedicated high-bandwidth memory chips to help NVIDIA’s GPU process more data. NVIDIA’s latest-generation Blackwell architecture is also more likely to overheat than the Hopper chips, so data centers are moving more and more towards liquid-powered cooling equipment, with plates and pipes extending across the servers. As AI chips get faster, the network switches also need to be updated so they can relay more data across the data center at faster rates. And the incessant demand for more electricity to ensure smooth data center operations is leading to a proliferation of small, modular, and flexible power generation units, like modular nuclear reactors and aeroderivative gas turbines. The spectralization and diffusion of increasingly powerful AI chips is changing the entire technological environment and technosocial structure of modern civilization, or at least of American civilization, prompting some pundits to declare that the energy transition is over.

Catalytic Spectralization and Energy Dynamics

We’re now in a good place to understand why catalytic spectralization expands the scale of energy systems under capitalism. At a conceptual level, when we make changes to major catalysts like AI chips and steam engines, those changes affect the surrounding coronet because of the entangled interdependence of conversional devices. New conversional technologies can only emerge and proliferate throughout the economy because they are mass produced, and that mass production might require developing new production methods and techniques. The new technologies have to be transported or distributed. Perhaps different fuel inputs also need to be used for the new conversional devices, in which case those inputs might need to be extracted, transported, refined, and distributed. And once the new technologies start working as intended, other surrounding technologies in their operational environment might need to change so they can be better adapted to interact with the new technology. Doing all of this takes a lot of energy, and if there are no downscaling effects from other sectors or technologies, it will lead to an expansion in total exergy capacity and energy consumption.

Catalytic spectralization has another curious and critical impact on the aggregate efficiency dynamics of our economic systems. Because the rate of spectralization under capitalism is so high, new conversional technologies are introduced quite often. These new technologies are by necessity inefficient relative to many existing technological forms, and as more of them diffuse across the economy, they amplify the energy scale of the economic system, in the absence of countervailing constraints. That’s because the evolution of all conversional technologies follows a general sigmoid-shaped efficiency curve, which means conversional devices start off as highly inefficient in their initial forms, then later versions of the technology are refined and improved, significantly boosting their efficiency (see figure 2 below). But at some point, the efficiency curve reaches a plateau and no further improvements are made. This can happen because of natural and physical limits on further efficiency gains or because of social, political, and economic factors that no longer make it profitable or realistic to keep developing the technology.

Figure 2: The general pattern of efficiency for conversional devices is similar to a logistic growth curve. The first generation of any conversional device starts off low on the efficiency curve, then improvements are made to later generations and the efficiency improves, and finally there comes a point in time when, for various reasons, the efficiency of later generations peaks and plateaus.

This basic process applies to just about any conversional device you can think of. The first steam engines in the early 18th century were very inefficient compared to later steam engines in the following century. The first internal combustion engines had a thermodynamic efficiency of less than 5% whereas current engines hover around 30% efficiency. The first air conditioners were much more inefficient than later waves of air conditioners. NVIDIA’s Hopper chips are less efficient than their Blackwell counterparts, and Blackwell itself will be less efficient than future generations of AI chips. And on and on it goes. This universal feature in the development of conversional technologies happens because the central goal when developing a new technology is to ensure basic functionality and operational capacity. Refinements and improvements are postponed for later as they often require large investments.

The end result is that aggregate economy-wide efficiencies experience a lot of downward pressure as we’re constantly flooding society with a barrage of new conversional technologies that are starting off very low on the efficiency frontier. Aggregate efficiencies can and often do improve over time, as I explain in the book, because many older technologies and their variations are still climbing the efficiency barrier shown above (the middle part of the chart), but these aggregate-level improvements are generally slow to materialize. In addition, newer technological descendants of older technologies might be more efficient than their ancestral variations, but that doesn’t mean that the older technologies magically go away. There are still hundreds of thousands of Hopper chips operating in numerous data centers even as NVIDIA is selling more and more Blackwell chips. At the same time, when dominant catalysts like steam engines and AI chips first emerge on the scene, they start off as very inefficient and greatly contribute to the inertia of aggregate efficiencies.

The Economic Dynamics of Technological Change

So far, I’ve described the biophysical and energetic dynamics of catalytic spectralization and how they affect the broader energy scale and aggregate efficiencies of our economies. But now let me say something about why this kind of technological change happens under capitalism in the first place. In the book, I clarify that technological innovation has no inherent teleology. There is no fundamental law of nature that says technologies are always aiming for maximizing efficiency or energy throughput or anything like that. Instead, technological change unfolds in the context of certain historical, political, economic, and geostrategic conditions. And it's these high-level social and class-power dynamics that operate as de facto selection mechanisms, deciding the broader economic contours of spectralization and also which new technologies diffuse throughout the economy.

The unique political economy, competitive dynamics, and class struggles of capitalism pressure major corporations towards pursuing automation, accelerated production, and industrial agglomeration. That’s because dominant capitalists are always looking to achieve capital accumulation: to boost stock prices, market capitalizations, revenues, profits, and any other aspects of the financial domain that would allow them to reinforce their class power and strategic control over society at large. And in the competitive dynamics that arise in the capitalist economy, capitalists are always aiming to boost accumulation relative to others. They want to beat out their rivals and competitors. They try to achieve that goal by controlling labor costs through automation, by accelerating production to undercut rivals in the market, by erecting barriers to competition for newcomers, by manipulating the political system to expand their advantages, and through so many other methods. Just look at the recent AI craze to understand how this affects catalytic spectralization. Because all of the hyperscalers want to have that critical first mover advantage in this new market, they are spending hundreds of billions on capex and racing to ensure that they can build the best and fastest AI models. That has meant launching a complete and total overhaul of their data center technologies and infrastructure. Companies like META are building out new data center “megaclusters” that will require vast amounts of energy and electricity to operate. In sum, the desire to control and dominate this new market is pushing elite capitalists to invest vast sums of money into developing new technologies.

The competitive dynamics of capitalism generally push major corporations towards finding or creating new markets, a seemingly incessant and perpetual agoragenesis. That’s because existing markets eventually become saturated and companies start hitting diminishing returns. Capitalism is all about growth, and this push for more growth intensifies agoragenesis. Establishing a new market could mean expanding the geographic scale of market activity, like when McDonald's opened its first store in China back in 1990. But it can also mean introducing a new product or service within the same country or operational area. NVIDIA was once known for making GPUs for video games, and it still does that. But then deep learning and generative AI came along, and suddenly NVIDIA found an explosive new market for its AI chips, catapulting the company to the top of the stock charts in 2024.

Even in an era of entrenched monopolization and cartelization in the American economy, these competitive dynamics still persist. Google is the de facto monopolist in online search, Meta is the de facto monopolist in social media, Amazon is the de facto monopolist in e-commerce, Microsoft is the de facto monopolist in PC operating systems, and yet all of them are engaged in a highly public corporate war to develop the fastest and most impressive AI models. In a very new market, there is generally an absence of a dominant monopolist. Monopolies and cartels only emerge later, once the market matures. Then the process of agoragenesis begins again, either because new competitors come along to introduce new markets (like when Netscape came along with its new web browser in the 1990s) or because the monopolies and cartels of established markets attempt to create other markets themselves. These new markets feature a whole new range of products, services, and technologies, and conversional technologies in particular are central to the story because they are implicated in just about everything we do. This is the ultimate social and economic basis for catalytic spectralization in the capitalist regime.

This is one of the major reasons why in the book I advocate for social and political transformation above and beyond mere technological innovation: because the energy-intensive nature of modern civilization is built on the class-power dynamics of modern capitalism. Reforming our social, political, and economic institutions and forms of organization should therefore be the primary concern. In a future post, I will explain how the concept of a modular society under the valerist system I introduced in the book can provide a comprehensive solution to the problem of catalytic spectralization. This post will delve into the concept of modularization and explain the changes we should make to the process of technological development, to industrial manufacturing and industrial strategy more broadly, and to the process of end-use consumption.