The Colonisation Never Ended. They Just Changed the Software.
GDP is growing on paper while debt accumulates in reality. AI output goes unmeasured in national accounts. And the same companies that built the first wave of technological dependency are now funding the second. This is not progress. It is extraction with better branding.
Every quarter, governments and international institutions announce that the economy is growing. The figures are cited in headlines, delivered in press conferences, and held up as evidence that the system is working. What they do not announce with equal prominence is that debt is growing faster. That the people doing the growing are not the people receiving the growth. And that the most consequential economic force of the present era, artificial intelligence, is not being measured in the national accounts at all.
Gross Domestic Product measures the total market value of goods and services produced in an economy. It does not measure how that production was financed. It does not subtract the debt accumulated to sustain it. And it does not distinguish between genuine expansion and borrowed consumption dressed up as output.
The International Monetary Fund has long identified 60 per cent of GDP as the threshold beyond which public debt becomes a structural risk for developing economies. For advanced economies, the comparable warning sits at 90 per cent. By 2025, global public debt had exceeded 100 per cent of global GDP. Several major economies were well above that figure. Japan exceeded 250 per cent. The United States crossed 120 per cent. Italy and Greece were both above 140 per cent.
This is not a cyclical problem. It is a structural one. The GDP number continues to rise not because economies are producing more sustainably, but in many cases because governments are borrowing more to sustain the appearance of growth. When the debt-to-GDP ratio rises faster than GDP itself, what is being reported as growth is partly a loan that has not yet been called in.
| Country / Region | Public Debt (% GDP, 2025 est.) | IMF Warning Threshold | Status |
|---|---|---|---|
| Japan | ~255% | 90% (advanced) | Critical excess |
| United States | ~122% | 90% (advanced) | Above threshold |
| Italy | ~140% | 90% (advanced) | Above threshold |
| Mauritius | ~82% | 60% (developing) | Above threshold |
| Sub-Saharan Africa (avg.) | ~62% | 60% (developing) | At threshold |
| Global average | ~100% | Varies | Historic high |
The debt is not abstract. It represents future taxation, future austerity, and future cuts to the public services that the people least served by the current system depend on most. The GDP headline does not tell that story. It was not designed to.
Artificial intelligence presents a measurement problem that economists are only beginning to grapple with honestly. When an AI system drafts a legal document, generates a financial analysis, writes a software module, or produces a marketing campaign, that output has economic value. But it does not appear as a discrete, measurable unit in national accounts. It is absorbed into the productivity of the firm that deployed it, without the human labour it replaced being separately recorded as a loss.
This creates a systematic distortion. GDP may be understating AI's productive contribution while simultaneously failing to capture the displacement it is causing. A law firm that reduced its paralegal headcount by 40 per cent after deploying AI document review software has not recorded a labour market event. It has recorded a productivity gain. The human beings who no longer work there have recorded unemployment. These two entries are in entirely different ledgers, and no official account reconciles them.
The methodology does not connect what people see in the labour market with what the accounts show as growth. Both observations are correct. The system prefers not to notice.
The result is an economy in which the official growth statistics and the lived experience of the labour market are increasingly divergent. People see the jobs disappearing. The accounts show output rising. Both observations are correct. The methodology does not connect them.
To understand the second wave of technological dependency now being built, it is useful to examine the first wave clearly and without sentiment.
Bill Gates and Microsoft did not simply build a successful product. They built a global system of dependency so pervasive that it became invisible. By the mid-1990s, Windows held approximately 90 per cent of the global desktop operating system market. Microsoft Office dominated enterprise computing across every sector and every continent. The decisions made in Redmond, Washington determined the daily working environment of hundreds of millions of people who had no voice in those decisions and no meaningful alternative.
What made this colonisation distinctive was its extension into human capital. Word processing, spreadsheet management, and presentation software became certifiable qualifications. ICDL, MOS, MCP. Professionals across the developing world paid course fees, sat examinations, and added certifications to their CVs not because they had acquired a generalised skill, but because they had demonstrated proficiency in a specific proprietary ecosystem. A typist who could produce a perfect document was considered less employable than one certified in Microsoft Word. The qualification was not about capability. It was about compliance with a standard that one private company controlled and could change at will.
The qualification was not about capability. It was about compliance with a standard that one private company controlled and could change at will.
In 1998, the United States Department of Justice filed antitrust charges against Microsoft, finding that the company had violated the Sherman Antitrust Act through predatory bundling, exclusionary contracts with manufacturers, and deliberate strategies to destroy competing software. The finding was that Microsoft's dominance was not purely the result of a superior product but of conduct designed to eliminate alternatives before consumers could evaluate them. The remedies were modest. The dependency remained.
In many developing countries, including Mauritius, Microsoft Office certifications are still required by employers in 2026, in the full era of AI, for roles in which an AI system could produce the relevant output from a plain-language instruction in under thirty seconds. The cage was built in the 1990s. The certification industry it created is still charging admission.
Microsoft did not retreat when AI arrived. It invested approximately thirteen billion dollars into OpenAI between 2019 and 2024, securing a preferred commercial partner relationship that placed its products at the centre of the most consequential AI deployment in history. Azure became the cloud infrastructure for OpenAI's models. Copilot was integrated into every Microsoft Office product. The company that colonised the world with Windows and Office is now embedding itself into the world's AI infrastructure with the same logic and the same method.
The pattern is consistent. Identify the emerging standard. Invest early and at scale. Integrate vertically so that the product becomes inseparable from the platform. Make switching prohibitively expensive. And do all of this quickly enough that by the time regulators and competitors understand what has happened, the dependency is already structural.
Other companies are running variations of the same play. Google controls the search layer through which most of the world's information is accessed, and is now embedding AI into that layer. Amazon controls the cloud infrastructure on which the majority of the world's digital services run. Meta controls the social communication layer for approximately three billion people and is building its own AI models on top of that distribution. These are not independent companies competing in a neutral market. They are interlocking dependencies that together constitute a private infrastructure on which the global economy increasingly runs, and over which democratic institutions exercise almost no meaningful control.
Beneath the macroeconomic argument lies a human one that is simpler and harder to dismiss.
A registered nurse in England, working under the National Health Service, earns approximately £28,000 to £35,000 per year at the entry band. A registered nurse in Mauritius, performing the same clinical tasks, holding qualifications that meet the same professional standards, and caring for patients with the same degree of skill and dedication, earns the equivalent of approximately £5,400 to £7,600 per year. The pay differential is four to five times. The qualification requirements are the same. The skills are the same. The hours are comparable. The only variable is the country of practice.
| Country | Registered Nurse Pay (annual, approx.) | Ratio vs UK |
|---|---|---|
| United Kingdom | £28,000 to £35,000 | 1x (baseline) |
| Mauritius | £5,400 to £7,600 | ~0.2x |
| Kenya | £3,000 to £5,000 | ~0.12x |
| India | £3,500 to £6,000 | ~0.14x |
| Philippines | £4,000 to £6,500 | ~0.16x |
What is also not the same is the state of the hospital. The NHS, for all its pressures, operates within an infrastructure of capital investment, medical technology, pharmaceutical supply, and estate maintenance that is categorically different from what a Mauritian nurse navigates. The Global South worker is not only paid less for the same skill. She is asked to exercise that skill in a materially inferior environment, with fewer resources, and against a higher burden of unmet need. The investment never arrived. The expectations never changed.
The Global South does not export cheap labour. It exports skilled labour that is paid as though it were cheap. The difference between those two descriptions is not semantic. It is the difference between a market outcome and a structural injustice.
The arrival of AI in this context does not resolve the labour paradox. It compounds it.
AI automation will disproportionately replace the lower-to-middle skill tasks that have historically provided the entry point into formal employment for workers in developing economies: data entry, document processing, customer service, basic legal and financial work, routine coding. These are precisely the roles through which Global South workers have accessed the formal economy and built careers. As these roles are automated, the ladder is removed.
At the same time, the premium tasks that AI cannot yet replicate, those requiring contextual judgement, physical presence, complex interpersonal care, and creative origination, remain in the hands of workers in economies that can afford to pay for them at market rates. A nurse cannot be automated away. But the medical records clerk who supports her can. The administrative infrastructure around care work is already being dismantled. The Global South loses its footholds in the formal economy while the specialised roles that remain are still priced at Global South wages.
AI does not liberate the Global South worker. It gives the global system another reason to pay her less, employ her in fewer roles, and extract more output per unit of cost.
GDP is growing. The debt required to sustain that growth is growing faster. The output generated by AI is not counted in the accounts. The jobs displaced by AI are not counted as losses. The skilled workers of the Global South earn a fraction of what their equivalent counterparts earn in the developed world. The companies that built the first generation of technological dependency are funding the second. And the institutional frameworks that are supposed to provide accountability were designed by the same economies that benefit most from the absence of it.
None of this is accidental. None of it is the natural outcome of a neutral market. It is the predictable result of a system that has consistently priced the labour, the resources, and the participation of the Global South at the minimum necessary to sustain extraction, while calling the aggregate outcome growth.
The colonisation never ended. They just changed the software.
Data notes: Public debt figures sourced from IMF World Economic Outlook 2025. Microsoft antitrust findings from United States v. Microsoft Corporation, 253 F.3d 34 (D.C. Cir. 2001). Microsoft OpenAI investment figures from published company disclosures. Nurse pay comparisons derived from NHS Agenda for Change pay scales 2024-2025 and ILO wage data. AI output measurement gap documented in OECD Going Digital working papers.