According to Microsoft’s "Global AI Adoption in 2025 – A Widening Digital Divide" report, the share of working-age people using AI in Hungary rose from 27.9% to 29.8% between the first and second half of 2025, placing the country among the world’s top 20 AI-using economies. This relatively high user penetration can ease the introduction of AI-based systems in companies because many employees already know and use the technology.
Despite that, Varga Botond, AI and Innovation Director at SERCO Informatika Zrt., reports a high failure rate for AI initiatives. He cites industry figures suggesting 70–80% of corporate AI projects fail, while the small and medium enterprise (SME) segment performs worse, with failure rates of 95–97%.
Shadow AI and cloud-based service risks
Varga highlighted the so-called "shadow AI" phenomenon, where employees use AI tools outside corporate control via personal or shared cloud subscriptions. This practice raises major data-protection risks because sensitive business information, passwords and internal data can end up in clouds beyond the company’s control.
He pointed to concrete incidents: in one Samsung-related case, engineers pasted confidential semiconductor development source code into a generative model for debugging, potentially allowing those assets to enter the model’s knowledge base. Another example involved Microsoft Copilot and an "indirect prompt injection" attack, where hidden instructions embedded in an email caused Copilot to extract sensitive data from a user account and forward it to an attacker’s server. These episodes show that cloud-hosted AI services create new attack surfaces.
Varga also called attention to the geopolitical exposure of cloud infrastructure: many subscription-based AI services run on North American or Asian servers that users cannot access. In a volatile geopolitical scenario, restrictions could abruptly cut off services.
Sovereign infrastructure and on-device AI chips
Hardware manufacturers have responded by adding AI-supporting chips to servers, laptops and workstations, enabling companies to run AI on their own infrastructure. Varga considers this the best option from a data-security perspective because it reduces the need to send data to the cloud.
SERCO Informatika Zrt. has launched a consulting division built on this idea: supporting successful AI deployments using sovereign, on-premise infrastructure to protect against hackers, leaks and geopolitical risks. The company emphasizes its manufacturer partnerships — notably with Lenovo and NVIDIA — which it can involve directly in complex use cases.
Deploying a sovereign setup does not always require a full infrastructure rewrite: often a single AI-capable server can be connected to existing systems to extract needed information. However, legacy systems that do not support data extraction can become a bottleneck, potentially requiring software or ERP replacement.
Why AI projects fail — strategy, data and culture
Varga argues many AI rollouts fail because the challenge is multi-faceted and less about pure IT than commonly believed. Key failure drivers include:
- Lack of or poor-quality data: Varga estimates about 85% of projects fail due to insufficient or low-quality data. AI cannot structure what is not already present in coherent, structured form.
- Absence of a genuine AI strategy: companies often perform "innovation theatre," showcasing AI deployments without a long-term plan.
- Management and cultural issues: success depends on leadership commitment and organization-wide understanding; lack of buy-in or employee cooperation can derail projects.
He recommends starting with tangible pain points — small, fast-to-deploy use cases like chatbots or automation that deliver measurable results — and expanding gradually. Engaging an experienced consultant is advised to plan structured, measurable steps.
Prerequisites for successful adoption
According to Varga, prerequisites for successful AI adoption include:
- Full executive commitment and active involvement;
- Structured, coherent and real data sources;
- Organization-wide process understanding and cultural readiness;
- Beginning with a small, quick-win pilot and scaling gradually.
While complex projects can be costly, a well-structured solution that produces immediate impact can often be realized with a relatively modest budget.
Implications for competitiveness and the workforce
Varga cautioned that AI itself does not automatically remove jobs — employees or competitors who use AI more effectively will gain an advantage. Companies that fail to develop and implement AI strategies risk falling behind in productivity and competitiveness.
SERCO believes the current market noise creates an opportunity for experienced integrators with manufacturer ties to offer long-term, secure solutions. The company expects its mix of expertise, sales capacity and vendor relationships to be a competitive advantage.
The article was sponsored by SERCO Informatika Zrt.
