Under what conditions will Türkiye use AI, in which layers will it remain externally dependent, and who will retain control in critical domains? Türkiye’s primary objective should be to establish a sovereign usage regime, especially for priority and regulated sectors and for state institutions.

Türkiye is not starting from zero. TÜBİTAK ULAKBİM’s ARF-ACC system within the TRUBA infrastructure rose to 145th place in the global TOP500 rankings by the end of 2025. Its capacity increased substantially with the addition of 192 NVIDIA H200 GPUs in 2025. Türkiye is also a partner in the BSC AI Factory initiative. The 2026–2028 Medium-Term Program explicitly identifies expanding TRUBA’s capacity and developing a Turkish large language model as core goals.

In the short term, it does not seem realistic for Türkiye to build a model-training infrastructure on the scale of the United States or China. But that does not mean Türkiye cannot establish a sovereign usage regime suited to its own conditions. The goal should be to ensure that Türkiye can manage the data, security, cost, continuity, and service conditions of the AI systems used in critical areas.

Access to advanced computing capacity is not merely a market issue. Today, access to advanced chips is still not independent of politics and geopolitics. For that reason, it would be a mistake for Türkiye to assume that it can simply buy chips and AI services from the market whenever needed.

So who would build such an infrastructure? Türkiye does not have two giant private-sector players like Russia’s Yandex and Sber. By contrast, public influence is stronger in the telecommunications and infrastructure backbone. Turkcell, Türksat, and Türk Telekom, all within the portfolio of the Türkiye Wealth Fund, provide a critical base in terms of data centers and connectivity infrastructure. The realistic option is an infrastructure coalition that combines public direction with private-sector operating discipline. For example, a strategic AI partnership similar to TOGG, publicly supported but not directly run as a state enterprise, could make sense.

The role of such a structure could primarily be to provide secure inference infrastructure for the public sector and regulated industries, Turkish-language model adaptation, enterprise solutions based on open weights, and local data sovereignty.

Of course, there is also a serious governance risk here. Such a structure could easily turn into a black hole that consumes resources, expands its mandate uncontrollably, and delivers uncertain performance. That is why success would require a narrow and measurable mission definition, clear access rules, multiple suppliers, cost transparency, and strict performance auditing. The measure of success should not be rhetoric, but results.

For Türkiye, the objective should be to move toward sovereign use through assured access in model training, local capacity in inference, portability in software, and strategic selectivity in public procurement. This is certainly a difficult path, but engineering innovations that reduce model training costs, the open-model ecosystem, software portability, and inference infrastructure may also open a window of opportunity for Türkiye.