SEOUL — Samsung Electronics has issued a stark warning to the global technology ecosystem, projecting that the severe global shortage of high-performance memory semiconductors will persist unabated until at least 2028. Driven by the aggressive proliferation of agentic artificial intelligence systems and an explosive surge in AI token consumption, compute infrastructure demands are outpacing production capacities at a rate never before seen in the technology sector. In response to skyrocketing demand from hyperscalers, cloud infrastructure providers, and frontier AI model builders, the South Korean tech titan disclosed that it is locking up between 60% and 70% of its total memory capacity in binding, five-year long-term agreements (LTAs).

Surging AI Token Consumption and Infrastructure Bottlenecks

Speaking during Samsung’s second-quarter earnings call on July 30, corporate executives detailed how the structural transition toward agentic AI—autonomous AI software systems that continuously execute multi-step reasoning tasks—has fundamentally reshaped global computing requirements. As autonomous agents operate continuously, token throughput requirements have exploded, straining not only specialized AI training clusters but also general computing servers across global data centers.

This massive compute requirement has created a cascading bottleneck across the entire technology supply chain. Hyperscale cloud providers, unable to expand their service capacities rapidly enough to accommodate the demands of AI software developers, have forced leading AI model creators to turn toward "neo-cloud" providers for auxiliary capacity. These specialized cloud vendors have, in turn, inundated server manufacturers with unprecedented bulk orders for memory hardware, creating severe supply constraints.

Faced with persistent infrastructure deficits, major AI frontier model developers have recently bypassed traditional intermediaries, approaching Samsung directly to convey urgent buying intent and soliciting multi-year supply guarantees for high-bandwidth memory (HBM) and next-generation server DRAM.

Structural Constraints and Multi-Year Fab Lead Times

Samsung emphasized that the widening gap between supply and demand cannot be bridged through short-term operational adjustments. The capital expenditure (CAPEX) requirements for modern semiconductor manufacturing present immutable physical limits. Constructing a state-of-the-art fabrication facility (fab), installing extreme ultraviolet (EUV) lithography systems, and ramping up cleanroom wafer production requires a lead time exceeding three years.

Consequently, supply expansion via new capacity build-outs will remain severely constrained in the medium term. Samsung foresees that unfulfilled market demand from 2026 will inevitably roll over into 2027 and 2028, rendering the supply-demand imbalance progressively more severe over the next two to three years. While forecasting dynamic market conditions beyond 2029 remains challenging, long-term indicators suggest that the baseline infrastructure needs of sovereign AI initiatives and hyperscale clusters will keep utilization rates at peak levels for the foreseeable future.

Strategic Risk Mitigation via 5-Year Rolling LTAs and Deposits

To insulate its operations from the historic volatility of semiconductor memory cycles while providing critical supply guarantees to strategic partners, Samsung is fundamentally altering its commercial strategy. The company is capping long-term contract allocations at 60% to 70% of its maximum manufacturing capacity, intentionally reserving the remaining percentage to maintain operational flexibility for spot market dynamics and emerging consumer device cycles.

These long-term agreements are structured on a five-year baseline with a rolling mechanism, allowing an additional year to be appended annually subject to recurring mutual negotiations. Crucially, Samsung revealed that it has already finalized contracts with all five major global hyperscale data center operators, with negotiations for five additional large-scale enterprise AI clients reaching their final stages.

To enforce contract bindingness and mitigate cancellation risks, Samsung has mandated substantial upfront earnest money deposits as a key contractual terms. The company confirmed that it has already received approximately one-quarter of the total agreed-upon advance deposits, significantly strengthening its balance sheet liquidity and securing guaranteed revenue streams.

Record Financial Results Driven by Unprecedented Memory Margins

The acute imbalance in memory supply was reflected directly in Samsung's historical second-quarter financial performance. The company reported a consolidated quarterly operating profit of 89.49 trillion KRW, representing a monumental 1,813.8% increase compared to the same period last year and a 56.4% jump quarter-over-quarter.

The Device Solutions (DS) division, which oversees Samsung's semiconductor operations, posted consolidated revenue of 127.5 trillion KRW and an operating profit of 89.2 trillion KRW. This yielded an extraordinary operating profit margin of 70.0% for the DS segment. The soaring average selling prices (ASP) of advanced HBM3E, HBM4, and high-density LPDDR5X modules played a central role in driving profitability to record highs.

Market Outlook and Industry Implications

Industry analysts note that Samsung’s strategy reflects a broader structural shift within the semiconductor sector, where memory hardware is transitioning from a cyclical commodity to a mission-critical infrastructure asset. Competitors such as SK Hynix and Micron Technology are similarly operating at full capacity, with their advanced HBM output fully sold out through subsequent years. As AI workloads evolve from cloud-based training to ubiquitous, on-device inference and autonomous execution, the memory industry appears poised for a prolonged period of elevated pricing power and sustained structural growth.