The present book aims to supplement these other perspectives by giving an economic history of capitalism and digital technology, while recognising the diversity of economic forms and the competitive tensions inherent in the contemporary economy. The simple wager of the book is that we can learn a lot about major tech companies by taking them to be economic actors within a capitalist mode of production. This means abstracting from them as cultural actors defined by the values of the Californian ideology, or as political actors seeking to wield power. By contrast, these actors are compelled to seek out profits in order to fend off competition. This places strict limits on what constitutes possible and predictable expectations of what is likely to occur. Most notably, capitalism demands that firms constantly seek out new avenues for profit, new markets, new commodities, and new means of exploitation. For some, this focus on capital rather than labour may suggest a vulgar econo-mism; but, in a world where the labour movement has been significantly weakened, giving capital a priority of agency seems only to reflect reality.

We might turn to the technology sector, but, strictly speaking, this sector remains a relatively small part of the economy. In the United States it currently contributes around 6.8 per cent of the value added from private companies and employs about 2.5 per cent of the labour force. By comparison, manufacturing in the deindustrialised United States employs four times as many people. In the United Kingdom manufacturing employs nearly three times as many people as the tech sector. This is in part because tech companies are notoriously small. Google has around 60,000 direct employees, Facebook has 12,000, while WhatsApp had 55 employees when it was sold to Facebook for $19 billion and Instagram had 13 when it was purchased for $1 billion. By comparison, in 1962 the most significant companies employed far larger numbers of workers: AT&T had 564,000 employees, Exxon had 150,000 workers, and GM had 605,000 employees. Thus, when we discuss the digital economy, we should bear in mind that it is something broader than just the tech sector defined according to standard classifications.

As a preliminary definition, we can say that the digital economy refers to those businesses that increasingly rely upon information technology, data, and the internet for their business models. This is an area that cuts across traditional sectors – including manufacturing, services, transportation, mining, and telecommunications – and is in fact becoming essential to much of the economy today. Understood in this way, the digital economy is far more important than a simple sectoral analysis might suggest. In the first place, it appears to be the most dynamic sector of the contemporary economy – an area from which constant innovation is purportedly emerging and that seems to be guiding economic growth forward. The digital economy appears to be a leading light in an otherwise rather stagnant economic context. Secondly, digital technology is becoming systematically important, much in the same way as finance. As the digital economy is an increasingly pervasive infrastructure for the contemporary economy, its collapse would be economically devastating. Lastly, because of its dynamism, the digital economy is presented as an ideal that can legitimate contemporary capitalism more broadly. The digital economy is becoming a hegemonic model: cities are to become smart, businesses must be disruptive, workers are to become flexible, and governments must be lean and intelligent. In this environment those who work hard can take advantage of the changes and win out. Or so we are told.

The argument of this book is that, with a long decline in manufacturing profitability, capitalism has turned to data as one way to maintain economic growth and vitality in the face of a sluggish production sector. In the twenty-first century, on the basis of changes in digital technologies, data have become increasingly central to firms and their relations with workers, customers, and other capitalists. The platform has emerged as a new business model, capable of extracting and controlling immense amounts of data, and with this shift we have seen the rise of large monopolistic firms.

In part, this book is a synthesis of existing work. The discussion in Chapter 1 should be familiar to economic historians, as it outlines the various crises that have laid the groundwork for today's post-2008 economy. It attempts to historicise emerging technologies as an outcome of deeper capitalist tendencies, showing how they are implicated within a system of exploitation, exclusion, and competition. The material in Chapter 2 should be fairly well known to those who follow the business of technology. In many ways, the chapter is an attempt to give clarity to various ongoing discussions in that world, as it lays out a typology and genesis of platforms. By contrast, Chapter 3 hopefully offers something new to everyone. On the basis of the preceding chapters, it attempts to draw out some likely tendencies and to make some broad-brush predictions about the future of platform capitalism. These forward-looking prognoses are essential to any political project. How we conceptualise the past and the future is important for how we think strategically and develop political tactics to transform society today. In short, it makes a difference whether we see emerging technologies as inaugurating a new regime of accumulation or as continuing earlier regimes. This has consequences on the possibility of a crisis and on deciding where that crisis might emerge from; and it has consequences on our envisaging the likely future of labour under capitalism. Part of the argument of this book is that the apparent novelties of the situation obscure the persistence of longer term trends, but also that today presents important changes that must be grasped by a twenty-first-century left.

Capitalism, uniquely among all modes of production to date, is immensely successful at raising productivity levels. This is the key dynamic that expresses capitalist economies' unprecedented capacity to grow at a rapid pace and to raise living standards. What makes capitalism different? This cannot be explained through psychological mechanisms, as though at some time we collectively decided to become greedier or more efficient at producing than our ancestors did. Instead, what explains capitalism's productivity growth is a change in social relationships, particularly property relationships. In precapitalist societies, producers had direct access to their means of subsistence: land for farming and housing. Under those conditions, survival did not systematically depend on how efficiently one's production process was. The vagaries of natural cycles may mean that a crop did not grow at adequate levels for one year, but these were contingent constraints rather than systemic ones. Working sufficiently hard to gain the resources necessary for survival was all that was needed. Under capitalism, this changes. Economic agents are now separated from the means of subsistence and, in order to secure the goods they need for survival, they must now turn to the market. While markets had existed for thousands of years, under capitalism economic agents were uniquely faced with generalised market dependence. Production therefore became oriented towards the market: one had to sell goods in order to make the money needed for purchasing subsistence goods. But, as vast numbers of people were now relying upon selling on the market, producers faced competitive pressures. If too costly, their goods would not sell, and they would quickly face the collapse of their business. As a result, generalised market dependency led to a systemic imperative to reduce production costs in relation to prices. … The outcome of these competitive actions was eventually expressed in the mediumterm tendencies of capitalism: prices tangentially declined to the level of costs, profits across different industries tended to become equal, and relentless growth imposed itself as the ultimate logic of capitalism.

One of the most important consequences of this schematic model of capitalism is that it demands constant technological change. In the effort to cut costs, beat out competitors, control workers, reduce turnover time, and gain market share, capitalists are incentivised to continually transform the labour process. This was the source of capitalism's immense dynamism, as capitalists tend to increase labour productivity constantly and to outdo one another in generating profits efficiently.

As its nearest competitors were devastated by the war, American manufacturing profited and was the powerhouse of the postwar era. Yet Japan and Germany had their own comparative advantages – notably relatively low labour costs, skilled labour forces, advantageous exchange rates, and, in Japan's case, a highly supportive institutional structure between government, banks, and key firms. Furthermore, the American Marshall Plan laid the groundwork for expanding export markets and for rising investment levels across these countries. Between the 1950s and the 1960s Japanese and German manufacturing grew rapidly in terms of output and productivity. Most importantly, as the world market developed and global demand grew, Japanese and German firms began to cut into the share of American firms. Suddenly there were multiple major manufacturers that produced for the world market. The consequence was that global manufacturing reached a point of overcapacity and overproduction that put downward pressure on the prices of manufactured goods. By the mid-1960s, American manufacturing was being undercut in terms of prices by its Japanese and German competitors, which led to a crisis of profitability for domestic firms. The high, fixed costs of the United States were simply no longer able to beat the prices of its competitors. Through a series of exchange rate adaptations, this crisis of profitability was eventually transmitted to Japan and Germany, and the global crisis of the 1970s was underway.

In fact the 1990s' boom is redolent of much of today's fascination with the sharing economy, the internet of things, and other tech-enabled businesses.

The 1970s therefore set the stage for the lengthy slump in manufacturing profitability that has since been the baseline of advanced economies. A period of healthy manufacturing growth in the United States began when the dollar was devalued in the Plaza Accord (1985); but manufacturing slumped again when the yen and the mark were devalued over fears of Japanese collapse. And, while economic growth recovered from its 1970s lows, nevertheless the G7 countries have all seen both economic and productivity growth trend downwards.

The bust was staved off through a series of rapid interest rate reductions made by the US Federal Reserve; and these reductions marked the beginning of a lengthy period of ultra-easy monetary policy. Implicitly the goal was to let equity markets continue to rise despite their 'irrational exuberance', in an effort to increase the nominal wealth of companies and households and hence their propensity to invest and consume. In a world where the US government was trying to reduce its deficits, fiscal stimulus was out of the question. This 'asset-price Keynesianism' offered an alternative way to get the economy growing in the absence of deficit spending and competitive manufacturing.

In 2006 US housing prices reached a turning point, and their decline began to weigh on the rest of the economy. Household wealth decreased in tandem, leading to lowered consumption and eventually to a series of mortgage non-payments. As the financial system had become increasingly tied to the mortgage market, it was inevitable that the decline in housing prices would wreak havoc on the financial sector.

Since these companies only need to move intellectual property (rather than entire factories) to different tax jurisdictions, tax evasion is particularly easy for them.

At one end, tax evasion and cash hoarding have left US companies – particularly tech companies – with a vast amount of money to invest. This glut of corporate savings has – both directly and indirectly – combined with a loose monetary policy to strengthen the pursuit of riskier investments for the sake of a decent return. At the other end, tax evasion is, by definition, a drain on government revenues and therefore has exacerbated austerity. The vast amount of tax money that goes missing in tax havens must be made up elsewhere. The result is further limitations on fiscal stimulus and a greater need for unorthodox monetary policies. Tax evasion, austerity, and extraordinary monetary policies are all mutually reinforcing.

Even today, long-term unemployment remains at levels higher than anything seen before the crisis. The effect of all this has been pressure on the remaining employed population – lower weekly earnings, fewer household savings, and increased household debt. In the United States personal savings have been declining from above 10.0 per cent in the 1970s to around 5.0 per cent after the crisis. In the United Kingdom household savings have decreased to 3.8 per cent – a 50-year low and a secular trend since the 1990s.

But the 1970s created a major shift within these general conditions, away from secure employment and unwieldy industrial behemoths and towards flexible labour and lean business models. During the 1990s a technological revolution was laid out when finance drove a bubble in the new internet industry that led to massive investment in the built environment. This phenomenon also heralded a turn towards a new model of growth: America was definitively giving up on its manufacturing base and turning towards asset-price Keynesianism as the best viable option.

There is some truth in this, but the argument goes awry when it situates this class outside of capitalism. Given that the imperatives of capitalism hold for these companies as much as for any other, the companies remain capitalist. Yet there is something new here, and it is worth trying to discern exactly what it is.

A key argument of this chapter is that in the twenty-first century advanced capitalism came to be centred upon extracting and using a particular kind of raw material: data. But it is important to be clear about what data are. In the first place, we will distinguish data (information that something happened) from knowledge (information about why something happened). Data may involve knowledge, but this is not a necessary condition. Data also entail recording, and therefore a material medium of some kind. As a recorded entity, any datum requires sensors to capture it and massive storage systems to maintain it. Data are not immaterial, as any glance at the energy consumption of data centres will quickly prove (and the internet as a whole is responsible for about 9.2 per cent of the world's electricity consumption). We should also be wary of thinking that data collection and analysis are frictionless or automated processes. Most data must be cleaned and organised into standardised formats in order to be usable. Likewise, generating the proper algorithms can involve the manual entry of learning sets into a system. Altogether, this means that the collection of data today is dependent on a vast infrastructure to sense, record, and analyse. What is recorded? Simply put, we should consider data to be the raw material that must be extracted, and the activities of users to be the natural source of this raw material. Just like oil, data are a material to be extracted, refined, and used in a variety of ways. The more data one has, the more uses one can make of them.

Data were a resource that had been available for some time and used to lesser degrees in previous business models (particularly in coordinating the global logistics of lean production). In the twenty-first century, however, the technology needed for turning simple activities into recorded data became increasingly cheap; and the move to digital-based communications made recording exceedingly simple. Massive new expanses of potential data were opened up, and new industries arose to extract these data and to use them so as to optimise production processes, give insight into consumer preferences, control workers, provide the foundation for new products and services (e.g. Google Maps, self-driving cars, Siri), and sell to advertisers. All of this had historical precedents in earlier periods of capitalism, but what was novel with the shift in technology was the sheer amount of data that could now be used.

The problem for capitalist firms that continues to the present day is that old business models were not particularly well designed to extract and use data. Their method of operating was to produce a good in a factory where most of the information was lost, then to sell it, and never to learn anything about the customer or how the product was being used. While the global logistics network of lean production was an improvement in this respect, with few exceptions it remained a lossy model as well. A different business model was necessary if capitalist firms were to take full advantage of dwindling recording costs. This chapter argues that the new business model that eventually emerged is a powerful new type of firm: the platform.

What are platforms? At the most general level, platforms are digital infrastructures that enable two or more groups to interact. They therefore position themselves as intermediaries that bring together different users: customers, advertisers, service providers, producers, suppliers, and even physical objects. More often than not, these platforms also come with a series of tools that enable their users to build their own products, services, and marketplaces. … Rather than having to build a marketplace from the ground up, a platform provides the basic infrastructure to mediate between different groups. This is the key to its advantage over traditional business models when it comes to data, since a platform positions itself (1) between users, and (2) as the ground upon which their activities occur, which thus gives it privileged access to record them.

Finally, platforms are also designed in a way that makes them attractive to its varied users. While often presenting themselves as empty spaces for others to interact on, they in fact embody a politics. The rules of product and service development, as well as marketplace interactions, are set by the platform owner. Uber, despite presenting itself as an empty vessel for market forces, shapes the appearance of a market. It predicts where the demand for drivers will be and raises surge prices in advance of actual demand, while also creating phantom cabs to give an illusion of greater supply. In their position as an intermediary, platforms gain not only access to more data but also control and governance over the rules of the game.

In the wake of the dot-com bust, Google increasingly needed a way to generate revenues, yet a fee-based service risked alienating the users who were the basis of its success. Eventually it began to use the search data, along with cookies and other bits of information, to sell targeted ad space to advertisers through an increasingly automated auction system. When the National Association of Securities Dealers Automated Quotations (NASDAQ) market peaked in March 2000, Google unveiled AdWords in October 2000 and began its transformation into a revenue-generating company. The extracted data moved from being a way to improve services to becoming a way to collect advertising revenues. Today Google and Facebook remain almost entirely dependent on them: in the first quarter of 2016, 89.0 per cent of Google's and 96.6 per cent of Facebook's revenues came from advertisers.

For critical theorists of the web, this rhetoric obscured a shift to business models premised upon the exploitation of 'free labour'. From this perspective, the story of how Google and Facebook generate profit has been a simple one: users are unwaged labourers who produce goods (data and content) that are then taken and sold by the companies to advertisers and other interested parties. There are a number of problems with this account, however. A first issue with the free labour argument is that it often slides into grand metaphysical claims. All social interaction becomes free labour for capitalism, and we begin to worry that there is no outside to capitalism. Work becomes inseparable from non-work and precise categories become blunt banalities. It is important, however, to draw distinctions between interactions done on platforms and interactions done elsewhere, as well as between interactions done on profit-oriented platforms and interactions done on other platforms. Not all – and not even most – of our social interactions are co-opted into a system of profit generation. In fact one of the reasons why companies must compete to build platforms is that most of our social interactions do not enter into a valorisation process. If all of our actions were already captured within capitalist valorisation, it is hard to see why there would be a need to build the extractive apparatus of platforms.

Rather than exploiting free labour, the position taken here is that advertising platforms appropriate data as a raw material. The activities of users and institutions, if they are recorded and transformed into data, become a raw material that can be refined and used in a variety of ways by platforms. With advertising platforms in particular, revenue is generated through the extraction of data from users' activities online, from the analysis of those data, and from the auctioning of ad space to advertisers. This involves achieving two processes. First, advertising platforms need to monitor and record online activities. The more users interact with a site, the more information can be collected and used. Equally, as users wander around the internet, they are tracked via cookies and other means, and these data become ever more extensive and valuable to advertisers. There is a convergence of surveillance and profit making in the digital economy, which leads some to speak of 'surveillance capitalism'. Key to revenues, however, is not just the collection of data, but also the analysis of data. Advertisers are interested less in unorganised data and more in data that give them insights or match them to likely consumers. These are data that have been worked on. They have had some process applied to them, whether through the skilled labour of a data scientist or the automated labour of a machine-learning algorithm. What is sold to advertisers is therefore not the data themselves (advertisers do not receive personalised data), but rather the promise that Google's software will adeptly match an advertiser with the correct users when needed.

If advertising platforms like Google and Facebook laid the groundwork for extracting and using massive amounts of data, then the emerging cloud platforms are the step that has consolidated the platform as a unique and powerful business model.

In the process of building a massive logistical network, Amazon Web Services (AWS) was developed as an internal platform, to handle the increasingly complex logistics of the company. Indeed, a common theme in the genesis of platforms is that they often emerge out of internal company needs.

The logic behind them is akin to how utilities function. Jeff Bezos, Amazon’s chief executive officer, compares it to electricity provision: whereas early factories had each its own power generator, eventually electricity generation became centralised and rented out on an ‘as needed’ basis. Today every area of the economy is increasingly integrated with a digital layer; therefore owning the infrastructure that is necessary to every other industry is an immensely powerful and profitable position to be in.

By moving businesses’ activities onto cloud platforms, companies like Amazon gain direct access to whole new datasets (even if some remain occluded to the platform). … Rather than relying on advertisers’ buying data, these cloud platforms are building up the basic infrastructure of the digital economy in a way that can be rented out profitably to others, while they collect data for their own uses.

As one report puts it, with the industrial internet 'the big winners will be platform owners'. It is therefore no surprise to see traditional manufacturing powerhouses like General Electric (GE) and Siemens, as well as traditional tech titans like Intel and Microsoft, make a major push to develop industrial internet platforms. Siemens has spent over €4 billion to acquire smart manufacturing capabilities and to build its industrial platform MindSphere, while GE has been working rapidly to develop its own platform, Predix. The field has so far been dominated by these established companies rather than being subject to an influx of new start-ups.

This is where industrial platforms come in, functioning as the basic core framework for linking together sensors and actuators, factories and suppliers, producers and consumers, software and hardware. … By positioning themselves as the intermediary between factories, consumers, and app developers, these platforms are ideally placed to monitor much of how global manufacturing operates, from the smallest actuator to the largest factory, and they draw upon these data to further solidify their monopoly position.

Zipcar owns the assets it rents out – the vehicles; Uber does not. The former is a product platform, while the latter is a lean platform that attempts to outsource nearly every possible cost. (Uber aims, however, eventually to command a fleet of self-driving cars, which would transform it into a product platform.) Zipcar, by contrast, might be considered a 'goods as a service' type of platform.

While subscription models have been around for centuries, for example in newspapers, what is novel today is their expansion to new realms: housing, cars, toothbrushes, razors, even private jets. Part of what has enabled these product platforms to flourish in recent years is the stagnation in wages and the decline in savings that we noted in Chapter 1. As less money is saved up, big-ticket purchases like cars and houses become nearly impossible and seemingly cheaper upfront fees appear more enticing.

The classic model of building an engine and then selling it to an airline was a relatively low margin business with high levels of competition. … But margins on the engines themselves remain small, and competition tight. By contrast, the maintenance of these engines involves much higher profit margins – seven times higher, according to estimates. … This prompted Rolls Royce to introduce the 'goods as a service' model, whereby airlines do not purchase the jet engine but pay a fee for every hour one is used. In turn, Rolls Royce provides maintenance and replacement parts.

The raw material of data remains as central to this platform as to any other. Sensors are placed on all the engines and massive amounts of data are extracted from every flight, combined with weather data and information on air traffic control, and sent to a command centre in the United Kingdom. Information on the wear and tear on engines, possible problems, and times for scheduling maintenance are all derived. These data are immensely useful in blocking out competitors and in securing a competitive advantage against any outside maintenance firm that may hope to break into the market. Data on how the engines perform have also been crucial for developing new models: they enabled Rolls Royce to improve fuel efficiency and to increase the life of the engines, and generated another competitive advantage over other jet engine manufacturers. Once again, platforms appear as an optimal form for extracting data and using them to gain an edge over competitors. Data and the network effects of extracting them have enabled the company to establish dominance.

The most notorious part of these firms is their outsourcing of workers. In America, these platforms legally understand their workers as 'independent contractors' rather than 'employees'. This enables the companies to save around 30 per cent on labour costs by cutting out benefits, overtime, sick days, and other costs. It also means outsourcing training costs, since training is only permitted for employees; and this process has led to alternatives forms of control via reputation systems, which often transmit the gendered and racist biases of society.

Most strikingly, between 2005 and 2015, the US labour market added 9.1 million jobs – including 9.4 million alternative arrangement jobs. This means that the net increase in US jobs since 2005 has been solely from these sorts of (often precarious) positions. Similar trends can be seen in self-employment. While the number of people who identify as self-employed has decreased, the number of people who filed the 1099 tax form for self-employment in the United States has increased. What we see here is effectively an acceleration of the long-term tendency towards more precarious employment, particularly after 2008. The same trends are observable in the United Kingdom, where self-employment has created 66.5 per cent of net employment after 2008 and is the only thing that has staved off much higher levels of unemployment.

In this context, self-employment is not a freely chosen path, but rather a forced imposition. A look at the demographics of lean platform workers seems to support this. Of the workers on TaskRabbit, 70 per cent have Bachelor's degrees, while 5 per cent have PhDs. An International Labour Organization (ILO) survey found that workers on Amazon's Mechanical Turk (AMT) also tend to be highly educated, 37 per cent using crowd work as their main job. And Uber admits that around a third of its drivers in London come from neighbourhoods with unemployment rates of more than 10 per cent. In a healthy economy these people would have no need to be microtasking, as they would have proper jobs.

While the other platform types have all developed novel elements, is there anything new about lean platforms? Given the broader context just outlined, we can see that they are simply extending earlier trends into new areas. Whereas outsourcing once primarily took place in manufacturing, administration, and hospitality, today it is extending to a range of new jobs: cabs, haircuts, stylists, cleaning, plumbing, painting, moving, content moderation, and so on. It is even pushing into white-collar jobs – copy-editing, programming and management, for instance. And, in terms of the labour market, lean platforms have turned what was once non-tradable services into tradable services, effectively expanding the labour supply to a near-global level. … These companies have also offloaded costs from their balance sheets and shifted them to their workers: things like investment costs (accommodations for Airbnb, vehicles for Uber and Lyft), maintenance costs, insurance costs, and depreciation costs.

Uber is perhaps the best example of this development, as it collects data on all of its rides, as well as data on drivers, even when they are not receiving a fare. Data about what drivers are doing and how they are driving are used in a variety of ways in order to beat out competitors. For instance, Uber uses the data to ensure that its drivers are not working for other taxi platforms; and its routing algorithms use the data on traffic patterns to plot out the most efficient path for a trip. Data are fed into other algorithms to match passengers with nearby drivers, as well as to make predictions about where demand is likely to arise. In China, Uber monitors even whether drivers go to protests.

Where is the money coming from? Broadly speaking, it is surplus capital seeking higher rates of return in a low interest rate environment. The low interest rates have depressed the returns on traditional financial investments, forcing investors to seek out new avenues for yield. Rather than a finance boom or a housing boom, surplus capital today appears to be building a technology boom.

In summary, lean platforms appear as the product of a few tendencies and moments: the tendencies towards outsourcing, surplus populations, and the digitisation of life, along with the post-2008 surge in unemployment and rise of an accommodative monetary policy, surplus capital, and cloud platforms that enable rapid scaling. … The profit-making capacity of most lean models likewise appears to be minimal and limited to a few specialised tasks. And, even there, the most successful of the lean models has been supported by VC welfare rather than by any meaningful revenue generation. Far from representing the future of work or that of the economy, these models seem likely to fall apart in the coming years.

We began this chapter by arguing that twenty-first-century capitalism has found a massive new raw material to appropriate: data. … The old business models of the Fordist era had only a rudimentary capacity to extract data from the production process or from customer usage. The era of lean production modified this slightly, as global 'just in time' supply chains demanded data about the status of inventories and the location of supplies. Yet data outside the firm remained nearly impossible to attain; and, even inside the firm, most of the activities went unrecorded. The platform, on the other hand, has data extraction built into its DNA, as a model that enables other services and goods and technologies to be built on top of it, as a model that demands more users in order to gain network effects, and as a digitally based medium that makes recording and storage simple.

Set in context, the lean platform economy ultimately appears as an outlet for surplus capital in an era of ultra-low interest rates and dire investment opportunities rather than the vanguard destined to revive capitalism.

Let us be clear, though: this is not the end of ownership, but rather the concentration of ownership. Pieties about an 'age of access' are just empty rhetoric that obscures the realities of the situation. Likewise, while lean platforms have aimed to be virtually asset-less, the most significant platforms are all building large infrastructures and spending significant amounts of money to purchase other companies and to invest in their own capacities. Far from being mere owners of information, these companies are becoming owners of the infrastructures of society.

With network effects, a tendency towards monopolisation is built into the DNA of platforms: the more numerous the users who interact on a platform, the more valuable the entire platform becomes for each one of them. Network effects, moreover, tend to mean that early advantages become solidified as permanent positions of industry leadership. Platforms also have a unique ability to link together and consolidate multiple network effects. Uber, for instance, benefits from the network effects of more and more drivers as well as from the network effects of more and more riders. Leading platforms tend consciously to perpetuate themselves in other ways as well. Advantages in data collection mean that the more activities a firm has access to, the more data it can extract and the more value it can generate from those data, and therefore the more activities it can gain access to. Equally, access to a multitude of data from different areas of our life makes prediction more useful, and this stimulates the centralisation of data within one platform. We give Google access to our email, our calendars, our video histories, our search histories, our locations – and, with each aspect provided to Google, we get better predictive services as a result. Likewise, platforms aim to facilitate complementary products: useful software built for Android leads more users to use Android, which leads more developers to develop for Android, and so on, in a virtuous circle. Platforms also seek to build up ecosystems of goods and services that close off competitors: apps that only work with Android, services that require Facebook logins. All these dynamics turn platforms into monopolies with centralised control over increasingly vast numbers of users and the data they generate. We can get a sense of how significant these monopolies already are by looking at how they consolidate ad revenue: in 2016 Facebook, Google, and Alibaba alone will take half of the world's digital advertising. In the United States, Facebook and Google receive 76 per cent of online advertising revenue and are taking 85 per cent of every new advertising dollar.

The challenge today, however, is that capital investment is not sufficient to overturn monopolies; access to data, network effects, and path dependency place even higher hurdles in the way of overcoming a monopoly like Google. This does not mean the end of competition or of the struggle for market power, but it means a change in the form of competition. In particular, this is a shift away from competition over prices (e.g. many services are offered for free). Here we come to an essential point. Unlike in manufacturing, in platforms competitiveness is not judged solely by the criterion of a maximal difference between costs and prices; data collection and analysis also contribute to how competitiveness is judged and ranked. This means that, if these platforms wish to remain competitive, they must intensify their extraction, analysis, and control of data – and they must invest in the fixed capital to do so.

Since platforms are grounded upon the extraction of data and the generation of network effects, certain tendencies emerge from the competitive dynamics of these large platforms: expansion of extraction, positioning as a gatekeeper, convergence of markets, and enclosure of ecosystems. These tendencies then go on to be installed in our economic systems.

If data collection is a key task of platforms, analysis is the necessary correlate. The proliferation of data-generating devices creates a vast new repository of data, which requires increasingly large and sophisticated storage and analysis tools, further driving the centralisation of these platforms. If expanding the capacity to collect data is one competitive imperative for these companies, developing corresponding means of analysis is another. Advances in hardware, database organisation, and network infrastructure therefore all play significant roles in gaining speed and insight advantages over one’s competitors. Much of Google’s initial success, for instance, stemmed from its pioneering work of creating useful internal software and innovative hardware architecture. Rather uniquely, Google designs and builds its own custom servers rather than purchasing standard servers off the market – again, in an effort to gain competitive advantage.

In 2016 Facebook began a major push for ‘chatbots’ – that is, low-level AI programmes that would converse with users on Facebook’s platform. (This is also why Facebook – and numerous other companies – are investing heavily in AI and the natural language processing needed to enable chatbots.) The bet is that these chatbots will become the preferred way for users to interact with the internet. On this open platform, businesses would be given the tools to develop their own bots and create intuitive means for users to order food, buy a train ticket, or make a dinner reservation. Rather than using a separate app or website for accessing businesses and services, users would simply access them through Facebook’s platform, which would make Facebook’s chatbot platform the primary interface for commercial transactions online. Rather than trying to compete against Google’s search engine or Amazon’s logistics network, Facebook is trying to dominate the e-commerce platform by dominating the interface.

Given the need to expand data extraction and to position oneself in strategic locations, it would appear that companies are tendentially drawn into similar areas. This means that, despite their differences, companies like Facebook, Google, Microsoft, Amazon, Alibaba, Uber, and General Electric (GE) are also direct competitors.

These first two expansionary tendencies give platform monopolies a distinct path of expansion by comparison to traditional business models premised on vertical integration, horizontal integration, or conglomeration. Instead, platform expansion is driven by the need for more data, which leads to what we might call the convergence thesis: the tendency for different platform companies to become increasingly similar as they encroach upon the same market and data areas.

Indeed, a major reason for Facebook's success is that, while Google dominated the open web through its search technology, Facebook was built as a closed platform that escaped the grips of Google. The aim for Facebook is to make it so that users never have to leave their enclosed ecosystem: news stories, videos, audio, messaging, email, and even buying consumer goods have all been progressively folded back into the platform itself.

A third dominant tendency is the funnelling of data extraction into siloed platforms. When extensive means are not sufficient for competitive advantage, this approach tries to tie users and data to the platform by locking them in through various measures: dependency on a service, inability to use alternatives, or lack of data portability, for instance.

This seems unlikely, as nothing in the industrial internet program appears to radically transform manufacturing, but rather simply to reduce costs and downtime. Rather than improving productivity or developing new markets, the industrial internet appears to drive prices still further down and to increase the competition for market share, thereby exacerbating one of the main impediments to global growth. The platform owners will simply siphon off more of the revenue generated, leaving direct manufacturers with even less.

For all the rhetoric of having overcome capitalism and of transitioning to a new mode of production – a rhetoric inherent in the postindustrial thesis of the 1960s, in the ideas of 'new economy' disciples in the 1990s, and in the radical and conservative paeans to the sharing economy of today – we still remain bound to a system of competition and profitability. Platforms offer new forms of competition and control, but in the end profitability is the great arbiter of success.

The low margins of the business model indicate that services that rely on infrequent tasks (grocery shopping, home cleaning, etc.) are poised to suffer, as they simply do not generate frequent enough revenue to survive. Uber, rather uniquely, has the sweet spot, since so many people need to travel at any point in time. The evidence also suggests that high-skill personal jobs are unlikely to be successful on a lean platform, as they require training (and therefore employees) and they are subject to workers setting out on their own (rather than remaining in an exploitative relationship with a platform). Independent house cleaners, for instance, can often make more than the platforms can profitably offer, and this was ultimately one of the reasons why Homejoy collapsed.

With a sudden drop in venture capital funding, businesses are being forced to become profitable faster. For many low margin services, there are two options: either go out of business or cut costs and increase prices. What is likely to happen is for a large number of these services to go out of business in the next couple of years, while others will move towards becoming luxury services, providing on-demand convenience at high prices. Whereas the tech boom of the 1990s at least left us with the basis for the internet, the tech boom of the 2010s looks as though it will simply leave us with premium services for the rich.

The investment in tech start-ups today is less an alternative to the centrality of finance and more an expression of it. Just like the original tech boom, it was initiated and sustained by a loose monetary policy and by large amounts of capital seeking higher returns.

Some have argued that we might fight these monopolistic trends by building up cooperative platforms. Yet all the traditional problems of coops (e.g. the necessity of self-exploitation under capitalist social relations) are made even worse by the monopolistic nature of platforms, the dominance of network effects, and the vast resources behind these companies. Even if all its software were made open-source, a platform like Facebook would still have the weight of its existing data, network effects, and financial resources to fight off any coop rival.

Rather than just regulating corporate platforms, efforts could be made to create public platforms – platforms owned and controlled by the people. (And, importantly, independent of the surveillance state apparatus.) This would mean investing the state's vast resources into the technology necessary to support these platforms and offering them as public utilities. More radically, we can push for postcapitalist platforms that make use of the data collected by these platforms in order to distribute resources, enable democratic participation, and generate further technological development. Perhaps today we must collectivise the platforms.