Global geopolitics

AI, data, and automation are remaking society, while energy, water, jobs are quietly redistributed and structural change is accelerating without public debate

A careful examination of the past twenty-four hours reveals not a collection of disconnected announcements, but a set of developments that, taken together, illustrate the current direction of travel in advanced economies: heavy capital deployment into artificial intelligence, limited short-term productivity gains, and a clear medium-term trajectory toward labour displacement across multiple sectors. At the same time, some high-profile voices, including David Sacks, have challenged the prevailing narrative, noting that AI has accounted for only a small share of reported layoffs, approximately 4.7 percent by some self-reported measures, suggesting that near-term labour impacts may be overstated or at least unevenly distributed. What is less frequently acknowledged, however, is that this reallocation is not occurring in a neutral institutional environment. The direction, pace, and ownership of AI deployment are being set largely by a small number of corporations and aligned policy actors, with minimal public deliberation. As a result, the current transition is not only an economic shift, but a redistribution of decision-making authority over core systems that structure everyday life.

The most consequential macroeconomic signal came from Goldman Sachs, whose chief economist publicly acknowledged that approximately $450 billion in recent AI-related investment has contributed “essentially zero” to measured US GDP growth to date. This aligns with broader empirical work on technology diffusion. Historical data from the adoption of electricity and information technology show that productivity gains typically lag initial investment by 5–15 years. However, Goldman Sachs itself has previously estimated that generative AI could eventually raise global GDP by around 7 percent over a decade, implying that current capital expenditure is being justified by anticipated, not realised, returns. The immediate implication is a widening gap between financial valuations and underlying economic output, a condition that has historically preceded periods of market correction or consolidation. It also raises the possibility, though not the certainty, that AI investment could follow the trajectory of previous speculative cycles such as the dot-com bubble or the 2008 financial crisis if expected returns fail to materialise. Beyond questions of valuation, this dynamic raises issues of democratic legitimacy. Trillions in capital are being deployed to reshape labour markets, infrastructure, and information systems without corresponding mechanisms for public consent. Unlike traditional public investment cycles, where spending priorities are at least indirectly mediated through political processes, AI capital allocation is primarily driven by private incentives, with public institutions often reacting after the fact.

At the firm level, capital-labour substitution is becoming explicit. Uber’s $1.25 billion investment into Rivian to produce autonomous vehicles represents a direct shift in cost structure. Labour accounts for a substantial share of ride-hailing expenses; removing drivers has the potential to improve margins significantly. Industry estimates suggest that autonomous fleets could reduce per-mile costs by 30–50 percent over time. The consequence for the labour market is equally clear: in the United States alone, several million individuals derive income from driving-related work. Even partial automation would exert downward pressure on wages and participation in that segment. This shift is not simply a matter of efficiency. It represents a rebalancing of bargaining power between labour and capital. As automation reduces reliance on human workers in key sectors, the ability of labour to negotiate wages or conditions weakens structurally. In this sense, automation functions not only as a cost-saving technology, but as a mechanism for altering the distribution of economic agency.

A parallel dynamic is visible in industrial strategy. Jeff Bezos’s reported effort to raise $100 billion for acquisitions in manufacturing points toward consolidation followed by automation. Manufacturing already accounts for a disproportionately high share of industrial robot usage; according to the International Federation of Robotics, global robot installations reached over 550,000 units annually in recent years, with China, Japan, and the United States leading adoption. The economic logic of acquiring existing firms rather than building new ones lies in immediate access to supply chains, customer bases, and physical infrastructure. Once acquired, these assets can be restructured with automation technologies, reducing labour intensity and increasing output per worker. The likely consequence is a further decoupling of industrial output from employment levels. Consolidation at this scale also has political implications. As ownership of productive assets becomes more concentrated, so too does influence over supply chains, pricing, and regional employment. This blurs the boundary between market power and systemic importance, where certain firms become not just participants in the economy, but infrastructural actors whose decisions carry economy-wide consequences.

The role of data as a production input is also becoming more visible. DoorDash’s initiative to pay individuals to record themselves performing routine tasks reflects a growing demand for high-quality, labelled data for training AI systems. In machine learning, performance improvements are closely tied to dataset scale and quality. By some estimates, the global market for data labelling and annotation is expected to exceed $10 billion within the next few years. The implication is that human activity itself is being commodified at a granular level, not merely as labour, but as training input. Over time, this creates a feedback loop in which human work generates the data that enables its own automation. This dynamic extends the logic of extraction into new domains. Individuals are not only selling their labour, but also generating the raw material, behavioural data, used to automate that labour. The asymmetry lies in ownership: while data is collectively produced, it is privately captured and monetised, reinforcing a model in which value creation is broadly distributed but value appropriation is highly concentrated.

In the software ecosystem, consolidation pressures are intensifying. OpenAI’s acquisition of widely used open-source tooling and its integration into proprietary systems reflects a broader pattern in the technology sector, where open innovation is absorbed into closed commercial platforms. This has precedent in cloud computing and mobile operating systems. The economic consequence is a shift in value capture: developers who once relied on open ecosystems may find themselves dependent on vertically integrated providers, with pricing power concentrated among a small number of firms. This shift toward vertically integrated ecosystems introduces a form of structural dependency. Developers and businesses operating within these environments may retain formal independence, but their practical ability to compete or innovate becomes contingent on access to privately controlled infrastructure. Over time, this can resemble a quasi-feudal relationship, where participation in the digital economy is conditional on adherence to platform-defined rules.

At the same time, competitive dynamics remain unusually fluid. Cursor’s release of a model reportedly outperforming systems from Anthropic on specific coding benchmarks highlights the rapid diffusion of capability. Benchmark results such as HumanEval and similar coding tests have shown that performance gaps between leading models can narrow quickly. The implication is that while capital requirements are high, innovation cycles are short, allowing smaller teams to compete in narrowly defined domains. This contributes to both rapid progress and market volatility, even as underlying infrastructure and distribution channels remain concentrated.

The cultural and legal dimensions are also evolving. The digital recreation of Val Kilmer raises questions about intellectual property, consent, and posthumous rights. The use of AI-generated likenesses is already the subject of ongoing negotiations between studios and unions, particularly in the United States. The Screen Actors Guild has previously highlighted concerns that digital replicas could be reused indefinitely without proportional compensation. The broader implication is that identity itself is becoming an asset class subject to licensing, replication, and dispute, further extending the reach of commodification into previously non-economic domains.

In financial services, restructuring is underway. HSBC’s consideration of workforce reductions in middle and back-office functions reflects a sector-wide trend. Banking operations involve large volumes of routine, rule-based processes, precisely the type of work most susceptible to automation. Consulting estimates suggest that up to 30–40 percent of banking roles could be significantly altered or eliminated by AI and automation technologies over the next decade. The immediate effect is cost reduction; the longer-term effect is a reconfiguration of employment toward higher-skill, lower-volume roles. The broader implication is that efficiency gains at the firm level may translate into instability at the system level. As employment becomes less secure and income more volatile, the resilience of debt-dependent economies may weaken, particularly where social safety nets have not adapted to these structural changes.

These developments are occurring alongside broader macroeconomic fragilities that extend beyond AI itself. Rising energy prices, particularly in the context of geopolitical instability such as conflicts in the Middle East, risk feeding directly into inflation through higher input costs, effectively acting as an “invisible tax” on households and firms. At the same time, high levels of public and private debt amplify systemic vulnerability. In the United States, federal debt has surpassed $36 trillion, while many advanced economies face similar structural pressures related to ageing populations, entitlement spending, and interest obligations. Because modern economies are deeply debt-based, even modest shocks, such as a small percentage of white-collar job losses, could trigger cascading defaults among households heavily exposed to mortgages, auto loans, and student debt.

This creates a set of interlocking risks. If AI adoption proves deflationary, reducing labour income and prices in certain sectors, it may clash with a system that depends on nominal growth to sustain debt repayment. If governments attempt fiscal consolidation, reduced spending may slow GDP growth, worsening debt-to-GDP ratios and limiting policy flexibility. Conversely, if debt becomes politically or mathematically unmanageable, the historical tendency has been toward monetary expansion, raising the risk of currency debasement or renewed inflation. None of these outcomes are predetermined, but their coexistence illustrates the narrow policy path available.

The scale of investment reinforces the systemic nature of the shift. Estimates from large corporations and industry groups place the total cost of AI infrastructure buildout at several trillion dollars, often cited around $4–5 trillion globally over the coming decade. This includes data centres, semiconductors, networking, and energy capacity. Such levels of capital expenditure are comparable to previous industrial transformations, including the expansion of railways and electrification.

At the centre of this infrastructure buildout is Nvidia, whose hardware underpins a substantial portion of current AI workloads. During the recent Nvidia GTC, Jensen Huang presented developments including humanoid robotics and digital avatars, underscoring the breadth of application domains being targeted. Nvidia’s revenue growth, driven largely by demand for AI chips, has already exceeded 200 percent year-on-year in recent reporting periods, illustrating how value is currently concentrated among suppliers of core infrastructure. Control over this infrastructure also carries strategic significance. Data centres, compute capacity, and AI models are increasingly analogous to critical national infrastructure, yet much of this capacity is owned and operated by private entities with transnational reach. This raises questions about sovereignty: who ultimately controls the systems that mediate communication, commerce, and information at scale?

Taken together, these developments indicate a transitional phase in which capital is being rapidly reallocated toward AI systems, while measurable economic gains remain limited and labour market impacts are only beginning to emerge. The historical pattern suggests that productivity improvements may eventually materialise, but the distribution of those gains, between capital owners and labour, remains uncertain. What is perhaps most striking is not simply the scale or speed of these developments, but the limited degree of meaningful public engagement around them.

The increasingly popular “doomer” framing of these dynamics, emphasising inevitable collapse through debt spirals, technological unemployment, or policy failure, captures real structural risks but can obscure the range of possible outcomes. The scenarios outlined above, from AI-driven deflation in a leveraged system to asset bubbles and monetary instability, are all plausible within current constraints. However, they are not deterministic endpoints. The path taken will depend on institutional responses, policy choices, and the distribution of economic power. Recognising systemic risk is analytically useful; defaulting to fatalism is not.

One reason for the broader disconnect lies in the way artificial intelligence is being framed. It is presented simultaneously as an inevitability and as a technical upgrade, an efficiency gain rather than a structural reordering. This framing limits scrutiny. Unlike trade agreements or austerity programmes, which are explicitly political and therefore contested, AI adoption is often treated as a neutral or apolitical process. Yet the distributional consequences are anything but neutral. Decisions about automation, data ownership, and capital allocation are, in effect, decisions about who captures economic value and who is displaced from it. The absence of widespread public engagement is not incidental. By framing AI as inevitable, the space for democratic contestation narrows, and by the time impacts become visible, the underlying structures may already be difficult to unwind.

There is also a temporal asymmetry at work. The benefits of AI, at least in financial terms, are being realised immediately by firms and investors through cost reductions and valuation increases. The costs, particularly in terms of employment displacement, are more diffuse and delayed. This weakens the incentives for early political intervention. By the time labour market effects become fully visible, the underlying economic structures may already be difficult to reverse.

A further factor is the opacity of the infrastructure itself. The expansion of large-scale data centres, operated by firms such as Amazon Web Services, Microsoft, and Google, is not widely understood outside specialist circles. These facilities are highly energy-intensive; recent estimates suggest that data centres could account for 3–5 percent of global electricity consumption within the next decade, with AI workloads significantly increasing demand. Water usage is also substantial, particularly for cooling systems. A single large data centre can consume millions of litres of water per day under certain conditions. These are not marginal resource demands. They compete directly with residential, agricultural, and industrial usage, particularly in regions already facing environmental stress. Yet these trade-offs are rarely made explicit in public discourse, reinforcing both opacity and the absence of meaningful consent.

The question of employment is even more fundamental. If, as many estimates suggest, a significant proportion of existing roles are partially or wholly automatable, then the issue is not simply job transition but income distribution. Modern economies are structured around wage labour as the primary means of accessing goods and services. If labour demand declines structurally, that model becomes unstable. Without structural adjustment, this trajectory risks decoupling economic participation from economic security, potentially eroding the social contract that underpins market economies.

Various policy responses have been proposed, though none have yet achieved consensus or large-scale implementation. Universal basic income is frequently discussed as a mechanism to decouple income from employment, but it raises questions about fiscal sustainability and political acceptance. Wage subsidies, job guarantees, and reduced working hours represent alternative approaches, each with different implications for productivity and public finance. There is also the possibility of expanding public or cooperative ownership of AI infrastructure, allowing the returns from automation to be distributed more broadly rather than concentrated among shareholders.

Another avenue lies in redefining what constitutes economically valuable work. Sectors such as care, education, and environmental restoration remain labour-intensive and socially necessary, yet are often undervalued in market terms. Redirecting investment and policy support toward these areas could partially offset displacement elsewhere, though this would require deliberate political choice rather than market-driven allocation.

The more extreme question, whether large-scale technological displacement implies or necessitates a reduction in population, is not supported by economic evidence. Historically, technological progress has altered the composition of employment rather than eliminating the need for human participation altogether. However, the speed and breadth of current developments do raise legitimate concerns about transitional dislocation. The relevant issue is not population size per se, but the capacity of institutions to adapt distribution mechanisms, education systems, and social safety nets to new conditions.

These dynamics, capital concentration, labour displacement, infrastructural control, and limited democratic input, are not isolated effects. They reinforce one another, producing a system in which economic, technological, and political power increasingly converge.

Ultimately, the core issue is institutional lag. Economic transformation is proceeding at a pace that exceeds the ability of political and social systems to respond. Without deliberate intervention, outcomes will be determined primarily by existing distributions of capital and power. That is not a technological inevitability; it is a policy choice, whether explicit or implicit. The central question is not simply how societies adapt to AI, but who has the authority to shape that adaptation, and whether that authority remains broadly distributed or becomes increasingly concentrated in ways that are difficult to contest or reverse.

Authored By: Global GeoPolitics

Thank you for visiting. This is a reader-supported publication. If you believe journalism should serve the public, not the powerful, and you’re in a position to help, becoming a PAID SUBSCRIBER truly makes a difference. Alternatively you can support by way of a cup of coffee:

https://buymeacoffee.com/ggtv |

https://ko-fi.com/globalgeopolitics |

Bitcoin: 3NiK8BoRZnkwJSHZSekuXKFizGPopkE7ns