Poland’s Shrinking Labor Cost Advantage Is Not an AI Growth Engine

Jaan Roose walked a slackline across the 500-metre gap between Warsaw’s Varso tower and the Palace of Culture and Science, Stalin’s gift to Poland. Natalia Kasperska, of HB Reavis, which owns Varso, called it “a symbolic walk from the past to the future.” From a higher floor, her colleague Cezary Błaszczyk was blunter: “We look down on it.” The 310-metre spire is billed as the tallest in the EU, and he has a reading ready for the view: “Since we are higher and taller than the Palace of Culture, we treat it as: ‘We got over communism, and now we are a country with a strong economy.’”

A skyline is not a model. The boom that filled those offices ran on skilled labour that cost less than the same work in western Europe. That discount is closing fast enough that a 2016 business case needs a new price, not a toast. The AI alternative being waved at the gap (a state stake in a voice company, a startup-access programme, a fund described in the billions) is not a substitute engine.

If you are placing work in Poland, reprice the labour. If you are buying AI for business there, name the hour you mean to remove before you applaud the press line.

A shrinking discount is not a vanished one

The bargain was simple. After EU accession in 2004, Poland had people who could do the work, inside the single market, at wages the west would not match. Volkswagen, Cadbury, and JP Morgan were among the employers drawn in. For a shared-services centre, the spreadsheet did the selling.

Rafał Brzoska, founder of Kraków-based InPost, has already called that edge dead.

“The next phase will depend less on catching up, and more on productivity. We don’t have labour cost advantage at all. We need to focus on productivity, technology, cheaper energy and better regulation.”

Argue with the middle sentence only with a figure you can stand behind. Eurostat’s hourly labour costs release for 2025, updated in March 2026, reports that Poland’s hourly labour costs in zloty rose 8.8% from 2024 to 2025. Among non-euro-area changes in that release, Romania was faster at 10.6% and Hungary a shade faster at 8.9%. Denmark rose 3.0%. In euros, hourly labour costs rose 4.1% in the EU and 3.8% in the euro area. Different currency, so not a clean comparison, and still a slower pace. Eurostat’s labour cost means what a finance director means: wages and salaries plus non-wage costs such as employers’ social contributions. The estimates cover firms with 10 or more employees.

That is fast. It is not “no advantage at all.” A level sometimes quoted beside this story, about €19 an hour and said to be half the German and French cost, is not in the figures behind that 8.8% change, and it will not be treated as one. Shrinking is not the same as gone. What an operator can use is the pace. A centre still justified by a 2016 salary gap is leaking its reason for existing, whether or not anyone has published a fresh euro comparison you would bet a contract on.

The voice comes with a kill switch

The company that makes the AI claim concrete is ElevenLabs, an audio-generation and voice-cloning firm named for 11 November, Poland’s independence day. Generated speech can cheapen support lines, training audio, and localised sales calls. It is also a consent problem and a fraud surface.

The deployment worth paying for is dull on purpose. Disclose that the voice is synthetic. Keep a permission chain you can show a customer or a regulator. Log the prompts. Put a human on anything that can move money or reputation. And keep a way to shut the system off when it starts sounding like the finance director on a call it should never have made.

None of that requires a national champion. It requires a buyer who knows which hour of labour they are trying to remove, and what a mistake in that hour costs.

One hour, priced properly

AI agents lean on the same fact that made Warsaw attractive. A team there can already serve a client in Chicago before lunch. Software can now draft the repeatable half of that work without a relocation package.

Take one hour that sits inside a lot of Polish business-service work: the first-pass sales follow-up after a demo. It is specific, frequent, and already written down. So is the weekly status note. So is the opening reply on a localised support ticket. A centre whose product was doing that hour cheaper than Frankfurt is now bidding against two things at once: rising zloty labour costs, and a tool that produces a usable first draft.

Hold the 8.8% steady as arithmetic, not as a forecast. A labour line written on 2024 rates is about 18% higher after two years and just under 29% higher after three, before the currency even moves. The nearshore advantage does not die on demo day. It thins every time the task is a template. Teams that only sold the template are exposed. Teams that sit on messy exceptions, customer judgment, or a physical network can use the same AI automation to raise output per hour. Those are different businesses. They should not share a strategy slide, and a deck about AI for sales does not make them the same business.

InPost is the second kind, minus the myth. Parcel lockers are won on drops per hour, failed deliveries, and energy. An agent that flags lockers about to overflow, or that ends a support call without inventing a refund, changes output on a physical network. That can earn its place. A valuation cannot.

Brzoska has already warned against looking for a shortcut around that distinction. “People may put trust in someone else that has a magic remedy.” There is, he said, no magic remedy, and no secret sauce that will magically change things. “In my opinion, unfortunately for all of us, we still need to work hard.”

Eleven million is a flag, not an engine

Poland has been presented as pushing to become a regional tech hub, backed by a $1 billion public-and-private fund for tech startups and AI companies. A companion reference points to Innovate Poland and “at least PLN 4 billion.” At recent rates, PLN 4 billion is roughly $1 billion. That is not a reconciliation. The open question is whether those phrases describe the same vehicle, and whether either sum is actually being deployed.

The step described in deal terms is smaller, and it is not the same thing. Tech Funding News reported that Poland’s state-backed Vinci fund invested $11 million in ElevenLabs, making the government a direct shareholder in what the post called an “$11 billion AI voice company.” The same account says the deal launches AI Lab Poland, offering early-stage AI startups funding, mentorship, and access to ElevenLabs’ investor and enterprise network.

The $11 billion figure is a reported private valuation. That account does not date the round, name the method, or point to a filing. Treat it as unaudited. Eleven million dollars into a company already discussed at eleven billion can open doors for younger founders. It does not show that a billion-dollar pool is at work, and it does not show that AI automation has replaced the labour-cost model that carried the catch-up. A shareholding and a startup-access programme can be useful without being mistaken for a growth engine.

Ambition is not the capacity plan

Michał Nowakowski, of CD Projekt Red, the studio behind The Witcher and Cyberpunk 2077, does not think the old brake is still on. “For young people now they have no inferiority complex. They don’t stop and think: ‘Can we be as good as Brits, French, Spanish?’ It’s not holding them back.” He has put it more flatly: “We’re a proper-sized country. Not too small, not too big; no inferiority complex any more. There’s nothing holding us back.”

Talent, and the loss of that complex, are real. The useful pattern is the studio itself: it began in the 1990s importing CD games from American publishers, and learned to export its own. Buying seats on someone else’s model is consumption. People who can wire a model into a locker network, a factory line, or a production pipeline are the production version of a tech hub.

Ambition is still not a capacity plan. Nowakowski credits universities and skills with a shot at “unprecedented heights.” Brzoska’s list is the answer to that confidence. Productivity, technology, cheaper energy, and better regulation have to move together, in live operations, by people who keep doing the unglamorous work. Confidence does not clear a permit, cut a power bill, or own an error.

Questions worth asking first

Has Poland actually lost its labour-cost advantage?

Not on the evidence worth underwriting. Eurostat’s March 2026 update shows zloty hourly labour costs up 8.8% from 2024 to 2025, behind Romania, a shade behind Hungary, and well above Denmark’s 3.0%, so the cushion is shrinking, not proven gone.

Does the ElevenLabs stake mean Poland has an AI growth engine?

No. Vinci’s $11 million, as Tech Funding News reported it, is a small public shareholding in a voice company with a reported, unaudited $11 billion private valuation, plus an access programme for early-stage startups. A described $1 billion fund, a sum that can sit near “at least PLN 4 billion” at recent rates, is a different claim until it is shown to be the same vehicle and to be deploying.

Will AI agents replace Polish nearshore teams?

Not as a single event. Pressure hits templated work first (a first-pass sales follow-up, a status report, the opening of a localised support reply), and a centre that only sold that hour cheaper is exposed as zloty labour costs rise. Operations with judgment, exceptions, or a physical network can use the same tools to raise output instead.

Is generated voice a safe productivity tool?

Only with a permission chain, disclosure, logged prompts, a human on anything that moves money or trust, and a way to turn it off. Audio generation can cheapen support and localisation; a cloned voice without clear rights is a liability, not an automation win.

Which hour are you removing, what does an error cost, and who owns it?

If you cannot name all three, you are buying a demonstration. Brzoska’s verdict still fits: there is no magic remedy. Productivity, technology, cheaper energy, and better regulation have to move together, and the work is still the work.

One comparison the tables will not make

A national income story still needs a number you can stand behind, and the usual world total is not one of them.