Central banks have stopped watching AI from the sidelines

The forecasters get a rival

The Czech National Bank's AI model now forecasts GDP almost as well as its economists. The average error of the human economists was 0.21 percentage points, while the AI model error was 0.22, Governor Aleš Michl told a lecture at Kyoto University, and Michl predicts that AI models will soon outperform human economists in GDP forecasting. For a central bank, that is not a gadget. The forecast is the job. m:…2026

The bank built the machinery deliberately. It created the CNB Lab in 2025 to test AI, tokenization, blockchain, stablecoins and bitcoin, launched a test portfolio of 1 million dollars in November 2025, and is testing a digital bond on a digital ledger. Michl calls the approach conservative innovation: monetary policy stays traditional while the experiments live in a separate lab. m:…2026

The bank earned the right to experiment by getting inflation down first. Inflation in the Czech Republic was 17.5 percent in mid-2022, and the bank achieved its 2 percent inflation target in February and March 2024. Its key interest rate is 4 percent, and it reduced its senior managers from 17 to 14 along the way. The lesson Michl draws is that a central bank can modernise its tools without loosening its grip. m:…2026

What the Fed's own research actually finds

The Federal Reserve's own research paper on AI and the US economy draws a cooler picture than the headlines. Investment in data centres has climbed sharply, driven by current and expected future demand for AI services, and AI model-producing firms now account for a noticeable fraction of software industry gross output. The buildout is real, and it is concentrated in the inputs: prices have climbed in sectors that provide direct inputs to data centres, and memory chip prices have climbed. m:…7864

But adoption is shallower than the spending suggests. AI adoption is growing rapidly among consumers and among firms, yet most firms are currently not using AI very intensively or at all. The effects of AI on the labour market appear small so far, and mixed, and the effects on aggregate productivity growth appear small so far as well. m:…7864

The paper's sharpest point is about timing. The recent pickup in productivity likely reflects factors that predated the diffusion of generative AI. In other words, the productivity gains being celebrated may belong to an earlier story, and AI's own contribution has not shown up in the aggregate numbers yet. m:…7864

The productivity gap behind it all

The governor of the Central Bank of Ireland, Gabriel Makhlouf, put the structural problem in numbers. US labour productivity growth fell from 2.7% per year in 1995-2007 to 1.3% in 2007-2019, and European labour productivity growth fell from 1.3% to 0.5% over the same periods. The US invests approximately 3.5% of GDP in research and development, compared with 2.2% in Europe. m:…8976

Makhlouf argues that European capital markets remain fragmented and reliant on bank debt, which is ill-suited for financing risky innovation, while Europeans hold approximately 12 trillion euros in savings and cash deposits that could be unlocked for productive investment. Industrial policy is no longer a taboo subject, he says, and he advocates for resilient and secure openness over open strategic autonomy. The through-line across all three central banks is the same: AI is no longer a technology story they watch from outside. It is a variable in their forecasts, their financing data and their inequality warnings, and they are now writing about it in their own words. m:…8976

The Fed's Barr sounds the warning. The labour share of national income reached a record low of 52.8 percent in the second quarter of 2026, and wealth concentration has increased at the very top, with the top 1 percent holding about one-third of all wealth. Barr cautioned that while AI could boost productivity, it risks widening inequality if market power becomes overly concentrated, with adoption favouring capital owners over labour. m:…4155