Artificial intelligence (AI) technology, which implements human learning ability in a computing system, has opened an era of mass-produced personalized, customized services. It is spurring a restructuring of industrial systems that goes beyond technological innovation. The era of Fordism, which mass-produces standardized products, is fading, and post-Fordism of multi-variety, flexible production is becoming commonplace.
Samsung Electronics, which had held hegemony in the memory semiconductor (hereafter memory) industry, struggled in the High Bandwidth Memory (HBM) market through 2025 because it failed to adapt to the AI ecosystem's "demand-customized production." Although the architecture of memory design shifted to meet fabless (semiconductor design) requirements, the existing memory production method—whose core is mass production capability for general-purpose DRAM and NAND—could not keep up.
Recently, beyond general-purpose HBM, demand is rapidly growing for "custom HBM" optimized for specific graphics processing unit (GPU) architectures and workloads, turning memory from a standardized product into a "customized component" designed to fit individual customers and systems. Even the memory industry is shifting to post-Fordist production.
On the back of a memory shortage and resulting price increases, Samsung Electronics and SK hynix saw operating profit surge to record levels. Samsung Electronics posted a historic performance with record quarterly operating profit in the 57 trillion won range in the first quarter, but if it does not break away from a production method centered on general-purpose products, it could struggle for leadership in the next-generation memory market. It is uncertain whether corporations with customized manufacturing capabilities will secure a competitive edge.
From production to learning, the rules of competition change
AI-led production draws capital into data centers and GPUs, and the large capital that can bear the initial expense builds a winner-take-all market structure through network effects that leverage accumulated data. If Fordism's core competency was "making more, cheaper," the AI era's competitiveness is "for whom, and how much optimization you deliver."
The way value is created is also changing. In a structure where performance is competitiveness, corporations that secure cutting-edge technology gain overwhelming profitability, market dominance, and pricing power at the same time. This technology gap translates directly into a revenue gap, freeing corporations from the uncertainty of profits whipsawed by price cycles. This is the core rule of the AI economy.
Nvidia, which changed the semiconductor industry paradigm with GPUs for AI computation, is a good example of this rule. Nvidia's operating margin in the 70% range was possible not only because of GPU performance but also because it locked developers and corporations into its architecture through the CUDA ecosystem. As GPUs established themselves as the "standard infrastructure" of AI semiconductors, a structure emerged in which the platform corporation can set prices.
This structure also applies to foundries (contract semiconductor manufacturing). Taiwan's TSMC, which makes big tech (large information technology corporations) AI chips, has gone beyond ultrafine processes such as 2 nm (nanometers; 1 nm is 1 billionth of a meter) and 3 nm to dominate packaging technologies across mobile and AI with InFO (Integrated Fan-Out) and CoWoS (Chip on Wafer on Substrate). The core competitiveness is controlling the process of connecting chips to implement computation. It has built a technological system that scales from an ultra-small image sensor measuring 0.57 mm on each side for medical devices to wafer-scale ultra-large AI chips. As a result, orders for TSMC's most advanced processes are fully booked through 2028. The unprecedented operating margin in the 50%–60% range for a manufacturing corporation stems from TSMC leading the AI supply chain with diverse and flexible technologies.
From factories to models, the industry's center shifts
Korea's competitiveness, which elevated it to the ranks of manufacturing powerhouses, was the Fordist production capacity to make general-purpose products "faster, cheaper, and more" through process efficiency. But in the AI era, the "learning model" is the key means of production. Because the ability to advance intangible knowledge and algorithms generates revenue, the marginal cost of production converges to zero (0). In effect, the era of a "learning economy" based on data has opened.
The growing importance of packaging in the semiconductor industry highlights the core of the change. Advanced packaging, which integrates chips with different functions into one, has emerged as a design realm that determines the computational efficiency of AI learning models and as the process that maximizes added value.
Amid this change, Korea, weak in packaging capability, is at risk of being pushed to the periphery of the AI supply chain. According to market research firm TechSearch, in 2024 Korea's global OSAT (Outsourced Semiconductor Assembly and Test; corporations that specialize in semiconductor packaging and testing) market share is 4.3%, one-tenth that of No. 1 Taiwan (46.2%). No Korean corporations are among the global top 10 OSAT players led by Taiwanese corporations such as ASE. According to research by the Electronics and Telecommunications Research Institute (ETRI), the technological level is only 66% of the top country, and the gap is 3.4 years.
For Korea to play a central role in the AI supply chain, it must work with domestic materials, parts, and equipment (MPE) corporations to secure proprietary packaging solutions. The existing system of pressuring supplier prices to cut costs for large corporations undermines competitiveness. In customized production, the capabilities of small and midsize corporations within the supply chain determine the competitiveness of the finished product. Some also note that to narrow the gap with technology leaders, Samsung Electronics and SK hynix should serve as platforms that support research and development (R&D) by cutting-edge technology startups.
Rewriting big-factory-based laws and regulations points to growth solutions
TSMC's position as infrastructure for the AI ecosystem is backed by an immersion-oriented R&D system. Its "Night Hawk" system, which runs core engineering staff in three shifts for 24-hour research, enabled rapid technology development tailored to customer needs. Big tech such as Nvidia, Apple, and Tesla also maximize the immersion of highly skilled talent through flexible, project-based work systems.
By contrast, Korea's R&D is trapped in "nine to six" (working 9 a.m. to 6 p.m.). The 52-hour workweek, which constrains research focus, is a product of factory-centered Fordism. It does not fit the AI era, in which creative ideas and immersion generate added value. With a work style bound to set hours, it is hard to close the technology gap. In the AI era, labor is a competition of quality and speed, not quantity. To remain a manufacturing powerhouse in the AI era, institutional reform that adds software-like flexibility to laws and regulations is essential. Korea must seek a new growth model that excels not only at making things but also at system design. Competitiveness in the AI era depends on the ability to design structures.