On September 10, 2026 Pablo Hernández de Cos, General Manager of the Bank for International Settlements (BIS), spoke at Global Fintech Fest Mumbai on AI's effect on growth and financial stability risks: from a productivity leap to valuation concentration, "circular" financing and cyber risk.
On September 10, 2026 Pablo Hernández de Cos, General Manager of the Bank for International Settlements (BIS), gave a speech at Global Fintech Fest Mumbai titled "Artificial intelligence, growth and financial stability: challenges for central banks". This text is a summary and rewrite of the speech's main points.
On growth: he said AI shapes demand through a massive investment wave in data centers, semiconductors and the cloud (projections of roughly $3 to $4 trillion by 2030). Micro-level productivity gains are estimated at 10 to 65 percent, and total factor productivity (TFP) growth could potentially rise by about 0.5 percent a year. But the benefits are concentrated in advanced economies and highly exposed sectors such as finance and professional services, while emerging markets lag because of lower readiness. Labor displacement has also begun in customer support, programming and clerical jobs.
He also pointed to the concentration of the AI supply chain: the US, China, Taiwan, South Korea and the Netherlands dominate across five layers — semiconductors, cloud, data, foundation models and applications — and the five big AI companies made more than $1 trillion in capital expenditure (capex) in 2025 and 2026.
The financial stability risks he listed: (1) equity valuation concentrated in a few AI giants; (2) herd behavior and overinvestment, with capital spending outpacing cash flows; (3) opaque financing through private debt and "circular" arrangements in which counterparties hold each other's shares and are hard to track; (4) a domestic asset bubble driven by AI-related export revenue; (5) contagion from US equity concentration to the rest of the world, amplified by household-held shares; and (6) cyber risk, where AI tools help attackers asymmetrically — because an attacker needs only one vulnerability.
He compared the situation with historical technology waves — canals in the 1830s, railways in the 1840s, electrification in the 1920s and the dot-com boom of the 1990s — all of which attracted more capital than their eventual returns. The comparison is a warning about the pattern, not a firm prediction of this wave's fate.
His policy recommendations for central banks: keep monetary mandates unchanged but strengthen measurement, judgment and policy agility; accept more uncertainty about potential output, natural rates and transmission; pursue international cooperation for shared learning; and invest in digital infrastructure, skills, competition, data governance and sound institutions — so the benefits of AI spread more widely across firms, workers and countries.
For technology leaders in financial institutions, three messages can be drawn from the speech (our reading, not the BIS's words): supply-chain concentration means supplier-dependency risk must show up in architecture and contracts; AI-amplified cyber risk means monitoring, incident response and operational resilience become priorities; and readiness — data, skills and governance — is what decides whether an organization benefits from this wave or just pays its cost.
This note is a summary and rewrite based on the report below; see the original source for the full report.
Bank for International Settlements (BIS)The 2026 report by the Cambridge Centre for Alternative Finance (CCAF), with BIS, IMF, WEF and others, shows 81% of financial firms use AI and software engineering is the most mature use — yet concern over lost human oversight and weak explainability is serious.
According to a SiliconANGLE report on September 18, 2026, a week that saw multi-billion-dollar AI infrastructure investments alongside safety tools to rein in rogue agents — the AI-agent industry's paradox.
Research by Medrano, Verma and Chhabra shows that in a production-like RAG system, RAG Fusion raises recall, but after re-ranking and context truncation much of the gain disappears and precision even dips slightly.
