If the AI Bubble Bursts, Who Runs Out of Medicine First? - Disruption Risk Intelligence, Edition 7

We are asking a narrow question: if the AI bubble bursts, how are health supply chains impacted?

Share

Verolyx · Cross-Sector Disruption Intelligence

Essay · Series opener: Finance and the Medical Supply Chain · 30 October 2026

This is not a prediction

We are not forecasting a crash, its timing or its size. This essay asks a narrower question: what does the historical record say happens to health supply chains when finance breaks, and what would that mean if the next outbreak arrived after, say, the AI bubble bursts? It opens a series of articles, each one examining the impact of an AI bubble burst on one type of health emergency using the same methodology as previous Verolyx Disruption Risk Intelligence editions.

Open almost any financial page this month and the question is the same: is the aggressive and rapid buildout of artificial intelligence infrastructure creating a bubble, and what happens when it pops? In its October Global Financial Stability Report, the International Monetary Fund named an AI market correction as one of three risks that could abruptly tighten financial conditions worldwide [1]. The Bank of England has warned of ‘stretched’ valuations and of AI firms’ increasingly complex and opaque debt [2].

Almost none of that coverage asks the question that matters to anyone who has run supplies into an emergency: if the money stops, which clinic runs out of what, and when?

The answer is not the one most people expect. A financial crash does not usually empty the world’s warehouses. It squeezes cash flows. Restricted and decreased cash flows in the financial system will first impact the countries least able to absorb the next outbreak.

A crash doesn’t empty the warehouse. It empties the wallet.

When speculative booms collapse, the first physical effect on supply chains is usually too much, not too little. After the 2008 crash, world trade fell about 20% relative to GDP [3], the main index of dry-bulk shipping rates fell 94% in seven months [4], and by March 2009 484 container ships sat idle [5]. When the dotcom and telecom boom ended in 2001, Cisco alone wrote off USD 2.25 billion of excess components [6]. Factories, ships and goods were still there. Demand and credit were not.

Look closer, though, and the damage is real, just unevenly spread. In 2008–09 banks pulled back trade finance and letters of credit that exporters rely on, and exports to smaller and poorer markets fell hardest [3]. Health systems feel this kind of shock through payment, not production:

  • Greece, 2010–11. As the sovereign crisis deepened, public hospitals stopped paying suppliers. Roche halted deliveries to hospitals in arrears, and the industry reported that almost nothing had been paid on roughly EUR 70 million of first-quarter 2011 purchases [7].
  • Sri Lanka, 2022. A country importing roughly 80–85% of its medical supplies ran out of foreign exchange. By May 2022, 311 medicines were out of stock, 190 of them essential [8][9].
  • United States, 2023. One over-indebted generic manufacturer, Akorn, was liquidated. Products it alone supplied were discontinued and 14 existing shortages worsened [10].

The pattern holds across very different settings. The medicine exists somewhere. The country, the hospital or the supplier can no longer pay for it, finance it or ship it. That is a different problem from scarcity, and it calls for different early warnings.

Not one crash. A ratchet.

Most discussion of an AI correction imagines a single event, followed by recovery. Low- and middle-income countries rarely experience shocks one at a time. They experience them in sequence, and each one spends a buffer that the next emergency will need.

Count the shocks since 2020: a pandemic; the 2022 rise in interest rates and the currency pressure that followed; the 2025–26 collapse in aid; and the 2026 energy shock from the Strait of Hormuz. The numbers are stark:

  • Official development assistance fell 23.1% in real terms in 2025, the largest single-year drop on record [11]. Aid for health is projected to be up to 63% below its 2022 peak in 2026 [12].
  • The Global Fund’s 2024–26 cycle has been trimmed by USD 1.4 billion [13], and Gavi cut about USD 200 million from vaccine stockpile funding [14].
  • The IMF rates 9 low-income countries as in debt distress and 23 more at high risk [15].
  • Oil was still near USD 100 a barrel in early October, against about USD 72 before the war, partly because of higher shipping and insurance costs [16].

Buffers refill slowly. The global stockpile of oral cholera vaccine fell to 35 million doses at its low point, and preventive vaccination was suspended for nearly four years before restarting in February 2026 [17]. That shortage was driven by demand, not finance, but it shows the timescale: a buffer drained in months can take years to rebuild.

Why this bubble is wired differently

Two features make the AI boom worth watching from a health supply perspective, beyond its sheer size.

First, the debt sits in unfamiliar places. More than USD 120 billion of data-centre spending has been moved off company balance sheets into special-purpose vehicles in under two years, funded largely by private credit [18]. US life insurers held USD 85.2 billion of AI-linked investments by mid-2026, a figure Moody’s says probably understates their exposure [19]. In 1987 and 2020, central banks stabilised markets within days or weeks [20][21]. In 2008 the backstop came later, and the damage spread further. How quickly losses in private credit and insurance would be caught is genuinely uncertain, and history says the speed of the backstop decides how far a shock travels.

Second, the boom is already competing with health for inputs. Memory-chip prices roughly quadrupled in a year, and medical-device makers have asked the US government for help securing chips [22]. Lead times for large power transformers now run to four or five years [23]. Oxygen plants, ventilators and hospitals draw on the same components. A bust would relieve some of this pressure even as it tightens finance: any serious analysis must hold both effects at once.

Three emergencies, three breaking points

WHO’s Disease Commodity Packages are organised by disease for a reason: each emergency needs different products, bought by different payers and moved in different ways. The same financial shock therefore lands differently depending on what is spreading. The rest of this series takes each in turn.

ARTICLE 1 · Respiratory pandemic

What it needs: Vaccines, diagnostics, oxygen, ventilators and respiratory PPE, at global scale and all at once.

Who pays: Rich countries sign advance purchase agreements; poorer countries rely on pooled mechanisms and donations.

Where it breaks: The queue. In 2009, vaccine arrived four to six months after the pandemic was declared [24]; global capacity was about 500 million doses [25]; and rich-country contracts came first. Poorer countries depended on donations, and only a mild pandemic created the surplus that made donations easy [26].

The thought exercise: A severe pandemic after a correction: a smaller donor cushion, finance deciding who stands where in the queue, and oxygen and ventilator makers still competing with data centres for chips.

ARTICLE 2 · Haemorrhagic fever

What it needs: High-grade PPE, infection-control kits, laboratory supplies and, where one exists, vaccine, delivered fast into fragile settings.

Who pays: Mostly donors and international agencies.

Where it breaks: Donor money and air capacity. In 2014, airlines suspended flights to Ebola-affected countries; Médecins Sans Frontières said the shutdowns ‘slowed and hampered the relief effort’ [27].

The thought exercise: An outbreak arriving after aid has already fallen by nearly a quarter, with donors under fresh fiscal pressure and air cargo capacity shrinking in a downturn.

ARTICLE 3 · Cholera

What it needs: Oral rehydration salts, IV fluids, antibiotics, rapid tests, chlorine and water treatment, and oral cholera vaccine.

Who pays: Often national budgets, in hard currency, for low-value, bulky goods, topped up by donors.

Where it breaks: The currency and the unpaid invoice, the Sri Lanka and Greece mechanism. Cholera is already endemic: WHO recorded more than 600,000 cases and nearly 7,600 deaths in its latest annual count [17]. It does not wait for a crisis; it is already there when one arrives.

The thought exercise: A currency slide in an import-dependent country, followed by the rainy season.

What we will test

This series rests on a hypothesis, and we would rather state it plainly than imply it:

The hypothesis

Financial shocks degrade emergency supply in low- and middle-income countries not by creating global scarcity but through payment, currency, credit and donor channels. The damage compounds across successive shocks, and it differs by type of emergency.

It would be wrong if: LMICs’ shortages follow global supply problems rather than their own ability to pay, or if their reserves, aid and stockpiles reliably recover between shocks.

Each of the three articles applies the methodology of our previous Disruption Risk Intelligence editions to one emergency type: the four driver types, explicit evidence tiers, and separate scores for how severe an event is and how hard it hits health markets.

Behind the articles, we are building a register of past episodes and candidate cause-and-effect links, from the 2008 trade-finance collapse to Greece, Sri Lanka and Akorn. After the current demonstration build, that register becomes a test case for Verolyx’s predictive model: can it find the chain from a financial shock to an empty pharmacy shelf without being told what the chain is? We will publish the method as we go, and we welcome challenge, data and cases we have missed.

Markets will settle whether AI is a bubble. The question that matters for health security is whether, when the next outbreak arrives, either during or after a bubble, the countries that need supplies most will still be able to pay for them.

How confident are we?

DOCUMENTED Crashes produce gluts and idle capacity; trade-finance contraction hits poorer markets hardest; payment and currency crises have caused medicine shortages (Greece, Sri Lanka); a single manufacturer’s failure can cause shortages (Akorn); aid, debt-distress and AI-finance exposure figures cited above.

PARTIALLY EVIDENCED That the 2008 crisis weakened the 2009 H1N1 response (the main constraints were vaccine capacity and contracts); that AI demand is reducing medical-device availability, not just raising prices.

STRUCTURALLY PLAUSIBLE That an AI correction would trigger the payment, currency and donor channels described here; that successive shocks compound the severity; and every ‘thought exercise’ in this essay. These are what the series and the model will test.

Sources

  1. IMF Blog, ‘Global Financial Markets Face Rising Risks From AI, Bond Yields, and Cyberattacks’ (Oct 2026). Link
  2. Alliance News via AJ Bell, ‘Rapid AI advances raising financial stability risks – Bank of England’ (2026). Link
  3. NY Fed Liberty Street Economics, ‘No Guarantees, No Trade!’ (2015). Link
  4. Risk.net, ‘Shipping out’ (2009). Link
  5. USNI Proceedings, ‘2009 U.S. Merchant Marine and World Maritime Review’. Link
  6. Cisco Systems Form 10-Q, FY2001 Q3 (SEC). Link
  7. Fierce Pharma, ‘Greek debt pushes drugmakers to cut off supplies’ (2011). Link
  8. The Core, ‘Lanka on life support’ (2022; Auditor General figures). Link
  9. Reuters via Ada Derana, ‘Sri Lankan medicine shortage a death sentence for some’ (2022). Link
  10. Becker’s Hospital Review, ‘Akorn: From financial troubles to a shutdown’ (2023). Link
  11. Donor Tracker, ‘Beyond the headline: Gavi 6.0 pledges’ (2026). Link
  12. OECD, ‘ODA projections for 2026 and the near-term’ (June 2026). Link
  13. Frontiers in Health Services (2026), donor cuts and India’s medicine supply. Link
  14. Rethink Priorities, ‘Gavi’s Marginal Cost-Effectiveness in 2026’. Link
  15. IMF, List of LIC DSAs for PRGT-eligible countries (as of 31 March 2026). Link
  16. Al Jazeera, ‘Hormuz ship attacks surge’ (6 Oct 2026). Link
  17. AP via WTOP, ‘WHO announces restart of preventive cholera vaccinations after nearly 4-year halt’ (Feb 2026). Link
  18. Quinn Emanuel, ‘Emerging Litigation Risks in Financing AI Data Centers Boom’ (2026). Link
  19. Reinsurance News, ‘US life insurers increase AI investments… says Moody’s’ (23 Oct 2026). Link
  20. Federal Reserve History, ‘Stock Market Crash of 1987’. Link
  21. Kansas City Fed, ‘G-Spread Suggests Federal Reserve Restored Calm’ (2020). Link
  22. Becker’s Hospital Review, ‘AI chip shortage threatens medical device supply’ (2026). Link
  23. AI Weekly summary of Financial Times on transformer lead times (2026). Link
  24. WHO Regional Office for Europe press release, 15 September 2009. Link
  25. Reuters via Fox News, ‘Countries close to deal on flu virus sharing, vaccines’. Link
  26. Kumar et al., ‘US Public Support for Vaccine Donation to Poorer Countries in the 2009 H1N1 Pandemic’ (PMC). Link
  27. Fortune, ‘Halting flights to Ebola regions could threaten relief efforts, experts warn’ (2014). Link