The World in 2046: Why AI Is Fast but Rebuilding the Economy Is Slow.

In 2000, investing in the internet was the right idea, yet the Nasdaq crashed 78% and took fifteen years to recover because the market mispriced who would actually capture the value. We are seeing the same pattern with artificial intelligence today. Everyone agrees the technology is revolutionary, but the real challenge is figuring out which parts of the economy can actually integrate and scale it.

To understand where the global economy is heading by 2046, I analyzed ten major papers from institutions like Goldman Sachs, McKinsey, the IEA, and Anthropic. While their forecasts diverge on the details, they point to a single structural shift: three forces—AI, aging demographics, and electrification—are colliding in a fragmented, highly political global market.

A shrinking working-age population raises the pressure to automate, while AI offers a way to produce more output with less human labor. However, scaling these systems requires electricity, data centers, grids, chips, and raw materials.

The bottleneck of this transition has moved from invention to absorption. The models are ready, but scaling them requires changing corporate structures, building massive power grids, training workers, and securing rare minerals. The speed of this transition depends on how fast societies can deploy physical assets, not how fast software improves.

1. Artificial intelligence: large potential, unclear pace

Goldman Sachs is among the most optimistic major institutions. The bank estimates that generative AI could raise global GDP by roughly seven trillion dollars over ten years. 300 million full-time positions would be at least partially exposed. As of early 2026, AI had made no measurable contribution to GDP growth. Goldman expects a broader macroeconomic productivity effect to become visible in the data starting around 2027.

Daron Acemoglu, MIT economist and 2024 Nobel laureate, reaches a significantly different conclusion. His number: at most 0.7% additional productivity growth over a decade. His argument: most AI studies measure simple tasks like call centers and code completion. The difficult, context-dependent tasks are far more resistant. Roughly 20% of all US work tasks are AI-exposed, but only about a quarter of those can be profitably automated — less than 5% of all tasks within ten years.

Seven trillion dollars versus less than one percent. The answer determines trillions in value creation and how radically work, wealth, and society will change.

The difference between the two positions lies less in their assessment of the technology than in their assumptions about implementation. Goldman looks at broad potential. Acemoglu looks at friction in real organizations. A task can't be automated simply because a model can handle it in principle. Processes need to change, data needs to be accessible, responsibilities need to be clarified, errors need to be managed. Both positions share one implication: the less human labor required per unit of output, the more value creation shifts toward those who own models, infrastructure, and data. Capital scales better than labor.

Anthropic published the first paper based on actual usage data in 2026. The finding: a massive gap. In some occupations, the theoretical automation potential exceeds 90%, but actual usage sits at around a third. Systematic unemployment from AI has not materialized through early 2026 — early evidence suggests instead that entry-level positions in exposed occupations are being filled less frequently.

Epoch AI has measured technical progress over more than a decade. Algorithmic efficiency doubles every eight months — three times faster than Moore's Law. Two exponential curves, hardware and algorithms, multiply each other. Whether they'll continue to do so is an open question. Ilya Sutskever, co-founder of OpenAI, said at NeurIPS 2024 that high-quality internet text data is largely exhausted. The industry needs to shift from scaling to algorithmic innovation. Whether that will work, nobody knows.

McKinsey’s 2025 survey confirms this integration bottleneck: although 88% of companies use AI in some form, only 6% scale it successfully enough to generate significant value, and only 39% see any impact on their operating income.

A plausible reading is that Acemoglu and Goldman describe different phases: Acemoglu the coming years of organizational friction, Goldman the potential after a broader reorganization of work and processes. Whether that reorganization happens by 2030, 2035, or later remains open.

2. Demographics: the least visible and most certain force

While technology is volatile, demographic shifts are locked in: the people who will run the economy in 2046 are already born. The IHME analysis published in the Lancet in 2024 shows a long-term decline in fertility across nearly all parts of the world. The global fertility rate stands at 2.2. In 1990 it was 3.3. By 2050, 76% of all countries will be below the replacement rate of 2.1. By 2100, nearly all of them — 97%.

Demographic baselines are more stable than technological forecasts. Their institutional consequences are not — they depend on migration, labor force participation, retirement age, and political reform.

Germany in 2046 will have significantly fewer working-age people per retiree under median population assumptions. The current balance of contributions, benefit levels, and retirement age is unlikely to hold. Median age around 52. Without substantially higher productivity, greater labor force participation, or immigration, the risk of lasting shortages in healthcare, public administration, skilled trades, and logistics increases.

China in 2046 will have 1.3 billion people and be shrinking. Over 400 million will be older than 60. China won't be without workers. But its industrial expansion can no longer draw on a nearly unlimited reservoir of young labor. The demographic burden is likely to constrain further growth significantly — unless automation compensates for the decline faster than in any other country before.

Japan in 2046 will have roughly 110 million people. Median age around 55. Japan has been living in this reality for fifteen years already. Its answer: robotics and automation in healthcare, logistics, agriculture, retail. What Japan is going through today, other developed economies will experience in the 2030s and 2040s.

The United States in 2046 remains comparatively well-positioned demographically among large developed economies, primarily as long as it continues to see net immigration. The US is therefore likely to have the youngest, most dynamic workforce among industrialized nations.

3. Energy: software scales fast, matter scales slowly

The third force is electrification. Transportation, heating, industry, and computing are increasingly powered by electricity.

The IEA estimates that data centers consumed 415 terawatt-hours of electricity worldwide in 2024. By 2030 that figure could more than double to 945 TWh. The more AI becomes part of the economy, the more it becomes an energy and infrastructure issue.

The Oxford analysis by Rupert Way and colleagues examines technological learning curves. Solar has had a learning rate of roughly 20% per capacity doubling for thirty years — each doubling of installed capacity reduces costs by about a fifth. In the probabilistic scenarios modeled by Way and colleagues, solar generation costs fall to a fraction of today's fossil generation costs by mid-century. This doesn't yet say anything about grids, storage, and firm power.

Because cheap generation is not the same as cheap energy. An economy needs electricity at night, in winter, and when the wind is calm. That requires grids, storage, flexible demand, and reserve capacity. The price of a solar panel can fall quickly. Rebuilding an energy system takes much longer.

And then there are the raw materials. The IEA's analysis of critical minerals shows potential supply gaps of roughly 30% for copper and 40% for lithium by 2035. No copper, no power cables. No lithium, no batteries. According to the IEA, new mines take over 16 years on average from discovery to production.

This creates a direct mismatch between digital speed and physical limits: a software model updates in seconds, but building the power lines, data centers, and copper mines to run it takes years.

4. Geopolitics: dependency remains, trust declines

These forces are colliding with an increasingly fragmented global economy. Since 2017, the US-China conflict, pandemics, and war have forced states to prioritize supply chain resilience over pure cost efficiency.

McKinsey's analysis of global trade flows shows that the widely expected broad reshoring has not materialized. The volume of world trade remains largely stable. What's changing are the routes: some production shifts to politically closer countries. At the same time, many strategically important supply chains remain concentrated. For goods dominated by three or fewer countries, roughly 40% of trade still takes place between geopolitically distant partners. This applies to semiconductors, battery raw materials, pharmaceutical precursors, and components for the energy system.

The Munich Security Report 2024 captures the shift in its title: "Lose-Lose?" — the global order is moving away from shared gains toward zero-sum competition.

The global economy will likely be more strongly organized into technological and industrial spheres of influence by 2046. No bloc will be fully self-sufficient. The critical dependencies — chips, energy, minerals, models, industrial manufacturing — will be more heavily politically controlled, subsidized, and restricted in case of conflict. In concrete terms: higher costs for strategic security, longer supply chains, redundant production capacity, and political screening of investments as the new normal.

5. The Friction of the Real World

Three consequences emerge from the papers.

Job losses and labor shortages at the same time. Anthropic's data shows AI displacing standardized knowledge work — research, drafting, analysis, documentation. The IHME projections show healthcare, public administration, and skilled trades running out of workers. A junior analyst loses relevance while a nursing home can't find staff. Both happen simultaneously. They don't cancel each other out — different people, different places, different training.

Productivity splits unevenly. Where AI meets functioning processes, cheap energy, and sufficient infrastructure, output rises fast. Where any of these is missing, the technology remains a tool without much productivity effect. A new model is available in Germany and Texas on the same day. The data center might not get its grid connection in the same year. Goldman and Acemoglu agree on one point: the less human labor is needed per unit of output, the more returns flow to those who own models, infrastructure, and data. How societies handle that will be one of the central political questions of the next twenty years.

The digital future will be decided by physical things. Models spread globally. Power grids, mines, data centers, and skilled workers don't. The IEA's mineral outlook and Way's energy analysis both point to the same constraint: the countries that electrify, build grids, and secure raw materials fastest will absorb AI first. The rest will have access to the same software but not to the same results.

6. Four Paths for the Global Transition

The ten papers show a clear direction but no uncertainty about pace. Two questions decide how fast the transformation arrives: how quickly do organizations reorganize work around AI? And how quickly does physical infrastructure keep up?

The world won't land in one quadrant. Different countries, industries, and companies will live in all four at the same time.

What the ten papers show together

The ten papers don't say the same thing. But they describe the same tension: technical capabilities grow faster than organizations, labor markets, and physical infrastructure can absorb them.

Aging societies cannot leave that gap open indefinitely. They need higher productivity while the very systems they'd need to rebuild for it — energy, institutions, training — respond slowly.

AI scales digitally, but it runs on electricity, chips, grids, and raw materials. Its economic reach remains bound to physical systems that take years to build.

How the world looks in 2046 will be decided less by what AI can do than by which societies absorb that capability into working systems fast enough.

Sources

The ten papers:

1. Goldman Sachs — The Potentially Large Effects of Artificial Intelligence on Economic Growth. Hatzius, Briggs, Kodnani, Pierdomenico. Goldman Sachs Global Investment Research, March 2023.
2. Acemoglu, Daron — The Simple Macroeconomics of AI. NBER Working Paper 32487, 2024. Published in: Economic Policy, 2025.
3. Epoch AI — Algorithmic Progress in Language Models. Ho, Besiroglu, Erdil et al. arXiv:2403.05812, 2024.
4. Anthropic — Labor Market Impacts of AI: Observed Exposure and Adoption. Massenkoff, McCrory. Anthropic Research, March 2026.
5. IHME / Lancet — Fertility, Mortality, Migration, and Population Scenarios for 204 Countries and Territories, 2024–2100. Vollset et al. The Lancet, 2024. DOI: 10.1016/S0140-6736(24)00550-6.
6. McKinsey Global Institute — Geopolitics and the Geometry of Global Trade. January 2024.
7. Munich Security Conference — Munich Security Report 2024: Lose-Lose? February 2024.
8. International Energy Agency — World Energy Outlook 2024. October 2024.
9. Way, Ives, Mealy, Farmer — Empirically Grounded Technology Forecasts and the Energy Transition. Joule, 2022. DOI: 10.1016/j.joule.2022.08.009.
10. International Energy Agency — Global Critical Minerals Outlook 2025. May 2025.

Additional sources:

• McKinsey & Company — The State of AI in 2025. Global Survey, November 2025.
• Sutskever, Ilya — Keynote, NeurIPS 2024.
• IMF — Gen-AI: Artificial Intelligence and the Future of Work. Staff Discussion Note SDN/2024/001.