What's Driving the Stock Market Rally, and Why Is the 'AI Semiconductor Supercycle' Coming?

the stock market's recent enthusiasm has been driven by the stock prices of semiconductor companies like Samsung Electronics and SK Hynix. the keyword behind this massive wave of new all-time highs is the "semiconductor supercycle".

but supercycles have happened before. remember the boom of 2017? So what, and 'why' is this cycle fundamentally different? This article goes beyond simple phenomenological analysis and delves deeper into the structural nature of the ongoing AI semiconductor supercycle and how it is reshaping our industry and investment landscape.

table of Contents

1. what is the 'semiconductor supercycle'?

2. qualitatively different from the past aI at the heart of this supercycle

3. The first front in the AI semiconductor war: AI accelerators and HBMs

4. the Secret of the 'Balloon Effect': Why HBMs Drive Up D-RAM Prices?

5. How AI eats up all the memory (deep dive)

6. the Semiconductor Tidal Wave: How AI is Changing the Industrial Landscape

7. (Case Analysis) When AI builds factories and makes cars run

8. a lighthouse in the storm: 3 hidden dangers of the supercycle

9. key Q&A: What Investors Want to Know Most (FAQ)

10. conclusion: The beginning of a new order

1. what is the 'semiconductor supercycle'?

the real meaning of the 'supercycle' term that has rocked stock markets

the term 'supercycle' originally originated in the commodity markets. It refers to a macro trend that goes beyond the normal business cycle, where prices continue to rise over an extended period of time - 10, 20 years or more - due to 'structural changes in demand'.

a classic example is China in the 2000s. as China accelerated its industrialization, emerging as the "factory of the world," demand for key commodities such as copper, iron ore, and crude oil exploded. Global supply could not keep up with this demand, causing commodity prices to skyrocket throughout the 2000s - the "commodity supercycle.

unlike commodities, semiconductors have their own 'special cycle'

however, the semiconductor supercycle is a bit different from commodities. instead of 10-20 years, it's a long-term boom that lasts about two years or more.

the key driver of the semiconductor cycle is 'technological change'. This is when new technologies, such as PCs, smartphones, and now AI, emerge and create huge, unprecedented demand for semiconductors, and supply cannot keep up with this demand for an extended period of time. In other words, the semiconductor supercycle is a cycle in which "technological innovation" triggers a "structural increase in demand," which leads to "supply shortages" and "rising prices.

2. qualitatively different from the past 'AI' at the heart of this supercycle

memories of 2017: The first boom, driven by smartphones and the cloud

we've had supercycles before. the PC proliferation of the 1990s, the smartphone popularization of the 2010s, and the cloud data center and cryptocurrency mining boom of 2017-2018 are all examples. In each of these cases, demand was "consumer" (B2C) driven: individuals buying new smartphones, or cloud companies investing to serve those individuals.

from 'consumer' to 'big tech': demand shifts

the current supercycle in 2024 is qualitatively different from the past. this is because the driver of demand is no longer the fickle "consumer," but rather "big tech" companies (B2B) themselves.

this is the most important point that explains the robustness of this supercycle. consumer demand is cyclical and can be easily deflated. but right now, demand is driven by big tech companies like Microsoft, Google, Amazon, and Meta. their investments are a fight for corporate 'survival' - the 'AI supremacy race' - so they are decoupled from the short-term consumer cycle, creating a much longer-term, stronger demand.

it's about survival: the $740 trillion AI supremacy war

big tech companies are pouring astronomical amounts of capital into building AI infrastructure. google plans to invest $85 billion, Microsoft $80 billion, and Meta $72 billion in 2025 alone. global AI investment is expected to reach 740 trillion won next year, up from 600 trillion won in 2025.

this is close to what SK Group Chairman Choi Tae-won calls a "demand shock". The "fear" of being left out of the market if they lose the AI race is driving their aggressive investment.

3. The first front in the AI semiconductor war: AI accelerators and HBMs

What is HBM semiconductor? The 'essential nutrient' of AI accelerators

the epicenter of this supercycle is undoubtedly HBM, or 'high bandwidth memory'. It was created to enable GPUs (AI accelerators), the heart of the AI era, to process massive amounts of data quickly.

if traditional D-RAM is a "two-lane highway," HBM is like a "vertically stacked 128-lane highway," with multiple D-RAMs stacked vertically to increase data pathways hundreds or thousands of times. A modern product like the HBM3E can process 40 UHD movies (1.2TB) at 30GB capacity in a single second. As AI models "learn" data, this highway is the essential nutrient that feeds the "mega-factory" that is the GPU with a steady stream of raw materials (data).

The dominant player in the HBM market: SK Hynix and Samsung are catching up

currently, this HBM market is dominated by SK Hynix. SK hynix's stock price has skyrocketed as a result of its dominance in the HBM market.

however, Samsung Electronics is nipping at its heels, entering the Nvidia supply chain with next-generation products such as HBM3E and aiming for more than 30% market share by 2026. the stock prices of both companies are the "report card" of this HBM supremacy race, and investors are watching to see if SK Hynix will continue its dominance or if Samsung will close the gap.

4. the secret of the 'balloon effect': why HBM is driving up D-ram prices?

Creating HBM leads to D-RAM shortages

HBMs don't fall from the sky - the base material for HBMs is D-RAM wafers, which means that making one HBM consumes the same amount of D-RAM production capacity.

HBM is much more profitable than traditional D-RAM. the three major players - Samsung, SK Hynix, and Micron - are "cannibalizing" their existing D-RAM production lines into HBM production lines to maximize profitability. this is the "hidden supply shock" that the market has overlooked. At the same time that AI is exploding HBM demand, chipmakers are intentionally reducing the supply of "commodity D-RAM" to make way for HBM.

(In-depth) Secrets of Samsung P4, Hynix M15X

this 'supply shortage' is structural. in the past, rising D-RAM prices would have led to new fabs being built, creating an oversupply and crashing prices.

but not now. even the latest new fabs, such as Samsung Electronics' Pyeongtaek P4 and SK Hynix's Cheongju M15X, are being prioritized as "HBM-only fabs" rather than "general-purpose D-RAM" fabs. if the semiconductor industry used to be a game of chicken of infinite expansion, now that it has been reorganized into a three-company oligopoly, it is playing much smarter. by focusing all new investment on the highest margin products (HBM), this strategy perpetuates a state of "structural" limitation of new supply in the commodity D-RAM market.

D-RAM price spike of 171%: a foreseen shortage

the consequences of this structural shortage are shocking. as of Q3 2025, D-RAM contract prices have skyrocketed by a whopping 171.8% year-over-year, outpacing the growth of the price of gold over the same period. one industry insider characterized Q4 2025 as the "beginning of the D-RAM bull market" and warned of "severe DRAM shortages" by 2026.

5. How AI is eating up all your memory (deep dive)

The Two Faces of AI: 'Training' and 'Inference'

Is the 171% spike in D-RAM prices solely due to the supply shift to HBM? no, because the "demand" for general-purpose D-RAM is also exploding because of AI.

the general public thinks that HBM is only needed when AI is 'learning', but the 'inference' process, where we ask a chatbot a question and get an answer, also requires a lot of memory.

HBM for 'learning', D-RAM and SSD for 'inference'

As the amount of text (Context Window) that AI models (LLMs) process at once has grown to over 1 million tokens, the amount of data that AI needs to access in real-time for 'inference' has grown exponentially.

as a result, HBMs alone cannot keep up with this demand, and hyperscalers like Amazon (AWS) and Google (GCP) are competitively increasing the amount of "high-capacity DDR5 RAM" and "ultra-fast enterprise SSDs (eSSDs)" on their AI servers, in addition to HBMs.

the perfect storm: double demand and double shortage

this is the "perfect storm.

1. (Supply drops) Semiconductor companies switch their D-RAM lines to make HBMs.

2. (Demand explodes) HBM demand for AI "learning" explodes, while AI "inference" demand for high-performance D-RAM and SSDs explodes.

The D-RAM market has entered a 'perfect storm' state where two huge forces of 'shrinking supply' and 'exploding demand' are acting simultaneously, which is the structural reason for the 171% surge in D-RAM prices and the forecast of a 'severe D-RAM shortage' in 2026, and the most powerful engine of this supercycle.

table 1: Comparison of past and present semiconductor supercycles

category 2017-2018 Cycle 2024-2026 (current) cycle key Drivers higher specification of smartphones, cloud/datacenter artificial Intelligence (AI) revolution, AI supremacy race demand Entities individual consumers (B2C), general enterprises (B2B) big Tech (MS, Google, Meta), Government (B2B/G) nature of Investment cyclical investment, reflecting consumer trends survival investments, long-term/structural investments core Products nAND Flash, General Purpose D-RAM HBM Semiconductor, High Performance D-RAM (DDR5), eSSDs

6. the Semiconductor Tidal Wave: How AI is Changing the Industrial Landscape

aI on the Road: The Era of Electric Vehicles and Autonomous Semiconductors

The AI-driven demand for semiconductors isn't just in the data center. The automotive industry is one of the first to see a massive shift.

while a typical internal combustion engine car contains about 200 to 300 semiconductors, an electric vehicle requires more than 1,000, and a fully autonomous vehicle requires even more high-performance AI semiconductors. Advances in AI technology are accelerating the development of autonomous driving technology, exploding the demand for automotive semiconductors.

the arrival of 'physical AI': the robotics industry and industrial IoT

When AI extends beyond the "virtual brain" of the data center and into the "physical world" of robots, factories, and smart devices, we call it "physical AI".

when every factory (industrial IoT), every robot, and every device (on-device AI) is equipped with AI, it will become the "third largest demand for semiconductors" after data centers (first) and inference servers (second). it's not chatbots that are the "final destination" of the giant AI models that big techs have invested hundreds of billions of dollars in, but physical AI, which means that this semiconductor supercycle is at the beginning of a long-term megatrend of "AI automation" across all industries.

7. (Case study) When AI builds factories and makes cars run

case 1: Hyundai and Nvidia's $3 billion alliance (Mobility)

hyundai Motor Group announced a joint project with Nvidia to create an 'AI-powered mobility' and 'physical AI ecosystem', investing $3 billion to build an 'AI factory' with 50,000 GPUs of Nvidia's next-generation AI chip, Blackwell.

this example clearly illustrates how AI is transforming industries: (1) Nvidia's AI will be used to "develop" the next generation of autonomous driving technology (autonomous driving), (2) robots will be used to "produce" cars in an AI-powered smart factory (robotics), and (3) those cars will in turn be powered by AI semiconductors. It's a perfect virtuous cycle of demand where AI creates AI, and AI creates AI cars.

case 2: Innovation at LG Electronics Changwon Smart Park (Factory)

LG Electronics' Changwon factory was recognized as a "Lighthouse Factory" by the World Economic Forum (WEF). The factory adopted AI, digital twins, and robotics to create an "intelligent process system." The results are staggering: productivity increased by 17-20%, energy efficiency improved by 30%, and the "cost of quality" due to defects decreased by a whopping 70%.

when other manufacturers see their competitors reduce their quality costs by 70% with AI, they have no choice but to adopt AI and robotics to survive. This shows that 'physical AI' is not just a buzzword, it is already demonstrating a clear return on investment (ROI) of 'cost savings' and 'productivity gains'.

example 3: 'AI makes AI semiconductors' (Samsung-Envidia alliance)

the most interesting point in this supercycle is 'AI makes AI semiconductors'. samsung is partnering with Envidia to build a "semiconductor AI factory" that brings AI into the semiconductor process itself.

the idea is to let AI optimize semiconductor design, process, and yield management through 'AI-Driven Manufacturing': 'Better AI' helps produce more 'better AI semiconductors' (HBMs), which in turn are used to train 'better AI'. This creates a powerful 'flywheel' effect where technology and production accelerate each other, ensuring the continuation of the AI supercycle.

table 2: The two faces of AI: Semiconductor requirements by task

AI task key role required semiconductor AI model training The process of "training" AI models (e.g., GPT-5 development) GPUs (AI accelerators) + HBM semiconductors (massively parallel computation, ultra-high bandwidth memory required) AI service inference the process of 'utilizing' trained AI models (e.g., chatbot responses) CPU/NPU + high performance/high capacity D-RAM (DDR5) + eSSD (real-time response, large data loading required)

8. a lighthouse in the storm: 3 hidden dangers of the supercycle

of course, it's not all rosy - there are three potential risks to this mega-trend.

risk 1: What if a global slowdown stops big tech from investing?

while Big Tech's investment is robust because its "survival" is on the line, it can't last forever. A sharp deterioration in the macroeconomy, such as a slowdown in the US and China, could dampen companies' investment sentiment. Even if it doesn't "stop" investment, it could "slow down" or "delay" the astronomical AI investment.

risk 2: Geopolitical risks and the never-ending supply chain wars

the semiconductor industry is inherently most exposed to 'geopolitical risk' from the US-China hegemonic rivalry. u.S. tariffs on semiconductors from populist countries, export controls on key technologies, and supply chain reorganization are unpredictable "downside risks" for Korean semiconductor companies.

risk 3: (Deep) AI innovation stalls? The paradox of 'data exhaustion'

this is the most fundamental risk. current AI advances (LLM) rely on a 'scaling law' of 'learning' more 'data'. but some studies warn that by 2026, AI could run out of "high-quality, human-generated, open text data" to learn from.

big tech's $740 trillion investment is based on the expectation that "AI will keep getting better. if data depletion causes AI models to hit a "plateau" in performance gains, AI innovation could stall, which could shake the rationale behind the astronomical demand for GPUs and HBM semiconductors.

9. key Q&A: Investors' top questions (FAQ)

Q1. Is it safe to invest in Samsung Electronics stock right now?

A: Analysts are giving a positive outlook. There are analysts who believe that Samsung Electronics will record record earnings in 2026 (82 trillion won in operating profit) due to the AI supercycle. Despite its late entry into HBM compared to SK hynix, Samsung Electronics is attractive because (1) the HBM chase is in full swing and (2) as analyzed in the text, it is expected to benefit from the "simultaneous explosion" of D-RAM and SSD demand due to the expansion of the AI "inference" market. It is also attractive because it is the most undervalued among global memory companies.

Q2. Is the semiconductor supercycle only a story for HBM Semiconductor?

A: Absolutely not, and this is the biggest pitfall of misunderstanding the market. AI needs HBMs for 'learning', but it needs high performance DDR5 and enterprise SSDs (eSSDs) in large quantities for 'inference'. however, as semiconductor companies focus on producing more profitable HBMs, the supply of D-RAM has shrunk, causing both HBM and D-RAM prices to rise, with D-RAM prices skyrocketing 171%.

Q3. We've seen supercycles before, why is this one any different?

A: The 'who' and 'what' of demand is different. while past cycles were 'cyclical' booms that relied on changing trends in individual 'consumers' (e.g. smartphone purchases), this is a much stronger, structural demand that is not easily swayed by short-term cycles - it's a 'survival of the fittest' long-term investment that 'big tech' companies like MS and Google are making to stay in the AI supremacy race.

Q4. Is there any chance that the AI market growth will stop, or that the bubble will burst?

A: There are always possibilities. as analyzed in the text, there is (1) the risk that a global economic downturn could delay big tech investment; (2) geopolitical risks, such as the US-China conflict; and (3) the risk of "innovation stagnation," which is when the performance advancement of AI models stalls due to "training data exhaustion," which could weaken the momentum of AI investment.

10. conclusion: The beginning of a new order

the "semiconductor supercycle" we are witnessing is not just another business cycle of the past; it is the prelude to a "new industrial revolution" in which the massive technological revolution of AI is reshaping the fundamental structure of industry to be "semiconductor-centric".

this cycle is a "structural boom" driven by a combination of (1) long-term demand spurred by Big Tech's "survival of the fittest" AI supremacy race, (2) a dual demand explosion in both AI "learning" (HBM) and "inference" (DDR5), and (3) a "structural shortage" of general-purpose D-RAM due to HBM production.

this massive trend is just beginning, spreading beyond the data center and into the world of "physical AI" in cars, robots, factories, and more. of course, there are risks of a slowdown or technology stagnation, but there's no stopping the tidal wave of semiconductors moving beyond the "rice" of 21st century industry to become the "intelligence" itself.

what do you think about the huge changes this AI semiconductor supercycle will bring? How do you think the double shortage of HBM and D-RAM will affect Samsung Electronics and Hynix stock prices?

share your insights in the comments, and be sure to subscribe and set up newsletter alerts to receive more in-depth industry analysis reports.