FPT Calls It a Foundry. The Question Is Whether That Word Means Anything.
In the late 1980s, TSMC decoupled chip design from manufacturing — a structural split that let fabless semiconductor companies flourish by offloading production to specialized factories. FPT Software is now making the same bet for software: architecture separated from production, with automation handling the production step. It calls the model a foundry. Whether that word has structural meaning or is just offshore delivery with extra steps is the question.
Flezi Foundry launched May 22 — an AI-augmented delivery platform pairing autonomous agents with human supervisors across software development (ADLC) and IT operations (AMS). ADLC covers planning, coding, review, testing, security, and documentation; AMS covers alert triage, incident resolution, remediation, and service improvement. FPT has framed the platform as a means to improve delivery efficiency within existing budgets — claiming up to 30% more output within the same budget and, at maturity, automating 60–90% of first-line support requests. Those are the company's stated targets; FPT has not published customer references or independently audited outcome data to verify them. The earnings report confirms the scale the bet is being placed against: $2.66 billion in 2025 revenue, 54,000 employees.
The timing is not neutral. Last December, Microsoft named TCS, Infosys, Wipro, and Cognizant its Frontier Firms — partners collectively deploying more than 200,000 Copilot licenses across enterprise environments and redesigning workflows around human-agent collaboration. FPT, the Hanoi-based firm, was the most notable absence. Flezi Foundry is FPT's counter to that exclusion.
The structural question the competitive landscape raises is whether automation at software scale can achieve the same separation of architecture from production that EDA tools enabled at chip scale. The major Indian firms are all pursuing automation-native delivery with Microsoft backing, which means any non-Indian challenger attempting the same structural break faces an uphill argument. As one market analysis of the EDA sector noted, the incumbents have embedded network effects — in this case, the hyperscaler partnership and the customer reference base — that a challenger must match not just on capability but on trust. FPT's answer is the product. Microsoft's current answer is the Frontier Firm list. The market has not decided between them.
TSMC's fab model worked because electronic design automation tools, established IP libraries, and fab process design kits gave chip designers a known abstraction layer — they could specify what they wanted without needing to know the fabrication details. FPT calls its equivalent the "Digital Brain and Skill Marketplace": a library of reusable standards, runbooks, and expertise that lets supervisors define what good looks like, then hand the execution to agents. Software delivery has no EDA tool equivalent that gives an architect confidence the output will match intent. Whether FPT's library achieves that fidelity is the open question the launch data cannot answer.
The direction of travel is not in dispute. FPT is betting on automation as the next services model. The Indian majors are betting the same thing, backed by Microsoft. If those firms define the model, enterprise AI delivery runs through the existing services giants and their Microsoft-backed agent buildouts. If FPT's Foundry model has structural validity — a genuine separation of architecture from delivery — it repositions the Vietnamese tech sector from cost competitor to automation-native platform. Who wins that argument shapes what the next decade of enterprise software costs, who builds it, and who captures the margin.