Infrastructure modernization was once viewed as a technology refresh cycle, focused on reducing maintenance costs, improving efficiency, and replacing aging systems. Today, it has become a business imperative. As enterprises race to operationalize AI, the limitations of legacy architectures can directly affect how quickly organizations can deploy, scale, and generate value from AI initiatives.

This shift is reflected in enterprise technology spending. Gartner forecasts that worldwide AI spending is expected to reach $2.59 trillion in 2026, with infrastructure accounting for more than 45% of total investment. This investment reflects a broader shift: AI readiness depends not only on access to more compute, but also on the ability to modernize the legacy systems, data, and architectures that AI must operate across.

Legacy Infrastructure Challenges

Traditional IT stacks were never designed to handle the compute intensity, memory demands, or data fluidity required by modern AI workloads. Yet many enterprises continue to operate environments shaped by decades of incremental technology decisions, resulting in complex ecosystems that are increasingly difficult to integrate, manage, and evolve.

The challenge is not simply that legacy systems are old. AI changes the economics of modernization. Processes that once tolerated fragmented data, batch operations, or limited interoperability now require continuous data access, real-time processing, and seamless integration across business functions. According to a global study by FPT and Forrester Consulting, 41% of enterprises cite integration complexity as the biggest barrier to scaling AI initiatives.

The challenge often stems from structural issues, including:

  • Monolithic architectures that slow down integration and change
  • Legacy middleware and batch processing models that limit real-time capabilities
  • Siloed and inconsistent data models that undermine AI reliability
  • A shrinking pool of specialists capable of maintaining legacy environments

For most enterprises, modernization centers on two priorities, including modernizing core systems to unlock data trapped within legacy environments, and expanding infrastructure capacity to support increasingly demanding AI workloads.

Transforming Core Systems

Modernizing core systems allows organizations to reduce technical debt while improving interoperability across the enterprise. By adopting modular architectures, streamlining data access, and introducing greater automation, businesses can create a more adaptable technology foundation without disrupting mission-critical operations. To help organizations accelerate this transformation, FPT developed xMainframe, an advanced large language model purpose-built for mainframe environments and COBOL codebases. In a modernization project for one of Japan’s largest steel manufacturers, FPT applied AI to analyze more than six million lines of code, reducing workload in the initial assessment phase by approximately 30%. To date, the company reported involvement in the modernization of more than 300 systems, transforming over 200 million lines of code for more than 40 enterprises globally.

Understanding legacy systems, however, is only the beginning. Enterprises must also transform decades of accumulated business logic into architectures that are easier to integrate, operate, and scale. To support this transition, FPT developed Flezi EMT (Enterprise Modernization Technology), an AI-enabled platform that combines modernization expertise with Azure OpenAI to help organizations assess, refactor, migrate, validate, and operate legacy applications at scale.


At the FPT Halong Sapphire Summit 2026, Teruyasu Kameyama explained how Flezi EMT enables organizations to modernize applications, from retaining and refactoring critical systems to rebuilding and replacing legacy components.

This was further explained by Teruyasu Kameyama, FPT Japan Managing Director of the Solution Consulting Department, who highlighted how the platform validates AI-generated code against established rule engines before deployment, helping organizations improve both speed and quality throughout the modernization process. By automating activities across the modernization lifecycle, the platform can deliver 50% higher productivity, reduce project preparation time by 70% with AI agents, and lower reliance on mainframe or midrange experts by 70%.

Technology alone does not solve the legacy modernization challenge. Enterprises also need to preserve the business knowledge embedded in decades-old systems and transfer it into modern architectures. FPT’s legacy transformation capabilities are supported by more than 15 years of experience and a workforce of over 3,500 COBOL, RPG, and PL/I engineers, alongside thousands of cloud, data, and AI specialists. Through dedicated training programs, modernization centers of excellence, and AI-enabled knowledge capture, FPT helps organizations retain critical expertise while accelerating the transition toward modern architectures.

Expanding Infrastructure Capacity

While modernizing core systems helps reduce infrastructure debt, enterprises must also ensure their infrastructure can supply the compute capacity required to support growing workloads. Large language models, multimodal applications, and AI agents place significantly different demands on infrastructure than traditional enterprise software. Building these environments independently often requires substantial capital investment, specialized expertise, and ongoing operational management. As a result, many organizations are exploring infrastructure models that provide scalable AI capacity without the complexity of building it themselves.

With AI Factories in Vietnam and Japan, FPT provides inference-ready GPU infrastructure designed for production AI workloads. Operating as a "token factory," the platform delivers the compute capacity needed to process, generate, and scale large volumes of AI activity efficiently.


FPT AI Factories deliver breakthrough performance on complex workloads in training, agentic systems and reasoning.

Powered by NVIDIA GPU Cloud infrastructure, including HGX B300, H100, and H200 systems, FPT AI Factories allow enterprises to run large-scale foundation models, multimodal AI applications, and long-context reasoning workloads while reducing the operational burden associated with managing complex AI infrastructure. By optimizing token generation and inference performance, FPT AI Factory can deliver up to 66% lower inference costs, 49% lower training costs, and up to 2.95x better cost-per-token optimization.

The Window is Open

The AI era is reshaping what infrastructure modernization means. Success is no longer measured by how effectively organizations replace aging systems, but by how well they transform legacy applications, enterprise data, and computing capacity into a foundation capable of supporting continuous innovation.

Organizations that modernize these foundations today will be better positioned to scale AI initiatives, respond to changing business demands, and adopt future technologies with confidence. By combining deep legacy modernization expertise with next-generation AI infrastructure, FPT helps enterprises build the foundations required for AI at scale.