Overview

On June 29, 2026, Samsung Electronics and SK Hynix are expected to unveil a combined investment blueprint of up to 2,000 trillion won ($1.3 trillion) over the next decade, a sum that dwarfs any previous capital commitment in the semiconductor industry [1]. The plans, to be presented during a government briefing chaired by South Korean President Lee Jae Myung, encompass new semiconductor fabrication plants, AI data centers, advanced packaging facilities, and supporting infrastructure across southwestern South Korea and the Yongin semiconductor cluster [1][27]. This historic outlay comes as the global memory market is already in the grip of an unprecedented supply-demand imbalance, with prices for DRAM and NAND flash surging at rates not seen in decades [7][11]. The investments are designed to cement the dominance of the two Korean giants in high-bandwidth memory (HBM) and advanced DRAM, while simultaneously reshaping the competitive landscape against Micron Technology and addressing the structural shortages that have sent memory prices soaring. This report examines the specific capacity expansion plans, the projected impact on memory pricing through 2030, the competitive dynamics with Micron, the supply chain implications, and the geopolitical forces that will define the next era of the global semiconductor industry.

The $1.3 Trillion Investment Blueprint: Samsung and SK Hynix's Decade-Long Vision

The scale of the announced investment is staggering. Samsung Group alone is expected to commit approximately 1,000 trillion won ($646 billion) over ten years, covering semiconductor fabs, AI data centers, advanced packaging, batteries, and displays [1]. Within that envelope, roughly 300 trillion won is earmarked for new fabs in southwestern South Korea, 360 trillion won for the Yongin semiconductor cluster, and more than 350 trillion won for AI data centers [1]. SK Hynix, which on June 22, 2026, surpassed Samsung Electronics to become South Korea's most valuable company with a market capitalization of 2.08 quadrillion won ($1.36 trillion), is expected to announce its own massive complementary investments [28]. The combined 2,000 trillion won figure represents a national industrial mobilization, with the South Korean government pledging support for power, water, land, infrastructure, workforce training, and housing [27].

The market's immediate reaction was cautious: Samsung shares fell 4.7% and SK Hynix dropped 3.1% on the day of the announcement, reflecting investor concerns about the sheer magnitude of capital expenditure and potential dilution [1]. However, the long-term rationale is clear. The artificial intelligence boom has transformed memory chip manufacturers from cyclical commodity businesses into high-growth, structurally supply-constrained sectors [4]. All three major memory makers—Samsung, SK Hynix, and Micron—have reached $1 trillion market capitalizations, driven by insatiable demand for HBM and DDR5 from hyperscale data centers [4]. SK Hynix's shares have more than quadrupled over the course of 2026, and the company has filed for a record $29 billion American Depositary Receipt (ADR) listing on Nasdaq, expected to begin trading on July 10, 2026, to broaden its investor base and fund further expansion [19][20].

HBM and Advanced Memory Capacity Expansion Plans

High-bandwidth memory is the cornerstone of the AI memory revolution. On June 5, 2026, Nvidia CEO Jensen Huang confirmed that all three major memory suppliers—Samsung, SK Hynix, and Micron—have been certified to supply HBM4 for Nvidia's next-generation Vera Rubin AI accelerator platform [2]. This certification opens a multi-year demand pipeline that the Korean firms are racing to capture.

SK Hynix, which holds approximately 40% of the global HBM market and is the leading supplier to Nvidia, has already shipped samples of its 12-layer HBM4E to major customers as of June 17, 2026 [16][12]. The HBM4E achieves a data processing speed of 16 Gbps per pin, a power efficiency improvement of over 20% compared to previous models, and a 48 GB capacity, utilizing Advanced MR-MUF technology that reduces heat resistance by 17% [12]. The company aims for timely mass production, building on its expertise across HBM3, HBM3E, and HBM4 [12].

In a significant strategic pivot, however, SK Hynix announced on June 23, 2026, that it will convert an existing HBM3E manufacturing facility to DDR5 production rather than upgrading it to HBM4 as originally planned [16]. The decision is driven by the extraordinary profit margins in the general-purpose DRAM market, where DDR5 margins could reach 90% this year, and by severe DDR5 shortages that make such production more lucrative in the short term [16]. SK Hynix believes this move will not undermine its HBM leadership, as it has already secured a solid position in the HBM market and sees no need to rush the transition to HBM4 and HBM4E [16]. The retooling of existing lines is expected to proceed relatively quickly compared to building new fabs, but much of the new DDR5 output will be purchased by hyperscalers and the broader AI industry, meaning it will not fully resolve the consumer supply deficit [16].

Samsung, meanwhile, has been investing heavily to narrow the technology gap with SK Hynix in HBM [1]. The company booked $50.4 billion in memory revenue in Q1 2026 and has been racing to qualify its HBM3E and HBM4 products with Nvidia [4]. The certification of all three suppliers for Vera Rubin HBM4 supply ensures that Samsung will have a seat at the table for the next generation of AI accelerators, though its market share in the most advanced HBM tiers remains behind SK Hynix [2].

Nvidia and SK Hynix deepened their collaboration on June 7, 2026, with a multiyear technology partnership to co-develop next-generation memory for AI factories, covering Nvidia Vera Rubin supercomputers, Vera CPUs, RTX Spark-powered PCs, and Jetson Thor robotic platforms [21]. The partnership includes supply commitments, AI-driven chip design using Nvidia CUDA-X libraries, and the creation of factory digital twins using Nvidia Omniverse to drive autonomous fab operations [21]. Nvidia CEO Jensen Huang warned that the global memory shortage "is going to persist for several years," framing it as a structural constraint rather than a cyclical inventory correction [24]. SK Hynix already supplies an estimated 50% to 70% of Nvidia's HBM4 requirements, making it a strategic dependency anchor across multiple GPU generations [24].

DDR5 and GDDR7: The General-Purpose Memory Crunch

While HBM captures headlines, the broader DRAM market is experiencing its own severe supply crisis. DDR5 memory prices are currently selling at four to five times their normal levels in many regions, and AMD's Vice President and General Manager of Client Channel Business, David McAfee, has warned that prices are unlikely to return to normal until 2028 [33]. The price surge is driven by the shift of production capacity from DDR4 to DDR5 to meet AI-driven demand, which has simultaneously caused DDR4 supply to decline and DDR5 prices to remain elevated [33]. This has forced many users to adopt older DDR4-based platforms, and manufacturers are increasing production of DDR4-compatible motherboards in response [33].

The GDDR7 market is similarly strained. Nvidia's RTX PRO 6000 Blackwell workstation GPU, featuring 96 GB of GDDR7 memory, has seen its price rise to $13,250—a more than 60% increase from its launch price of around $8,000—due to memory shortages and rising AI demand [32]. The consumer GeForce RTX 5090, launched at an MSRP of $1,999, now retails for over $4,000, more than double its original price [32]. These price increases reflect the intense competition for advanced memory between the consumer graphics market and the AI data center buildout.

The acute memory chip shortage is now pushing up prices across many industries, with the Wall Street Journal Editorial Board describing it as "another example of America's permitting and industrial policy dysfunction" [8]. Gartner analyst Ranjit Atwal has predicted that the sub-$500 entry-level PC segment will disappear by 2028 due to the memory shortage [30]. Apple CEO Tim Cook has publicly called for memory pricing and supply to return to reasonable levels, even as Apple raised prices on several MacBook and iPad models citing rising memory costs [30][31]. Valve's Steam Machine base price of $1,050 was forced by RAM makers' take-it-or-leave-it pricing, with a Valve engineer stating that memory suppliers give a price and quantity with no room for negotiation [30].

New Fabrication Facilities and Clean Room Expansions

The investment plans include a massive buildout of physical manufacturing capacity. Samsung's decade-long blueprint allocates roughly 300 trillion won for new fabs in southwestern South Korea, a region that includes Gwangju and South Jeolla province, and 360 trillion won for the Yongin semiconductor cluster [1][27]. The southwestern cluster is a new initiative, part of President Lee Jae Myung's "great leap" economic plan, and is intended to diversify chip investment beyond the Seoul metropolitan area to ease infrastructure bottlenecks [27]. However, industry experts have warned that building cutting-edge fabs in a new region requires vast electricity, water, advanced logistics, deep supplier networks, and highly skilled labor—elements that may not scale quickly enough to meet surging AI demand [27].

SK Hynix is building a new campus of memory chip fabrication plants in the Yongin cluster, with operations expected to begin in 2027 [19]. The company is also constructing a $4 billion advanced chip-packaging plant in Indiana, its first U.S. production facility, which will be critical for HBM assembly and test [19][20]. SK Hynix has stated that it plans to double wafer capacity over five years to ease memory chip shortages, with a goal of becoming an autonomous fab by 2030, leveraging digital twin technology developed with Nvidia and SK Telecom [23].

In the United States, Samsung's $17 billion Taylor, Texas fab is on track to begin mass production in 2027 [11]. Margaret Han, VP of Samsung's US foundry business, confirmed at the SAFE Forum on May 28, 2026, that "we are ready" and that customers are expected to start production next year [11]. The facility will manufacture Tesla's AI5 and AI6 autonomous driving chips using Samsung's advanced 2-nanometer process technology, including the SF2P+ process optimized for AI workloads, which delivers up to 30% better performance [11]. Samsung is also in discussions with Chinese EV manufacturer BYD to produce self-driving chips on its 4nm and advanced 2nm processes, a deal that would provide a major boost to its foundry business, which has faced underutilized production capacity [15].

Samsung's Pyeongtaek mega-factory remains a critical production hub. The company recently offered some full-time memory division employees bonuses of up to $400,000—roughly four times the company's average annual salary—to avert a strike threat and retain talent amid the AI boom [9]. Supporting infrastructure is also expanding: Hantok Chemical, a Lotte group affiliate, held a groundbreaking ceremony on June 19, 2026, for a new facility in Pyeongtaek to produce tetramethylammonium hydroxide (TMAH), a developer used in semiconductor manufacturing, with an investment of 130 billion won [10]. Soulbrain TX LLC, a U.S. subsidiary of South Korea's Soul Brain Co., received an $11.6 million grant from the Texas Semiconductor Innovation Fund to build a chemical manufacturing plant in Taylor, Texas, producing high-purity phosphoric acid for Samsung's nearby foundry, with total investment of up to $600 million [14].

Memory Pricing Dynamics Through 2030

The pricing outlook for memory through the end of the decade is shaped by a structural supply-demand imbalance that has no precedent in the industry's history. Jefferies Equity Research warns that memory prices will surge 40–50% quarter-over-quarter in Q3 2026, followed by another 30–40% increase in Q4 2026, with year-over-year price hikes of 40–45% in 2027 [7]. The firm sees no relief until 2028, when 15–20% new capacity is expected to come online, but even that supply growth is considered modest relative to rising AI and compute demand [7].

The semiconductor market achieved a record quarterly revenue of $319 billion in Q1 2026, a 27% sequential increase, with memory products driving the surge [37]. NAND revenue reached nearly $48 billion, up 96% quarter-over-quarter, and NAND average selling prices rose 95% sequentially [37]. Memory now accounts for over 40% of all semiconductor revenue, far above the historical average of 20% [37]. Micron Technology reported that DRAM prices rose more than 60% in the quarter ended May 28, 2026, compared with the previous quarter, while NAND prices increased more than 80% [31]. Micron's fiscal Q3 2026 revenue more than quadrupled to $41.46 billion, and the company guided current-quarter revenue to about $50 billion, far above analyst expectations [39][38].

TrendForce projects the global memory market will reach $1.3 trillion by 2027, up 44% from 2026, with DRAM revenue rising 303% in 2026 to $619 billion and expanding to $903 billion in 2027 [38]. JPMorgan's AI Capex 2.0 report now expects total AI capital expenditure to reach $5.5 trillion through 2030, with hyperscaler capex rising to around $650 billion in 2026 and potentially above $1.1 trillion in 2027 [34]. The core logic is a supply-demand mismatch: AI workloads are moving from training to large-scale inference and agents, requiring more accelerators, more HBM/DRAM, faster networking, and higher rack-level power density, while capacity expansion cannot ramp overnight [34].

Bernstein estimates that Nvidia's next-generation Vera Rubin NVL72 system will cost approximately $9.1 million per rack, with high-bandwidth memory prices more than tripling by the time the system ships at scale [29]. Memory and storage costs alone are projected to reach $3.2 million per rack—over a third of the total system cost [29]. A 1-gigawatt Vera Rubin data center would cost roughly $47 billion to build, compared to $40.5 billion for data centers using predecessor chips [29].

The long-term outlook is sobering for downstream customers. Lenovo executive director Martin Hiegl stated at the ISC 2026 conference that memory prices may "never" return to pre-surge levels, with a "new normal" for 2030 and onwards that will still be more expensive than before October 2025 [30]. Micron has introduced strategic customer agreements (SCAs) that lock in high prices through 2030, with customers committing $22 billion to secure supply [30][36]. Micron CEO Sanjay Mehrotra stated that the company can only meet 50–75% of customer demand and is deliberately pacing factory construction to avoid oversupply [4]. He has no line of sight for supply to catch up with demand even in 2028 [39].

Chinese memory producers CXMT and YMTC are not expected to disrupt the market in the near term, as their pricing matches incumbents and their output is primarily for domestic consumption [7]. However, by 2028, their expansion could begin to affect offshore markets [7]. Apple is reportedly lobbying the Trump administration for access to CXMT's blacklisted DRAM to mitigate rising costs [7]. CXMT's IPO on June 24, 2026, marks a significant step in China's push for memory self-sufficiency amid global tech decoupling, strengthening China's ability to finance domestic DRAM expansion and reduce reliance on foreign capital and technology [3].

Competitive Positioning: Samsung, SK Hynix, and Micron

The competitive dynamics among the three memory giants have been fundamentally reshaped by the AI boom. SK Hynix has emerged as the clear leader in HBM, supplying an estimated 50% to 70% of Nvidia's HBM4 requirements and becoming South Korea's most valuable company on June 22, 2026, with a market cap of $1.36 trillion [28][24]. The company reported record Q1 2026 revenue of $35.5 billion, a 198% year-over-year increase, and 2025 revenue of $68.1 billion [4]. Its $29 billion ADR listing on Nasdaq will further strengthen its global capital access [19].

Samsung, while still the largest memory company by revenue with $50.4 billion in Q1 2026 memory sales, has been playing catch-up in HBM [4]. Its decade-long investment plan is designed to close the technology gap and leverage its massive scale in both memory and foundry. The Tesla and potential BYD foundry deals demonstrate Samsung's unique ability to offer integrated memory and logic manufacturing, a competitive advantage that neither SK Hynix nor Micron can match [11][15]. Samsung's relocation of its U.S. headquarters from New Jersey to Plano, Texas, by the end of 2026 signals a strategic alignment with its expanding Texas manufacturing footprint [12][13].

Micron Technology, the only U.S.-based memory maker, has seen its revenue explode 196% year-over-year to approximately $24 billion in Q2 2026, with 2025 revenue of roughly $37 billion [4]. Its stock has climbed 270% in 2026 and 860% over 12 months, and in May 2026 its market cap surpassed $1 trillion [4]. Micron's CEO Sanjay Mehrotra killed the company's 29-year-old consumer business in early 2026 to focus entirely on AI data center supply, shifting from spot sales to long-term multi-year contracts [4]. Micron has signed 16 long-term agreements of three to five years with hyperscalers, automakers, and AI infrastructure companies, locking in both volume and pricing [39][38]. The company's entire 2026 HBM production capacity is sold out, and it sees DRAM shortages lingering past 2027 [38].

Wall Street analysts are overwhelmingly bullish on Micron, with a consensus "Strong Buy" rating from 41 analysts and price targets ranging from $1,150 to $2,000 [35]. Rosenblatt Securities raised its target to $1,200, citing constrained industry supply growth with no meaningful new wafer capacity expected for roughly another year. D.A. Davidson set a Street-high target of $2,000, calling it "a new era" with unprecedented visibility [36]. Bank of America's Vivek Arya argued that demand is rising four to five times faster than the industry can expand supply, a gap the memory market has rarely seen before [35].

The technology node race is intensifying. SK Hynix's 12-layer HBM4E with Advanced MR-MUF technology represents the current state of the art [12]. Samsung is introducing 2-nanometer foundry process technology at its Taylor fab, with the SF2P+ variant optimized for AI workloads [11]. Micron's technology roadmap, while not detailed in the available sources, is expected to keep pace, as all three suppliers have been certified for Nvidia's Vera Rubin HBM4 [2]. The key differentiator will be execution speed and yield ramp, areas where SK Hynix currently holds an advantage in HBM.

Supply Chain Implications: Equipment, Substrates, and Materials

The $1.3 trillion investment wave will have profound ripple effects across the semiconductor supply chain. The construction of multiple new fabs in South Korea and the United States will drive unprecedented demand for semiconductor manufacturing equipment from suppliers such as ASML, Tokyo Electron, Applied Materials, Lam Research, and KLA Corporation. While specific equipment procurement figures are not available in the current findings, the scale of the announced fab projects—including the Yongin cluster, the southwestern Korea cluster, and the Taylor, Texas fab—implies multi-year order books for lithography, deposition, etch, and inspection tools.

The advanced packaging segment, critical for HBM assembly, is a particular bottleneck. SK Hynix's $4 billion Indiana packaging plant and Samsung's advanced packaging investments are designed to address this constraint [19]. The HBM manufacturing process requires sophisticated through-silicon via (TSV) technology, microbumping, and thermal compression bonding, all of which demand specialized equipment and materials. The shift to HBM4 and HBM4E with 12-layer stacks will further increase the complexity and capital intensity of packaging.

Materials supply is also being scaled up. Hantok Chemical's new TMAH facility in Pyeongtaek, with an investment of 130 billion won, will expand production of a critical developer used in semiconductor and display manufacturing [10]. High-purity TMAH is currently produced only in South Korea, Taiwan, Japan, and the United States, making supply chain resilience a strategic priority [10]. Soulbrain's $600 million chemical plant in Taylor, Texas, supported by an $11.6 million state grant, will produce high-purity phosphoric acid essential for advanced semiconductor manufacturing, specifically for Samsung's nearby foundry [14]. These investments highlight the deepening of the local supply chain ecosystem around major fab clusters.

Substrate and package supply constraints, while not explicitly detailed in the available findings, are a well-known industry challenge. The rapid growth in HBM demand has strained the availability of ABF substrates and FC-BGA substrates, which are essential for advanced packaging. The Korean government's support for the semiconductor cluster includes measures for infrastructure, workforce training, and housing, but the ability of the substrate supply chain to keep pace with the explosive growth in HBM and advanced packaging demand remains a critical risk factor [27].

Geopolitical Factors and Government Support

The investment plans are unfolding against a backdrop of intensifying geopolitical fragmentation in the global semiconductor industry. The U.S. CHIPS Act and the Texas CHIPS Act have created financial incentives for domestic manufacturing, as evidenced by the $11.6 million grant to Soulbrain from the Texas Semiconductor Innovation Fund [14]. Samsung's $17 billion Taylor fab and SK Hynix's $4 billion Indiana packaging plant are both beneficiaries of this policy environment, which seeks to onshore critical semiconductor manufacturing capabilities.

South Korea's government is playing an active role in facilitating the investment plans. President Lee Jae Myung is personally chairing the briefing at which the investment plans will be unveiled, and the government has pledged comprehensive support covering power, water, land, infrastructure, workforce training, and housing [27]. The creation of a new semiconductor cluster in the southwestern region is explicitly framed as a national strategic initiative, not a regional favor, with the government arguing that it represents "the most rational semiconductor industrial center through the decisions of relevant companies under full government support" [27]. However, critics note that 85% of voters in the region backed Lee in the last election, raising questions about political motivation [27].

The U.S.-China technology rivalry is a critical factor. CXMT's IPO on June 24, 2026, and its push for memory self-sufficiency highlight the growing bifurcation of the global semiconductor supply chain [3]. As supply chains increasingly split along geopolitical lines, the dominance of the Korean and American memory producers in the high-performance segment is being reinforced, while China focuses on domestic consumption and mid-range products. Apple's reported lobbying of the Trump administration for access to CXMT memory underscores the tension between national security concerns and the commercial pressures created by the memory shortage [7].

The SK Hynix ADR listing on Nasdaq, expected to begin trading on July 10, 2026, is itself a geopolitical signal, as the company seeks to deepen its ties to U.S. capital markets and position itself as a global AI infrastructure player [19][20]. SK Group's broader strategy to become a supplier across the entire AI stack—chips, infrastructure, platforms, and energy—is exemplified by SK Telecom's $480 million investment in SK Hynix NAND Product Solutions, a U.S.-based subsidiary [6]. SK Telecom's early $100 million stake in Anthropic is now valued at approximately $2.5 billion, with Anthropic preparing for a 2027 IPO at a $965 billion valuation, illustrating the convergence of memory, AI, and financial markets [6].

Market Concentration and Pricing Power

The combined $1.3 trillion investment by Samsung and SK Hynix will further entrench the oligopolistic structure of the global memory market. The three major players—Samsung, SK Hynix, and Micron—already control virtually the entire market for advanced DRAM and HBM. The massive capital requirements and technological complexity of leading-edge memory manufacturing create insurmountable barriers to entry for new competitors, with the exception of Chinese firms like CXMT, which are constrained by export controls and technology gaps.

The shift from spot-market pricing to long-term strategic customer agreements is fundamentally altering the industry's pricing dynamics. Currently, 50% of total memory capacity is locked in long-term agreements with top tech firms, and this could rise to 70% [7]. Micron has signed 16 multi-year agreements with hyperscalers, automakers, and AI infrastructure companies, locking in both volume and pricing [39]. This structural shift reduces the boom-bust cyclicality that historically characterized the memory industry and gives producers greater pricing power and revenue visibility.

The deliberate pacing of capacity expansion by all three manufacturers is a rational response to the lessons of past cycles. Micron's CEO has explicitly stated that the company is deliberately pacing factory construction to keep supply below demand [4]. SK Hynix's decision to convert an HBM3E line to DDR5 rather than rushing to HBM4 reflects a similar calculus: maximizing profitability by allocating capacity to the highest-margin products in a supply-constrained environment [16]. AMD's David McAfee noted that "SK hynix and Samsung are both trying to minimise risk by not ramping up production too hard, just in case demand drops suddenly" [30]. This disciplined approach to capacity addition, combined with the multi-year lead times required to build new fabs, means that even the $1.3 trillion investment will not quickly alleviate the supply shortage.

The market concentration is further reinforced by the deep technological partnerships between memory makers and their primary customer, Nvidia. SK Hynix's multiyear technology partnership with Nvidia, which includes co-development of memory for AI factories and the use of Nvidia Omniverse for fab digital twins, creates a level of integration that makes it difficult for customers to switch suppliers [21]. Nvidia CEO Jensen Huang's warning that the memory shortage will persist for years, and his characterization of SK Hynix as a "strategic dependency" rather than merely a supplier of choice, underscores the locked-in nature of these relationships [24].

For downstream industries, the implications are severe. The autonomous vehicle industry faces potential disruption as memory manufacturers prioritize higher-margin AI customers, leaving automakers competing for limited supply [36]. Deutsche Bank warns that soaring HBM and DRAM demand could leave automakers scrambling, with future Level 4 autonomous vehicles potentially requiring over 300 GB of memory [36]. Consumer electronics companies, from Apple to Valve, are being forced to accept take-it-or-leave-it pricing from memory suppliers, with limited alternatives [30]. The sub-$500 PC segment is projected to disappear entirely by 2028 [30].

The geopolitical dimension adds another layer of pricing power concentration. With Chinese producers effectively walled off from the most advanced technology nodes by export controls, and with their output primarily serving domestic consumption, the global market for cutting-edge memory remains a tightly controlled triopoly. The $1.3 trillion investment by the two Korean leaders will widen the technology and capacity gap with any potential challengers, ensuring that pricing power remains firmly in the hands of the incumbents for the foreseeable future.

Conclusion

Samsung and SK Hynix's combined $1.3 trillion investment plan represents a watershed moment for the global semiconductor industry. It is a direct response to the structural transformation of the memory market driven by artificial intelligence, which has turned memory chips from cyclical commodities into strategic assets with persistent supply deficits. The investments will expand HBM, DDR5, and advanced memory capacity across new fabrication clusters in South Korea and the United States, with the first significant capacity additions not expected until 2028 at the earliest. In the interim, memory prices are projected to continue their sharp ascent, with Jefferies forecasting 40–50% quarterly increases through the end of 2026 and 40–45% year-over-year gains in 2027 [7]. The competitive landscape will remain a three-way race among Samsung, SK Hynix, and Micron, with SK Hynix holding a lead in HBM and Samsung leveraging its integrated memory and foundry capabilities. The supply chain implications are vast, driving demand for equipment, materials, and advanced packaging, while geopolitical forces reinforce the oligopolistic market structure. For customers across the technology spectrum, from hyperscalers to automakers to consumers, the era of cheap and abundant memory is over, replaced by a new normal of constrained supply, long-term contracts, and structurally higher prices that may persist well beyond 2030.