Poland is building Europe’s AI workshop. But geography, the OECD warns, decides who gains
Building the data centres may prove the easier task; building the companies that capture what those servers produce will decide whether Poland becomes not only Europe’s AI workshop but one of its technology powers.
PARIS – The biggest economic divides in the OECD are no longer between countries; they are increasingly inside them. That is the starting point of the OECD’s Employment Outlook 2026, published on Tuesday. Across the developed world, employment gaps between regions within the same country now exceed the differences between OECD countries themselves: unemployment in the worst-performing fifth of a country’s regions runs, on average, more than twice as high as in the best-performing fifth, and more than four times as high in Belgium, Canada, Italy and Slovakia. Capital regions keep drawing investment, talent and higher-paid work while weaker regions struggle to keep pace, and the report argues that artificial intelligence, like globalisation before it, risks deepening those divides unless governments deliberately steer new industries beyond the places where capital already wants to go.
Poland arrives at that moment from an unusually strong position. Employment is still rising while much of the OECD has begun to level off, and productivity has grown faster than its long-term trend, placing Poland alongside the United States among the few OECD economies to outpace their own historical rates. It has also become one of Europe’s fastest-growing destinations for AI infrastructure: last year developers announced more than $6bn in AI-related investment, and nearly 70 proc. of the country’s data-centre capacity now sits in and around Warsaw.
The build-out is already visible outside the capital: Switch is putting up what will be Poland’s largest data centre, a 90-megawatt facility in Mory, near Warsaw, while Data4 and Microsoft’s Azure Poland Central have anchored capacity across Mazovia and Google’s 2,000-strong Warsaw hub is its largest engineering centre in Europe. Further south-west, LG’s complex near Wrocław helped make Poland the EU’s largest battery exporter, and KGHM supplies almost half of Europe’s copper. But look at the map and most of it clusters around Warsaw and the industrial south-west; eastern Poland barely appears – a divide the report names directly, describing an east-west gradient in which its high-unemployment clusters sit in a less-developed east, long reliant on small-scale agriculture and marked by lower levels of schooling.
That geography is exactly what the OECD worries about. The report treats AI, alongside trade, the energy transition and demographic change, as a structural force reshaping labour markets unevenly: places already rich in infrastructure, research capacity and skilled workers tend to attract still more investment, while regions that fall behind often lose their youngest and best-educated residents first, making recovery harder rather than easier. The market rarely corrects the imbalance on its own. Workers who leave struggling regions are disproportionately young and highly educated – across the OECD, someone with education beyond secondary school is at least five percentage points more likely to leave a low-employment region than someone without – so the places most in need of skilled workers tend to lose the very people most able to attract new investment.
Poland’s boom illustrates both the opportunity and the dilemma. The location is not accidental: cheap land, growing power capacity and a cool climate – Warsaw averages about 8.9°C, against 12.9°C in Madrid and nearly 18°C in Athens, the Polish Data Center Association notes – trim the industry’s biggest operating costs. Yet the buildings consume far more electricity than labour. A hyperscale campus can draw hundreds of megawatts, the output of a medium-sized power station, while employing relatively few people once built. The benefits are real – construction work, power sales, property tax – but much of the lasting value created by AI is captured elsewhere. The OECD frames that gap as one between a country’s regions; it runs between countries too, and hosting the activity is not the same as gaining from it.
Goldman Sachs estimates cumulative global AI investment of roughly $7.6trn between 2026 and 2031, much of the return expected to accrue not to whoever owns the buildings but to whoever owns the chips, cloud platforms and foundation models running inside them. Hosting servers creates construction work, demand for electricity and industrial supply chains; designing semiconductors, training language models and selling software generate the profits that compound over decades.
Poland is trying to own that layer too, not just host it. Warsaw leads a four-country consortium seeking to build the €3bn Baltic AI Gigafactory, and the government finances two Polish large-language models, PLLuM and Bielik, framing AI as part of a broader European technology strategy. Deputy Digitalisation Minister Dariusz Standerski describes Poland not simply as a customer but as „a co-architect” of Europe’s AI future. He is also candid about the missing piece: private investment in Polish AI firms remains modest by European standards, which he attributes to the same caution that once made Polish companies slow to adopt cloud computing. Public money has begun to flow; private capital has not yet followed.
Poland’s neighbours offer different answers to the same question – whether hosting the boom is enough to gain from it. Hungary – the neighbour whose economy most resembles Poland’s own – made the boldest bet on hosting. Courting first Korean and then Chinese manufacturers, it became the world’s second-largest recipient of Chinese electric-vehicle investment, the centrepiece a €7.3bn CATL battery plant near Debrecen. The factories reshaped the region, but the profits flow mainly to their foreign owners, much of the workforce is on temporary agency contracts, and because the deals demanded no transfer of technology, Hungary risks being stranded in the low-value assembly links of a supply chain it does not control. The IMF reckons the shift could cost it around 1 proc. of GDP over five years. The bet proved politically costly, too: the backlash helped end Viktor Orbán’s sixteen years in power at this year’s election, bringing in Péter Magyar’s Tisza party on a promise of tighter oversight.
Türkiye has taken almost the opposite tack. Rather than court foreign manufacturers, Ankara has spent a decade building nationally owned champions – Baykar’s drones, the TOGG car, Aselsan’s defence electronics – and President Recep Tayyip Erdoğan now wants to extend the model to computing, pledging some $3bn in public money to catalyse more than $10bn of investment in what he calls „sovereign AI”. Whether Türkiye can afford the ambition is another matter. Consumer prices were still rising at 32.6 proc. in May, the highest rate in the OECD; growth has slowed for three straight quarters; and analysts, among them the Soufan Center, argue that a fraying rule of law – sharpened by the jailing of Istanbul’s opposition mayor – keeps deterring the foreign capital Ankara still needs.
Even countries that already own the technology are not exempt. Israel designs and exports the silicon at the heart of the AI boom, yet the analyst Amir Mizroch warns that „Startup Nation” risks becoming „Subcontractor Nation” if its engineers build the value into products owned elsewhere. Its public finances are strained, too: nearly three years of war have held the budget deficit above 5 proc. of GDP for a third year running and pushed public debt from around 60 proc. to 70 proc., prompting one recent analysis to warn that Mr Netanyahu’s government is on a „dangerous debt path”. Ownership, the Israeli case suggests, is not only about invention but about keeping enough of the surrounding ecosystem to capture what invention earns.
That is where the OECD parts company with a purely market-driven approach. Governments, it argues, should think not only about moving workers to jobs but about moving jobs to workers – aligning industrial strategy with local strengths, extending infrastructure beyond the metropolitan hubs, and encouraging diversification rather than dependence on a single industry.
Canada offers one model, directing AI investment towards regions that already have research institutions and technical talent; the EU’s critical-minerals strategy likewise ties industrial policy to regional skills and infrastructure.
For Poland, the OECD’s question is not whether the AI boom arrives – it has – but who gains when it does. That turns on whether the country’s role stays that of Europe’s preferred host for hyperscale infrastructure, or whether more Polish firms come to own the software, models and platforms running inside those buildings. Building the data centres may prove the easier task; building the companies that capture what those servers produce will decide whether Poland becomes not only Europe’s AI workshop but one of its technology powers.
