En la imagen
New technological landscapes [deeznutz1]
As we know, a huge part of the world’s economy and our everyday lives rely on microchips, and since 2024 a new crisis began to arise regarding the RAM memory (Random Access Memory), a supply chain disruption of one of the core elements of our computers, phones, consoles, webs and most critically servers and AIs. But unlike cyclical shortages of the past, the current global memory shortage represents an irreversible shift in the technological landscape driven by the explosion of AI.
The problem with AI is its massive and insatiable necessity for this kind of memory for it to work. The race to train and deploy these new advanced AI models has altered what these RAM memory chips are made for, transforming semiconductor memory from an affordable consumer product into a critical geopolitical chokepoint. The core technology that has altered the market in order to supply the massive needs of AI is the High-Bandwidth Memory (HBM), a new specialized technology that is now as important as to become a point of national interest specially to South Korea (main producer of HBM memory) and to reshape the US-China tech war. However, before assessing the geopolitics of it we must understand the technological basis of this crisis.
The core of this shortage is a key limitation for modern computing that we may call the ‘memory wall’. While the processors (CPU’s) that train and run AI models have become exponentially faster, the issue lies within the information that needs to be transferred to those chips. The key lies in the RAM memory, specifically it relies on the technology called DRAM (which is the standard for most consumer goods, but it is not fast enough for large servers or AI models). To overcome this lack of speed the industry shifted from the standard DRAM to HBM, a new kind of way of manufacturing memory that stacks chips vertically (instead of horizontally like DRAM), connecting them with microscopic tunnels called through-silicon vias (TSVs). This is a 3D architecture that allows for massive, lightning-fast data transfers (80x faster). This upgrade covers the needs of AI servers that require enormous amounts of this memory to function. To give an example of the massive RAM needs for AI, and although numbers are not official, it is estimated that Netflix uses between a few kilobytes and a few megabytes per user, while for example Chat GPT needs a minimum of between 100 and 200 gigabytes of VRAM (specific kind of RAM used normally for video processing) per user to only function, which makes the need for these chips abysmally bigger.
Concentrated production
However, producing this HBM RAM is not easy, and it creates an imbalance due to its manufacturing difficulty and resource intensity. Building these chips is extraordinarily complex and costly. A single gigabyte of HBM requires three times more silicon wafer capacity than standard memory. Because ‘Hyperscaler’ tech companies (Google, Apple, Microsoft, OpenAI) are paying billions to build AI data centers, memory manufacturers (SK Hynix, Samsung, and Micron) have pivoted their fabrication lines away from conventional DRAM to focus on HBM. This ‘cannibalization’ of production has triggered a massive supply deficit across the tech industry, driving up the cost of consumer electronics, automotive components, and standard enterprise hardware worldwide.
Another of the issues regarding the HBM RAM production is that its manufacturing is restricted to what we may call an oligopoly. Only three companies worldwide control roughly 95% of the global RAM market. These three companies being Micron from the US and SK Hynix and Samsung from South Korea. Within this triad, South Korea completely dominates the RAM and specifically HBM production, with SK Hynix and Samsung on the lead, producing around 35% of the market share each. Because the foundational hardware of the AI revolution relies on just a handful of factories, the physical production of AI has led to a highly vulnerable geographical chokepoint.
This concentration of the RAM manufacturing controlled by just a few factories, tightly links the memory chip manufacturers in South Korea to the microchip factories in Taiwan. This connection is a co-dependent relationship. Taiwan, primarily through the Taiwan Semiconductor Manufacturing Company (TSMC), has a monopoly on building the advanced logic chips (the ‘brains’) designed by companies like Nvidia or AMD. However, these processors are completely useless without South Korean memory. The GPUs manufactured in Taiwan are the engine, but High-Bandwidth Memory (HBM) is the fuel they need to run, providing the massive data transfer speeds required for AI to work. No country can fully profit from the AI boom without the other's specialized technology, making the global tech sector and economy dangerously dependent on this East Asian corridor.
Adding more to this fragile manufacturing situation, the reliance on Taiwan and South Korea puts the future of global tech in the middle of a very volatile, conflicted and militarized region. Beyond the direct threat of a military conflict in the Taiwan Strait or on the Korean Peninsula, also other geopolitical issues can bring this delicate system to a halt. For example, making semiconductors requires massive amounts of energy. Because Taiwan imports 97% of its power needs and relies heavily on Liquefied Natural Gas (LNG) from the Middle East, a disruption, such as the current blockade in the Strait of Hormuz, could be a disaster. Taiwan only has about 11 days of LNG [1] reserves, meaning that any energy disruption could paralyze global chip production within weeks.
The US-China tech war
The reality of chip production relying purely on the stability of these two countries has forced world powers to rethink their strategies and try to find new ways to become the new leaders of this tech race. But this leads us to the US-China tech war: China wants a piece of this cake and aims to become a new massive manufacturer of microchips and memory, but the boundaries for accessing are increasingly being dictated by the United States, which is focusing on strategies to contain this China rise. Through strict export controls, and international bans on the purchase of the highly specialized machinery needed to manufacture these chips (only produced by the Dutch company ASML), Washington is actively restricting China’s access to this market to protect its own interests and companies.
In fact, the US is the current world leader in semiconductor design and equipment, but it is tied down by the East Asian monopoly on manufacturing. The Washington strategy regarding this lack of ability to produce is investing billions in the creation of state-of-the-art factories aimed to produce key semiconductors. The main examples for this are the two new factories that Micron is building in Idaho and New York, or the ones being built in Texas and Arizona by Samsung and TSMC, financed through the CHIPS act. [2]
The CHIPS and Science Act is the main US ‘offensive’ to take the lead on this race, and it provides $52.7 billion in direct funding to incentivize companies to build manufacturing plants on US soil. Through these investments, the US aims to capture 30% of the world’s semiconductor production by 2032.
The other key is the diplomatic alliances (‘Pax Silica’) through which the US is aligning with key partners like Japan, South Korea, and the Netherlands. By coordinating with these nations (such as working with the Netherlands to restrict the export of ASML’s advanced chipmaking machines to China), the US aims to secure the technological frontier and ensure that critical hardware remains under Western influence.
But as expected China is not sitting on its hands, and it is pushing for self-sufficiency and innovation despite being cut off from the Western advanced technology. China’s leading memory manufacturer CXMT (ChangXin Memory Technologies) is the one leading this effort. Despite lacking access to the advanced lithography machines restricted by Western sanctions, CXMT has managed to manufacture DDR5 RAM and is dedicating 20% of its capacity to producing HBM3 memory. China achieves this by pushing older equipment to its limits using extraordinarily complex and costly multi-patterning techniques. It has also shown surprising resilience in the production of logic chips (CPU processors). For example, Semiconductor Manufacturing International Corp (SMIC) successfully produced an advanced 7nm processor, and the Chinese government is encouraging state-owned enterprises to buy these domestic chips to ensure local chip manufacturers have the revenue needed to survive and grow.
And more importantly, China also holds the retaliative power to strike back at the US and Western sanctions, as the Chinese are the holders of many of the raw materials required to feed the beginning of the memory and semiconductors supply chain. No matter how advanced Western technology gets, it requires specific minerals to function, and China has a near-monopoly on extracting and refining these necessary minerals and rare-earths such as gallium, germanium, or lithium.
China has already demonstrated its willingness to use this power. In retaliation for US tech blockades, Beijing imposed strict export controls on critical tech minerals, including gallium, germanium, and rare earth elements.
This creates a systemic vulnerability for the West. While the US and Europe are pouring billions into building new fabrication plants through their respective CHIPS Acts, they cannot easily replicate China's massive, state-subsidized mineral refining infrastructure. As a result, China holds the power to inflict severe supply chain shocks on the global semiconductor industry simply by restricting the export of the dirt and minerals required to build the hardware.
No clean decoupling
In conclusion, the ‘RAMaggedon’ is not just a technological disruption, but a clear example of how the AI revolution is reshaping international geopolitics. What was once a relatively cheap and abundant component has become a strategic asset, concentrated in a handful of companies and factories located in one of the most unstable and militarized regions in the world. As AI continues to expand, access to memory chips may become as important as access to energy resources or critical raw materials.
The current shortage also highlights a reality that both the United States and China are beginning to recognize complete technological independence is almost impossible. The United States may lead in chip design and advanced technology, while China dominates many of the minerals needed to manufacture them. At the same time, Taiwan and South Korea remain indispensable for producing the hardware that powers the global digital economy. Rather than a clean decoupling, the future of the semiconductor industry is likely to be defined by a complex relationship of competition, cooperation, and mutual dependence.
Ultimately, the RAM crisis demonstrates that the future of global power will not be determined only by military strength or economic size, but also by who controls the critical technologies that make artificial intelligence possible. In this new technological race, silicon has become the new oil, and the factories that produce and refine it have become some of the most important strategic assets in the world.