Rising with Algorithms: Serbia in the Global Race for AI leadership

Introduction

“AI will take our jobs!”

“Humans will be replaced by robots!”

“Artificial intelligence has become self-aware!”

Sensational headlines like these pound our newsfeeds every day. Yet beyond the clickbait lies undeniable momentum: artificial intelligence is already streamlining routine tasks and sparking scientific breakthroughs, detecting tumours on MRI scans months before expert radiologists, designing new molecules, even helping decode the language of proteins. Its progress is breathtaking; what felt state-of-the-art last year can seem quaint today.

Change at this speed naturally breeds anxiety. But history shows that bold societies turn disruption into advantage. So the real question isn’t whether AI will reshape our world, but how a small, inventive nation like Serbia can contribute to shaping that future instead of merely reacting to it. With a deep mathematical tradition, strong engineering talent, and a fast-growing start-up scene, Serbia may be much further ahead in this race than many realize.

To grasp the current AI race and Serbia’s place in it, we first need to return to the very origins of artificial intelligence. Understanding each stage of the field’s development is essential for making informed predictions about the next phases and the steps that ought to follow.

How can a computer be intelligent?

Although for most people, the concept of AI originated in 2022 with the release of ChatGPT, the very essence of the idea that computers can do things on their own, without exact instructions from a person, dates back to the 1940s and Alan Turing. Turing gave quite possibly the earliest public lecture (London, 1947) to mention computer intelligence, saying, “What we want is a machine that can learn from experience,” and that the “possibility of letting the machine alter its own instructions provides the mechanism for this.”

The main problem Turing faced – there was no computer on which he could program anything back then.

He proposed the so-called Turing test, which holds that a computer may be considered intelligent if an interrogator cannot distinguish between the computer and a human respondent. It has been claimed that ChatGPT-4.5 passed this test, but the scientific community remains skeptical.  Judge for yourself whether you can tell the difference between a human and ChatGPT.

In 1956, there was a conference called The Dartmouth Summer Research Project on Artificial Intelligence at Dartmouth College, in Hanover, New Hampshire. The workshop has been referred to as "The Constitutional Convention of AI". A group of researchers gathered to discuss the theoretical aspects of computers’ ability to be intelligent.

These “founding fathers” of artificial intelligence even coined the term itself. Some of them developed Logic Theorist, a program designed to prove theorems from the landmark work Principia Mathematica. In certain cases, the proofs generated by the program were more elegant than those published in the books.

In 1966, MIT’s Joseph Weizenbaum created ELIZA, a program that mimicked a therapist. Six years later, Stanford’s Kenneth Colby built PARRY, which simulated a paranoid patient so convincingly that psychiatrists often couldn’t tell it from a real one. Despite the novelty, both chatbots were essentially hard-coded: their replies were pre-written and triggered by simple pattern matching, far from the adaptive intelligence of today’s AI systems.

The rise of domain-specific “expert systems” produced the computer program MYCIN (1972), which prescribed antibiotics for blood infections, based on the patient’s data. Yet it lacked commonsense limits—MYCIN would still hunt for bacterial causes even if a gunshot victim was bleeding to death.

There have been efforts to develop systems capable of genuine awareness and understanding of the real world. Beginning in 1984, the Cyc project tried to endow AI with real-world awareness by encoding millions of commonsense rules—letting it infer, for example, that “Garcia is wet” from “Garcia finished a marathon.” Despite its enormous knowledge base, Cyc still falls short of human understanding and struggles to manage its ever-expanding data.

However, some researchers embraced a fundamentally different strategy: instead of encoding explicit, human‐crafted rules, they sought to emulate human thought processes—and that’s precisely where neural networks emerged.

Neural networks: teaching computers to think

Although neural networks are the driving force behind today’s AI breakthroughs, their underlying concepts date back decades.

It was as early as 1954 that researchers at MIT succeeded in running the first artificial neural network, albeit limited by computer memory to no more than 128 neurons. They were able to train their networks to recognize simple patterns. In addition, they discovered that the random destruction of up to 10 percent of the neurons in a trained network did not affect the network’s performance—a feature that is reminiscent of the brain’s ability to tolerate limited damage inflicted by surgery, accident, or disease.

In a 1986 study, researchers Rumelhart and McClelland trained a two-layer neural net (920 neurons) on English verbs. Afterwards, the network could correctly generate past tenses for unseen verbs—e.g., guard → guarded, weep → wept—showing genuine generalization, though it still produced occasional mistakes.

One of the first commercial uses of neural networks, and AI in general, came in 1989, when Yann LeCun and his team at AT&T Bell Labs developed a neural network that could recognize handwritten ZIP codes. The system helped automate postal services, demonstrating the potential of neural networks.

It’s worth noting the cyclical nature of AI development: periods of intense optimism are often followed by disillusionment when the technology fails to live up to investors’ and the public’s expectations. Progress has rarely been linear, and development routinely dips into so-called “AI winters.” This should be taken into account with regard to the most recent AI achievements; while short-term expectations may go unmet, the long-term potential and implications remain profoundly promising.

Final breakthrough

Looking back at the development of artificial intelligence, it is clear that the concept was initially shaped by pioneering academic research at universities across the United States. The turn of the 21st century marked a pivotal shift, as AI began transitioning from theoretical exploration to practical, commercial application.

In the early 2000s, AI made strides in natural language processing. In 2010, Apple introduced Siri, a voice-controlled virtual assistant. In 2011, IBM’s Watson drew global attention by winning the game show Jeopardy!, showcasing AI’s ability to understand and respond to human language.

Meanwhile, in 2020, Tesla released its Full Self-Driving (FSD) Beta system. It enabled vehicles to autonomously navigate streets and highways, though with driver supervision, marking a major step toward autonomous transportation.

Rapid data growth and steady research progress laid the foundation for today’s generative AI. In 2021, OpenAI introduced DALL·E, a generative model capable of producing detailed images from text prompts, marking the rise of creative AI tools now used by millions. In early 2024, Google launched Gemini, a powerful language model designed to handle extended context, advancing long-form understanding.

OpenAI soon followed with Sora, a text-to-video model that can generate up to one-minute video clips from textual input, pushing generative AI into dynamic media creation. In May 2024, Google DeepMind expanded AlphaFold to help detect cancer and genetic diseases, bringing AI into practical, real-world healthcare.

Perhaps most notably, ChatGPT, powered by OpenAI’s GPT-4, became a widely used AI assistant—capable of writing, coding, explaining, and conversing across languages and topics. These innovations signal a shift: AI has moved beyond research labs into tools we now use, see, and rely on every day.

The US as the center of AI development and investment

Since the early days of artificial intelligence (AI), the United States has consistently played a central role in shaping the field through groundbreaking research and innovation. Today, that leadership is not only evident in scientific contributions but also in financial power, global reach, and infrastructural capacity.

The scale of AI investment demonstrates US dominance. According to a joint report by PitchBook and McKinsey, from 2018 to the third quarter of 2023, AI companies in the European Union and the United Kingdom raised approximately €32.5 billion. In stark contrast, US-based AI firms raised over €120 billion in the same period, almost four times more. This investment disparity continues to widen.

In 2023 alone, AI venture capital investment in the United States reached $68 billion, compared to only $8 billion across Europe, approximately 12% of the US level. More recently, between February and May 2025, North America received 86% of global AI funding, while Europe’s share dropped from 18% to 12%. This shift is largely attributed to the size and frequency of large US-led funding rounds.

The graph below shows the proportion of investment in AI in different countries for the period 2013-2024, making it obvious that the key players are the United States.

Furthermore, even outside of the US, American venture capital plays a crucial role in supporting the global AI ecosystem. European AI startups frequently rely on US-based investors to close large, late-stage funding rounds, filling a critical gap in the European investment landscape. This dependency means that US financial influence extends well beyond its borders.

Beyond investment, the United States maintains strategic advantages in AI-enabling infrastructure. American cloud service providers, such as Amazon Web Services, Google Cloud, and Microsoft Azure, host over 70% of Europe’s AI computing capacity. Even EU supercomputing initiatives often rely on US-funded or US-designed technology components.

In July 2025, Oracle announced a $2 billion investment in AI and cloud infrastructure in Germany, aiming to support the growing demand for AI services in Europe. While the investment is local, it underscores how American companies continue to drive both capital and technological expertise behind Europe’s AI infrastructure, further reinforcing US leadership in the global AI ecosystem.

This infrastructural leadership is set to expand further. A megaproject dubbed Stargate, launched in 2024, is expected to inject up to $500 billion into US AI infrastructure over four years. This initiative will support the construction of high-performance data centers and strengthen the foundational computing backbone of the AI ecosystem, largely reinforcing American dominance.

China: A Rising but Limited Challenger

China is frequently cited as the United States’ main competitor in AI. The Chinese government has invested heavily in AI research, infrastructure, and national strategy. Companies such as DeepSeek have made notable advancements in large language models and multimodal AI systems.

However, China continues to trail the United States in both research leadership and investment scale. Most Chinese AI advancements follow US breakthroughs rather than precede them. Furthermore, national security concerns have led many Western countries to limit or scrutinize the adoption of Chinese AI technologies, particularly in sectors such as telecommunications, infrastructure, and defense. This has restricted China’s ability to scale its AI influence globally.

Despite domestic momentum, China has yet to match the US in global trust, investment openness, and technological reach.

The evidence is clear: the United States is the global leader in artificial intelligence by every measure, investment, infrastructure, research, and global impact. With massive capital deployment, an unmatched cloud ecosystem, and a strong pipeline of innovation, the US is defining the trajectory of AI development for the foreseeable future.

For any country, startup, or institution aiming to be a serious player in the AI space, collaboration with US companies and alignment with US-backed investment ecosystems are essential. The future of AI is being shaped in America, and participation in that future requires active engagement with its platforms, infrastructure, and capital.

Serbia: Rising Tech Star

Serbia is increasingly recognized as a rising hub of technological innovation in Southeast Europe, with artificial intelligence (AI) emerging as a strategic sector. While still developing compared to global leaders, Serbia’s AI ecosystem is rapidly expanding, fueled by global competitors as well as emerging local startups.

Serbia’s strength in AI development begins with its highly skilled workforce, particularly in computer science and mathematics. A strong academic foundation has allowed Serbia to attract and retain global tech companies and foster homegrown innovation.

Microsoft Development Center Serbia (MDCS), based in Belgrade since 2005, plays a critical role in the development of global Microsoft AI and cloud products, including Azure Synapse, Security Copilot, Microsoft Mesh, and AI-powered features in Microsoft 365. When you see Copilot’s suggestion in Word, know that it is made in Serbia to some extent.

Databricks, the US company behind the Lakehouse AI platform, has established a growing engineering presence in Belgrade. The Belgrade team contributes to core elements of Databricks’ Lakehouse AI platform, including Databricks SQL, Spark Engine optimization, and data governance.

AMD Serbia, with offices in Belgrade and Niš, is a key part of the company’s global engineering network, focusing on advanced hardware design, including AI accelerators, GPUs, and adaptive computing solutions. Engineers in Serbia contribute to cutting-edge technologies used in data centers, AI model training, and high-performance computing worldwide. Whether it’s the chip powering an AI inference engine or accelerating neural network workloads in the cloud, parts of that innovation are designed in Serbia. AMD’s growing presence highlights the country’s potential not just in software but also in the core hardware that fuels the AI revolution.

Serbia’s private sector has produced several AI-oriented companies that are gaining international attention and investment. Originally founded in Belgrade, HTEC Group has grown into a regional leader in AI and software engineering. HTEC delivers end-to-end AI solutions to clients globally, from early-stage startups to Fortune 500 companies. It offers full-stack AI services from low-level hardware optimization (GPUs, FPGAs) to model design (Machine Learning, Deep Learning, Reinforcement Learning), AI/ML Ops, and generative AI capabilities.

Many Serbian software companies have taken bold steps to enhance their products using AI.

Nordeus, a mobile game development studio based in Belgrade, best known for its globally popular Top Eleven football manager game, incorporates AI in key areas such as player modeling, game personalization, and real-time analytics, common industry practices in modern game development.

Serbia’s startup scene is flourishing, with several companies focused exclusively on artificial intelligence, driving innovation across diverse sectors:

• Agremo utilizes AI-powered analytics to transform aerial crop imagery into actionable insights, enabling precision agriculture through advanced computer vision and machine learning. Their technology helps farmers monitor crop health, estimate yields, and optimize resource use for improved productivity and sustainability.

  
  

• Anari AI develops cloud-based platforms for customizable AI chip design. Their solutions combine AI with hardware design automation, allowing engineers to create optimized AI accelerators tailored to specific workloads, thereby improving performance and energy efficiency in data centers and edge devices.

  
  

• Deep Netts offers a Java-based deep learning framework aimed at making AI development accessible to enterprise software developers. Their open-source tools support building, training, and deploying neural networks, facilitating the integration of AI capabilities into business applications.

  
  

• Kagera AI provides machine learning tools designed to optimize operations within the energy sector. Through predictive analytics and optimization algorithms, Kagera AI enhances energy production efficiency, grid management, and maintenance planning, contributing to cost reduction and sustainability.

  
  

• Abstract (Seif.ai) specializes in cybersecurity assessments powered by AI risk modeling. Their platform analyzes large datasets and threat intelligence to evaluate vulnerabilities and predict risk exposure, enabling organizations to proactively address cybersecurity challenges with data-driven insights.

  
  

• Prepia is an innovative educational technology startup that employs AI to personalize learning and assessment experiences in real time. By adapting educational content to individual student needs, Prepia exemplifies how artificial intelligence is extending beyond traditional industry applications to foster social impact and transform education.

  
  

• Wonder Dynamics, an American startup with a Serbian cofounder and an office in Novi Sad (acquired by Autodesk in 2024), applies artificial intelligence to automate visual effects in filmmaking. By leveraging AI-driven computer vision and animation tools, their platform integrates digital characters seamlessly into live-action footage, reducing production time and costs while enhancing creative possibilities.

Many others exist already, and more are expected in the coming years.

Government Support and Infrastructure

The Serbian government has taken active steps to position AI as a national priority. Key initiatives include:

• National Strategy for AI: Serbia was among the first countries in the region to adopt a national AI strategy, aligned with ethical guidelines and EU standards.

  
  

• The Science Fund’s Support Program for AI plans to invest €8.4 million in the coming years to fund scientific research in artificial intelligence.

  
  

• S4AI_HUB: A European Digital Innovation Hub based in Serbia offering infrastructure, training, and consulting services to startups, small and medium-sized enterprises, and public sector organizations.

  
  

• Belgrade Data Innovation Hub: Located at the University of Belgrade's ETF (School of Electrical Engineering), this center supports academic-industry collaboration in AI.

Originally announced in 2024, Serbia’s national plan for AI and digital infrastructure included €30 million for a new supercomputer to support researchers and startups. In 2025, this initiative was significantly expanded with the signing of a €50 million agreement between the Serbian government and French company Bull SAS (part of the Eviden group), as part of a broader intergovernmental cooperation with France. The supercomputer, which will be free to use for researchers and startups, aims to provide high-performance computing resources to accelerate scientific work and innovation in AI.

This investment is accompanied by €20 million for AI software implementation in critical public sectors such as healthcare, energy, and transport, and an additional €30 million in incentives to support the development and adoption of artificial intelligence. These efforts together signal a clear and strategic commitment by the Serbian government to strengthening the country’s digital infrastructure and innovation ecosystem.

Challenges and further steps

According to research published in August 2023,  sponsored by the European Commission, two-thirds of IT professionals in Serbia are employed in outsourcing, reflecting an industry built largely around servicing foreign demand rather than developing local intellectual property or product ecosystems. This is further illustrated by Serbia’s planned €120 million investment in AI infrastructure by 2026, including a €50 million high-performance computing system, when compared to Oracle’s recent €2 billion investment in AI and cloud infrastructure in Germany alone. Such disparity reveals the scale gap between global tech giants and smaller national efforts. To move beyond the outsourcing model, Serbia must deepen its integration with the US tech sector, not just as a service provider, but as a meaningful partner in innovation and product development. Positive examples exist: Microsoft and Databricks operate advanced R&D hubs in Belgrade, AMD supports local teams in hardware design and AI acceleration, and the mentioned domestic startups show promising product innovation. However, to build a resilient and competitive AI sector, Serbia needs more investment in indigenous products, infrastructure, and research capacity, alongside continued collaboration with global AI leaders.

Policy Recommendations for Strengthening the AI Ecosystem in Serbia

To position itself as a credible regional hub for artificial intelligence, Serbia must undertake a set of targeted reforms and strategic initiatives. These should focus on improving the quality of talent, fostering innovation-oriented investment, and deepening integration with the main global AI competitor, the US tech sector, which continues to set the pace globally. Suggested steps include, but are not limited to:

1. Enhancing the Quality of Higher Education in AI and Related Fields

   A foundational step in building a competitive AI ecosystem is raising the quality of education in key technical disciplines. While Serbia produces a large number of STEM graduates, there is a growing consensus that greater emphasis must be placed on the quality of knowledge over the quantity of diplomas. This includes introducing modernized master’s programs in artificial intelligence, machine learning, data science, and computational engineering. Curricula should be designed in collaboration with industry leaders and research institutions, ensuring students gain both theoretical grounding and practical, project-based experience. Achieving this will require universities to have access to the full range of necessary equipment and resources.

   
   

2. Attract US Venture Capital to Scale Domestic Innovation

   While public funding initiatives are commendable, state-subsidized funds often lack the scale, risk appetite, and specialized mentorship that AI startups require to grow globally. The Serbian government and startup ecosystem should therefore create conditions to attract US-based venture capital, which has a proven track record in scaling AI companies. This could involve legal and tax reforms to simplify cross-border investment, co-investment schemes, and the establishment of dedicated liaison offices or “landing zones” in Silicon Valley or New York to connect Serbian founders with American investors.

   
   

3. Proactively Attract US AI Companies through Talent and Diaspora Channels

   Given the dominance of US firms in the global AI sector, Serbia should actively court American technology companies to open development and research centers in the country. The experience with companies such as Microsoft demonstrates that Serbian professionals already working in top-tier US tech firms can serve as a highly effective entry point. Targeted diaspora engagement policies, talent ambassador programs, and similar initiatives could help identify and encourage successful Serbian engineers, researchers, and executives abroad to advocate for investment in Serbia.

Such strategies would not only bring in capital and know-how but also increase the country's integration into global innovation networks, helping Serbia move beyond an outsourcing economy toward a product-based ecosystem.

Conclusion

Although the current state of AI development in Serbia is promising, with tangible progress in infrastructure, talent, and international cooperation, there is still much more to be done. Serbia must move beyond its traditional outsourcing model and focus on building a sustainable, innovation-driven ecosystem rooted in high-quality education, homegrown research, and strategic integration with global AI leaders.

Importantly, the emphasis should not be on short-term visibility or quick results, but rather on long-term impact. The history of artificial intelligence is marked by cycles of overhype and disillusionment, so-called “AI winters”, when early expectations went unmet. Yet, in the long run, those who continued to invest in research and infrastructure were ultimately the ones who enabled the breakthroughs we see today. Serbia should follow this path, patiently and strategically, knowing that true value in AI emerges not from momentary trends but from sustained investment in knowledge, capacity, and trust.