---

Discussion

Kate Burvish:

It is both. The forecast reflects real spending on chips, data centers, equipment, and related services. According to the IMF, AI-related goods trade grew 29% between 2023 and 2025, compared with 8.3% for global trade overall. But when one investment cycle supports global demand, expectations become crucial. If productivity or profits disappoint, the correction can hit asset prices, construction plans, corporate borrowing, and household wealth together.

Omar Seidren:

The underlying technology is real. There is genuine demand for more computation, better models, and systems that perform useful work. The important distinction is between demonstrated capability and the financial assumptions built around it. A model can be impressive while the business case for deploying it everywhere remains unsettled. Useful does not automatically mean highly profitable at the speed markets expect.

Eric Arcan:

The boom is also a physical construction program. It needs electricity, transmission, transformers, advanced chips, memory, cooling, and acceptable sites. The International Energy Agency estimates that data-center electricity use could rise from 485 terawatt-hours in 2025 to 950 in 2030, around 3% of global electricity demand. That may be manageable globally, but difficult locally when a cluster arrives before the grid does.

Red Velhouse:

So the question is whether infrastructure and financing can keep up. Omar, could bottlenecks produce a healthy slowdown rather than a market panic?

Omar Seidren:

They could. Shortages of chips, high-bandwidth memory, grid connections, or power equipment can delay projects and raise costs, disciplining unrealistic plans. But scarcity can also concentrate the ecosystem around firms and regions that already control infrastructure. A delay is not automatically a crash; sometimes it is the market discovering that a spreadsheet cannot manufacture a transformer.

Eric Arcan:

Exactly. The binding constraint is often the connection between generation and the load, not generation in the abstract. A gas turbine can be ordered faster than a transmission line. Nuclear power may be excellent long term but cannot solve every near-term timetable. Permitting, transformers, substations, and reliability standards determine whether the electrons arrive when promised.

Kate Burvish:

Those constraints also distribute gains unevenly. The United States and China account for more than three-quarters of net demand for imports of computer chips and other AI-related equipment, while energy importers and lower-income economies are more exposed to the war shock. A resilient global average can conceal very different combinations of investment, inflation, and fiscal pressure.

Red Velhouse:

The forecast also assumes oil and gas shipments through the Strait of Hormuz resume by March 2027 and shipping conditions move toward pre-war conditions by October 2027. Kate, what changes if that normalization does not happen?

Kate Burvish:

The forecast becomes more vulnerable on both growth and inflation. The IMF estimates that a deeper conflict keeping oil above $130 a barrel through the first quarter of 2027 would reduce 2027 global growth by about half a percentage point. That is a scenario, not the baseline. Energy raises costs across the economy while making it harder for central banks to cut rates.

Eric Arcan:

Energy shocks reach transport, chemicals, food production, heating, and electricity. Governments then face pressure to subsidize bills. Temporary, targeted support can protect vulnerable households, but broad subsidies become expensive and preserve price signals poorly. Voters feel the price immediately; fiscal deterioration arrives later.

Omar Seidren:

AI infrastructure is not insulated from that shock. Data centers need reliable power, not merely cheap annual averages. If fuel prices and financing costs rise, facilities may be delayed or concentrated among the largest providers and regions with excess power. The technology can keep advancing while the investment map changes sharply.

Red Velhouse:

And fiscal room is already limited. Global public debt was approximately 93.9% of GDP in 2025, with an IMF projection near 100% by 2029 on current trajectories. Kate, why does that make an investment correction more dangerous alongside an energy shock?

Kate Burvish:

Governments could be asked to cushion several shocks at once. If energy inflation keeps interest rates higher, refinancing costs consume more budget space. Then an AI correction could weaken investment and tax receipts just as households need support. The state can still act, but the response becomes more selective, slower, and politically contested. Public debt turns a manageable shock into a problem of limited options.

Omar Seidren:

There is a related concentration issue in private finance. According to the IMF’s financial-stability analysis, hyperscalers represented about 10% of global investment-grade bond issuance, up from less than 1% in 2024. The IMF also says they had raised more than $100 billion in bond financing since January 2025. These are major companies with strong balance sheets, so default is not the claim. The point is that more of the buildout depends on capital markets believing the revenue story.

Kate Burvish:

Leverage does not create a crisis by itself; it amplifies one when assumptions reverse, liquidity dries up, or spending is concentrated in weak-cash-flow projects. IMF stress tests found aggregate bank capital broadly resilient under a combined AI slowdown and bond-market shock, but identified weak-capital pockets involving about 22% of global banking assets. A shock need not begin inside banks to reach them.

Red Velhouse:

What would distinguish normal cooling from destabilizing repricing? Omar, what should viewers watch beyond technology-stock prices?

Omar Seidren:

Watch the relationship between capability, usage, and monetization. Normal cooling would mean slower capacity additions, more efficient models, and investors demanding clearer returns while useful deployments continue. Destabilizing repricing would be synchronized: falling valuations, canceled data-center and chip orders, weaker corporate borrowing, and evidence that customers are not converting experiments into recurring revenue. The issue is whether the financing chain breaks.

Eric Arcan:

On the physical side, watch project cancellations and equipment orders. A project delayed because a connection is unavailable differs from one abandoned because the economics no longer work. Delays may leave useful grid assets for manufacturing or broader electrification. But specialized facilities and oversized connections can become stranded if demand assumptions collapse.

Kate Burvish:

I would add labor income and trade. The IMF estimates that roughly 40% of jobs globally could be affected by AI, but affected does not mean eliminated. If productivity gains arrive with worker mobility and wage growth, demand can broaden. If bargaining power weakens and transitions concentrate in particular occupations or regions, output may rise while insecurity rises with it.

Red Velhouse:

The IMF also identifies a major cyber incident as a possible catalyst. Omar, how does an AI-enabled attack become a macroeconomic event rather than a problem for one company?

Omar Seidren:

Through common dependencies. AI can accelerate vulnerability discovery, exploitation, fraud, and attacks on shared cloud, software, payment, or telecommunications infrastructure. If many financial institutions rely on the same service and an attack moves quickly, firms may not be able to verify transactions, settle payments, or trust their data. A small number of shared systems can create a very large blast radius.

Kate Burvish:

That shock could arrive when balance sheets are already fragile. Markets might react before technical facts are clear, especially if investors fear payment disruption or forced liquidation. That is why the IMF places cyber incidents alongside an AI correction and sharply higher sovereign yields. The triggers differ, but leverage, concentration, and liquidity transmit them through similar channels.

Eric Arcan:

Energy infrastructure is part of that shared-dependency problem. Modern grids need digital controls, communications, and complex coordination. More computing can improve forecasting and maintenance, but it also increases the systems that must be secured. Reliability is not just having enough megawatts; it is operating under stress and recovering when something fails.

Red Velhouse:

So does AI improve energy security by accelerating investment, or worsen it through dependence on gas, specialized chips, and concentrated supply chains? Eric, what is the practical test?

Eric Arcan:

It can do either. AI demand may justify new generation, grid upgrades, storage, and efficiency investments that benefit other industries. But if expansion outruns planning, it can intensify local price pressure and dependence on whatever can be built fastest, often gas turbines or imported equipment. The test is whether infrastructure serves a durable power system, not merely the next data-center queue.

Omar Seidren:

Efficiency is not a magic eraser. The International Energy Agency reports that global data-center electricity demand grew 17% in 2025 while AI-focused demand grew 50%. Energy per task may fall while total usage rises because people do more tasks, and newer applications such as video, reasoning, and agentic systems are more intensive. The machine gets thriftier; humans ask it to do more.

Kate Burvish:

That is the broader economic tension. AI may raise productivity and expand what firms can produce, yet the gains can be geographically and socially concentrated. Policymakers need fiscal buffers, targeted protection during energy shocks, and investment in skills. Those policies compete with debt service and other spending. The question is who receives the gains and who absorbs the transition costs.

Red Velhouse:

The IMF’s modeled AI downside would lower global gross domestic product, or GDP, by 1.1% by 2028 and United States output by 1.8%, but probably would not, by itself, push global growth into recession. Is the combination the real danger?

Kate Burvish:

Yes. Any one shock may be manageable. The danger is feedback: an AI repricing reduces investment, an energy shock raises inflation and yields, debt limits fiscal support, and weaker income reduces demand. These scenarios are not predictions, but they explain why resilience today does not eliminate vulnerability tomorrow.

Omar Seidren:

I would watch whether useful AI adoption broadens beyond infrastructure spending and a small group of firms. If capability becomes durable productivity across many sectors, the boom has a stronger foundation. If capital keeps racing ahead of real use, correction risk rises. The technology is real; the valuation story remains a live question.

Eric Arcan:

And watch whether power systems keep pace affordably and reliably. Grid connections, transformers, generation, permitting, and cybersecurity will show whether this is a scalable industrial transition or a collection of bottlenecked projects. Physical reality will have the final vote, but good planning determines whether that vote arrives as progress or an expensive surprise.

Red Velhouse:

The unresolved issue is whether AI investment is becoming a durable productivity engine or a concentrated financial cycle vulnerable to disappointment. Over the next six to twelve months, watch AI revenue and deployment, technology valuations and borrowing, data-center cancellations, electricity and energy prices, sovereign yields, labor-market transitions, and signs of cyber disruption. The IMF’s warning is not a forecast of collapse; it is a reminder that separate shocks can become one another’s accelerants. Sources and references for this discussion are available with the episode at Factolio.com.

---

Sources and References

These sources supported the factual material used in this discussion. Factolio’s panel discussion is AI-generated from researched evidence and is written in original language.

1. Reuters — IMF flags risks to global economy from AI market correction, war and debt (NEWS)
2. Associated Press — IMF raises its forecast for world growth this year but warns of risks from AI boom and debt (NEWS)
3. Reuters — IMF flags risks to global economy from AI market correction, war and debt (NEWS)
4. International Monetary Fund — Global Financial Stability Report, October 2026: Resilience, Risk, and Transformation (PRIMARY)
5. International Monetary Fund — Global Financial Stability Report Annual Meetings 2026 (PRIMARY)
6. Reuters — IMF cites AI slowdown, spiking yields as looming global market risks (NEWS)
7. International Monetary Fund — Global Financial Stability Report, April 2026: Global Financial Markets Confront the War in the Middle East and Amplification Risks (PRIMARY)
8. International Monetary Fund — Fiscal Monitor, April 2026: Fiscal Policy under Pressure: High Debt, Rising Risks (PRIMARY)
9. International Monetary Fund — Rising Sovereign Yields Raise the Fiscal Stakes (PRIMARY)
10. International Energy Agency — Key Questions on Energy and AI (PRIMARY)
11. International Monetary Fund — Artificial Intelligence and the Economics of Adjustment (ANALYSIS)
12. International Monetary Fund — New Skills and AI Are Reshaping the Future of Work (ANALYSIS)
13. International Monetary Fund — Artificial Intelligence and Cybersecurity in the Financial Sector (PRIMARY)
14. International Monetary Fund — The Rise of Cyber Events and Digital Fraud in the Financial Sector (PRIMARY)