Janisz Varufakis recently argued that artificial intelligence (AI) will destroy capitalism by eliminating wage labour and concentrating corporate profits in large tech platforms, shrinking monetary measures of economic value. That diagnosis captures some real trends — which the author calls the “productivity illusion” — but its conclusion rests on a narrow picture of an economy made mainly of digitally deliverable services and omits several important constraints.
What the "productivity illusion" gets right
It is observable that the volume of performed tasks can decouple from their market value: when the price of a translation or an administrative task falls sharply, statistical turnover may decline. For consumers and small firms, however, the saved money does not vanish but is spent elsewhere — on machinery, renovations or consumption. The electricity market provides an analogy: abundant solar production can push prices to zero or negative values during midday, yet that does not make the economy poorer; it reallocates value to storage, flexibility and network investment.
Work does not disappear, it relocates
In many parts of Europe the immediate labour problem is not mass unemployment but labour shortages: plumbers, electricians, carpenters and nurses are hard to find. AI is unlikely to replace these place-bound, manual jobs; at best it can streamline paperwork, diagnostics or design. Economics long recognizes this mechanism: where productivity rises fast, labour tends to become relatively cheaper; where it does not, labour can become relatively more expensive. A decline in the value of office work could therefore raise wages and prestige in manual trades — a reallocation rather than the end of the economic system. The transition will nonetheless create reskilling and policy challenges for affected workers.
Trust networks resist full platformization
Varufakis cites online retail platforms where algorithms match buyers and sellers. That model works well for standardized, easily comparable goods, but in construction, renovation or bespoke services quality often emerges only later — sometimes years after completion. In these cases star ratings and platform mechanisms are weak filters; personal recommendations and trust networks matter more. AI can amplify the problem by generating large numbers of fake reviews, profile texts and reference photos, reducing the credibility of online information and boosting the relative value of personally verifiable references.
A premium for verifiable human work
Mass production did not eliminate craftsmanship; it transformed it into a luxury segment. The mechanical-watch industry survived the quartz crisis by becoming a status category precisely because of human workmanship. A similar dynamic could appear now: items or services that can be credibly certified as human-made — furniture, artworks, texts, live performances — may command a premium because of scarcity. But that premium requires verifiability and limited supply; if human-made goods cannot be distinguished from machine-generated ones, or if too many suppliers enter the market, the price advantage will disappear.
Demography and regional variation matter
AI’s consequences depend heavily on demographic and regional context. In an ageing, shrinking labour-force region like Europe, AI primarily fills gaps left by retiring cohorts. By contrast, in many African countries a young and fast-growing population is seeking jobs, and their development strategies have relied on outsourced services, call-centres and data processing. Those laddering opportunities are more directly threatened by AI. Thus Varufakis’s concerns may be more relevant in some regions than in others — it is a crisis of a development model, not necessarily of capitalism globally.
Energy is a limiting factor: numbers and implications
Varufakis often treats AI as a technology that can scale without physical constraint. But AI runs on physical infrastructure, and electricity and network capacity impose real limits. The International Energy Agency (IEA) projects that data-centre electricity use will rise from roughly 485 TWh in 2025 to about 950 TWh in 2030; AI-specialized data centres may grow even faster. Globally this demand is manageable (it would represent a few percent of total electricity), yet locally it can create severe bottlenecks: grid connections, permitting and storage are common constraints. If electricity becomes the binding limit, rent will not necessarily accrue to cloud owners but to places with cheap, reliable power and strong transmission — strengthening the bargaining position of the energy sector and power-rich countries.
Shared errors of the debate’s protagonists
Varufakis is not alone in drawing sweeping conclusions. Daron Acemoglu, the MIT economist, offers a skeptical estimate that AI will raise total factor productivity by at most 0.66 percent over the next decade and will likely widen income gaps between capital and labour. Elon Musk, by contrast, predicts that within 10–20 years work will become optional and AI and robotics will create abundance and universal high incomes. Both positions err in similar fashion: they reason along a single axis — capital dynamics, task-based statistics or exponential technology curves — and therefore understate the complexity introduced by energy, trust-based work, demographics and geopolitical shocks.
Implications
The future will be determined by multiple interacting factors: productivity shifts, energy and network constraints, the structure of work, demographic trends, regulation and geopolitics. The productivity illusion exists — tasks can become cheaper without implying overall impoverishment — but equally illusory are forecasts that declare the end of capitalism, the definitive low impact of AI based on narrow ten-year models, or a guaranteed, work-free abundance. Robust policy must address transition costs (training, grid and storage investment), account for regional differences, and prepare for shocks that could rapidly change which skills and infrastructures are most valuable.
Final remark
AI can drive significant economic reallocation: some activities will become far cheaper while others gain relative value. The political task is not to adjudicate a single deterministic outcome but to manage the redistribution of work and value, invest in networks and energy, and design education and social policies that reduce the human costs of technological change.
