Polcom to expand sovereign AI cloud. Data Center 3 in Skawina to reach 40 MW

The Polish operator Polcom has announced an expansion of its Polcom AI Cloud platform, in which the infrastructure, models and training data are to fall under Polish jurisdiction. It is an attempt to claim a place in a segment that until now has been served mainly by the big American clouds and the largest Western European data centres.

The scale of the investment is concrete. Polcom has already spent PLN 42 million on modernising its existing IT and power infrastructure. The next stage is to be the Data Center 3 facility in Skawina, with a capacity of up to 40 MW, a budget of around PLN 150 million and a planned launch in 2030.

Sovereignty as an argument, not a slogan

The declaration of Polish jurisdiction over data and models fits into a broader European debate about digital sovereignty, which European industry media have been covering over the same period. Alongside the talks about regulation, there is a contest over who physically controls computing power and training data.

The geopolitical context completes the picture. The rp.pl service reports that artificial intelligence has become one of the main topics in US-China relations. During a meeting between Donald Trump and Xi Jinping, AI appeared among the subjects of discussion alongside security, Taiwan and technology, and both leaders spoke about responsible management of AI development. In the same text, rp.pl points to an opportunity for Poland in the semiconductor chain and to the dispute between France and Germany over the "Made in EU" rule.

A proving ground for models

For the Polish market, a managed model means one thing. If open-weight models such as DeepSeek-V4.1-Flash under the MIT licence or Xiaomi's open models are to be run in the country and under national jurisdiction, a local infrastructure layer is needed. Demand for capacity is enormous on a global scale. The French outlet Le Monde Informatique writes, citing a Gartner forecast, that global spending on AI is to reach USD 2.7 trillion in 2026, or 49.5 per cent more than a year earlier, with most of it going to infrastructure and software.

Whether an investment of around PLN 150 million and a 40 MW facility in 2030 will be enough to mean anything at that scale remains an open question. Whatever the answer, it is one of the few projects in Poland that directly addresses the whole stack, from capacity through models to data jurisdiction.

The business model of such a platform rests on two pillars. The first is the model layer: with open weights under the MIT licence, such as DeepSeek-V4.1-Flash, the operator does not pay for the right to use the model itself; it pays for capacity, cooling and support. The second is the compliance layer: public institutions and companies from regulated sectors need access logs, data location and data processing agreements, which global clouds do not always provide in the expected form.

The risk is scale and timing. A 40 MW facility launched in 2030 will enter a market where the largest players are building clusters with capacities in the hundreds of megawatts today, and successive generations of accelerators change the profitability calculation every few quarters. A sensible strategy for smaller operators therefore usually consists not in chasing training volumes but in handling inference, sensitive data and integration with systems that cannot be moved to the public cloud.