Chip stocks have been hammered by the market, with many investors losing their shirts. Jensen Huang, the man at the peak of his fame, rushed to South Korea to finalize a long-term technology partnership with SK Hynix!
This is no minor matter of NVIDIA buying more memory or SK Hynix gaining a big customer—Jensen Huang is directly slapping the short-term panicked market! The AI rally is far from over; we have only just stepped into the second phase of the AI industry!
The core logic of AI Phase 1 was simple: whoever holds GPUs, has the most computing power, can sell the "industry shovel", or rides the AI trend, sees capital pouring in with valuation premiums and growth stories. So NVIDIA soared, AMD soared, Broadcom and Marvell followed, and the entire semiconductor sector rocketed up!
But now the winds have completely reversed! Phase 1 competed on "who has the best GPU," while the core competition of Phase 2 is who can keep GPU computing power fully loaded—this is the most valuable core logic of the AI industry today!
An AI factory never makes money just by stacking GPUs! Even the best GPUs need HBM for data transfer, advanced packaging to interconnect chips, network transmission for signals, power supply support, cooling systems, and servers and data centers for stable delivery.
On the surface, the NVIDIA-SK Hynix partnership looks like just another supply chain news item, but reading it at face value means missing the biggest industry trend.
The real core point is: Jensen Huang is deeply tying NVIDIA's roadmap for AI infrastructure over the next several years to the advanced memory industry, locking in collaboration ahead of time.
This means that the memory industry, previously viewed as cyclical, is now being repriced by Wall Street with a new "AI infrastructure" logic.
This also explains why, just as the semiconductor sector experienced a sharp sell-off, Jensen Huang chose this moment to firmly lock in advanced memory. He is not unaware of the market decline; rather, he understands better than the short-term market that the next bottleneck for AI factories likely lies in the memory segment.
U.S. semiconductor stocks recently suffered a violent sell-off, with the Philadelphia Semiconductor Index plummeting and U.S.-listed chip stocks losing about $1.3 trillion in market cap in a single day. Core AI names like NVIDIA, Micron, AMD, and Marvell all saw indiscriminate selling and violent swings.
Most investors' first reaction is: Has the AI rally peaked? Has Wall Street stopped believing in AI's growth story? But the more panicked the market, the less one should focus solely on stock price movements. Stock prices reflect short-term emotions; industry actions reflect long-term trends.
At the moment of greatest market panic, NVIDIA and SK Hynix announced a multi-year deep technology collaboration, aiming to develop and commercialize the advanced memory technology needed for next-gen AI factories, and to fully align SK Hynix's next-generation memory R&D with NVIDIA's future AI infrastructure roadmap.
Make no mistake: this is not an ordinary commercial procurement. Ordinary procurement is a passive supply of "out of stock today, deliver tomorrow"; this collaboration is NVIDIA publicly disclosing its industry layout for the next several years in advance—including next-gen AI factory architectures, personal AI terminals, and robot computing platforms—letting SK Hynix align R&D and iterate synchronously rather than wait passively for orders.
The collaboration covers a wide scope, not only data center GPUs, but also the Vera Rubin AI supercomputing platform, Vera CPU, RTX Spark personal AI PC, and the Jetson Thor robot computing platform.
Advanced memory is no longer an ordinary component of AI servers; it has become a core pillar of the NVIDIA AI ecosystem!
In the past, when the market talked about AI, all eyes were on NVIDIA's GPU. Even with GPU shortages, the public assumed the core AI logic remained unchanged. But today, AI factory architectures are growing increasingly complex; a single top-tier GPU alone cannot support the efficient operation of the entire system.
A GPU is like a top-tier engine in a super factory—powerful and fast-spinning. But to keep it running, it must constantly read, transmit, and exchange massive amounts of data. Once data supply is interrupted or transmission lags, the GPU can only idle and wait. The issue is not insufficient GPU performance, but that top computing power cannot be fully utilized.
HBM (High Bandwidth Memory) was born to solve this problem. It is not a minor upgrade from ordinary memory; it is a core component tailored for high-performance computing and AI accelerators. Its key advantage is not capacity increase, but ultra-high bandwidth and ultra-low latency, enabling massive data to be fed to the GPU instantly, completely solving the idle computing issue.
The larger the AI model, the longer the context window, the more complex the inference scenario, and the more frequent multi-agent collaboration, the demand for memory bandwidth and capacity grows exponentially. At this stage, evaluating AI computing power cannot rely solely on GPU chip performance. Precisely, computing power is never about a single chip's capability, but the overall operational efficiency of the entire system.
The GPU is the brightest star on the stage, but HBM, advanced packaging, network interconnection, power supply, and cooling systems—these seemingly low-key components—are the backstage key supporting the entire AI industry. If the backend infrastructure cannot keep up, even the strongest front-end GPU will suffer stuttering, efficiency loss, and system bottlenecks.
This is like a top-tier super restaurant: the GPU is the world's best chef, but if the kitchen lacks ingredients, warehouse deliveries are slow, serving is delayed, and utilities are unstable, even the most skilled chef can only wait idly, unable to perform.
The same logic applies to AI factories. No matter how expensive or advanced the GPU, if data cannot be fed in, transmitted, or cooled properly, overall system efficiency takes a massive hit. Therefore, what Jensen Huang locked in this time is not ordinary memory, but the core granary for the long-term development of the AI industry.
I recommend you like and bookmark this article. In the future, when we analyze NVIDIA, Micron, Broadcom, TSMC, AMD, and even the power, optical communication, and data center industries, we will use the same core judgment: the core opportunity in AI is never in the most obvious GPU, but in the bottleneck that most limits industry growth.
Why is SK Hynix the biggest direct beneficiary of this collaboration?
The answer is straightforward: it currently sits firmly in the first tier of the HBM industry and is NVIDIA's most core advanced memory partner for AI platforms.
More importantly, top executives of the SK Group have repeatedly stated clearly that AI-driven demand for advanced memory is extremely strong, and the high-end memory shortage is not a single-quarter or short-term phenomenon, but a long-term trend lasting several years.
These statements hide a core transformation in the industry. In the past, the biggest risk for the memory industry was "frenzied capacity expansion." Traditional DRAM and NAND are classic cyclical products: when shortages raise prices, all manufacturers expand capacity; once capacity is released and demand cools, inventories pile up, prices plunge, and profits swing dramatically. Veteran investors have a collective cyclical trauma when it comes to memory stocks—profits double in up cycles and losses are severe in down cycles, with extreme volatility.
But the current AI-driven memory cycle is completely different. It is not that cycles have disappeared, but that AI has redefined the pricing logic for high-end memory. In the past, memory cycles were driven by inventory cycles in smartphones, PCs, and ordinary servers; now demand for high-end HBM and data center DRAM is dominated entirely by AI capital expenditure.
This brings a qualitative change: high-end memory is no longer a simple up-and-down commodity, but a "strategic asset" with irreplaceability. SK Hynix's core competitive advantage is thus established.
SK Hynix's advantage lies not just in capacity, but in being deeply embedded in NVIDIA's core supply chain, with its technology roadmap fully tied to AI's future layout.
In the short term, it is the biggest winner from this collaboration news; in the medium term, as AI factories continue to expand and HBM supply and demand remain tight, its pricing power and industry scarcity will keep rising; but long-term competitiveness depends on the yield, power consumption control, and mass production progress of next-generation products like HBM4. Overall, SK Hynix has locked in a key AI bottleneck, with the only risk being strong competition from Samsung and Micron.
Now look at Micron: not just hype, earnings have truly materialized
Many think that if SK Hynix rises, Micron will be weak. This is a very one-sided view. In the short term, SK Hynix leads, but the supply gap in the high-end HBM track is huge, and Micron remains firmly in the core track.
Data does not lie: Micron's fiscal Q2 2026 revenue grew nearly three times year-over-year, with non-GAAP gross margin of 74.9%, and it guided next quarter's revenue at about $33.5 billion with gross margin reaching 81%. This level of profitability completely breaks the traditional perception of "memory cyclical stocks."
Micron's investment logic is never about "SK Hynix signs a deal, Micron rides the coattails." Instead, the NVIDIA-SK Hynix collaboration once again validates that "high-end memory is the lifeline of AI factories." As one of the world's top three HBM suppliers, Micron is inherently on the industry's core chain.
More importantly, Micron has already moved beyond the "pie-in-the-sky storytelling" stage; earnings are fully materialized. In the past, the market was most worried about the cyclical nature of memory stocks: gross margins shine at cyclical peaks but get cut in half during downturns. But the AI era logic has turned. Micron clearly states that in the AI wave, memory has become a strategic asset for tech giants.
These words carry significant weight! Customers are purchasing high-end memory not to restock or expand ordinary servers, but to support efficient AI platform operation and ensure full computing power utilization.
Micron also publicly expects that by 2026, the data center DRAM and NAND market will surpass the total consumer electronics market. Simply put, the demand center of gravity for the memory industry is formally shifting from smartphones and PCs to AI servers and data centers. The era dominated by consumer electronics has ended, and AI hardware has become the core driver.
Investing in Micron doesn't require blindly chasing hype; just look at core indicators: whether AI demand is truly materializing and whether earnings are consistently delivered. But risks must also be watched: Micron is fundamentally a memory company and cannot completely escape industry cycles. Key follow-up indicators: HBM capacity expansion speed, long-term order growth, whether high gross margins can be maintained, and whether price increases suppress end demand. Once AI capital expenditure cools, memory prices can still fall back; invest cautiously.
Samsung: the biggest variable in the industry, potential and risk coexist
Samsung, with its massive capital, capacity, and technology heritage, is a global memory giant but has fallen into an awkward lagging position in the current HBM race. The market's core observation point on Samsung is clear: can it quickly close the technology gap with SK Hynix and successfully tie into NVIDIA's AI ecosystem?
HBM competition is never about capacity scale; it's about hard strengths like yield, power consumption, cooling, and packaging. It's like an exam: reputation doesn't matter as much as solid scores. Currently, SK Hynix is far ahead in both technology and mass production.
But Samsung is by no means without a chance! It is pouring efforts into developing next-gen advanced memory products like HBM4. If it can break through technology bottlenecks, catch up with industry progress, and successfully penetrate NVIDIA's supply chain, the market competitive landscape will be completely rewritten. If it continues to lag, the high-end HBM market will become a duopoly between SK Hynix and Micron.
Samsung is the biggest X-factor in the entire track: once its technology lands and capacity releases, industry competition will heat up rapidly; if it continues to lag, the scarcity and pricing power of the existing duopoly will further increase.
AI industry logic fully upgraded: from GPU shortage hype to system efficiency battle
Looking back at this core news, we can see a phase shift in the industry: in AI Phase 1, the market hyped GPU shortages and computing power scarcity, with NVIDIA dominating and upstream/downstream stocks rising along. This was essentially a "scarcity theme play."
Entering Phase 2, a GPU alone can no longer sustain industry development. An AI factory is a complete system engineering project: memory, advanced packaging, optical network interconnection, power supply, cooling solutions, and server systems—every link is indispensable.
Future AI investing cannot just stare at NVIDIA's stock price; one must follow the demand direction of its next-gen AI platforms. The hottest themes often lack excess returns. Real explosive profit opportunities lie in segments that have not yet been fully hyped but are rigid bottlenecks for the industry. The core thinking of capital has shifted: after GPUs, which segment—HBM, advanced packaging, optical communication, etc.—will become the next core industry bottleneck?
But always remember: not everything with an AI label will rise. A true golden track must satisfy three rigid conditions: first, demand is real, not hype; second, expansion barriers are extremely high, so price increases don't flood in new capacity; third, earnings are backed by financial reports and orders—stocks with only hype but no actual earnings will eventually correct.
HBM perfectly matches all three conditions: on the demand side, it has rigid support from NVIDIA, cloud providers, and AI server makers; on the supply side, it requires high-end chips and advanced packaging, with long expansion cycles and very high barriers; on the earnings side, SK Hynix and Micron's financial reports have fully validated profitability.
Short-term risk warning: don't mistake long-term trends for short-term buying points
Key reminder: don't blindly chase highs or impulsively buy just because you understand the industry logic!
Short-term stock price volatility is driven by market expectations and sentiment, not long-term industry logic. No matter how perfect the thematic logic, if the stock has already risen too much and valuation is bubbly, a violent sell-off can still happen in the short term. The June 5 semiconductor sell-off is the most real case: AI themes don't only go up; once capital crowds in and profit-taking sentiment rises, leading names become the escape hatch for fund outflows.
The surface reasons for this semiconductor crash were strong employment data leading to expectations of high interest rates, Broadcom earnings concerns, and market doubts about AI spending paying off. The core reason was that prior AI tech stocks had accumulated too much gain, with overly concentrated positions, and short-term sentiment reversal triggered a stampede sell-off.
Mature investors never rush in on good news or panic sell on short-term plunges. The core of investing is to distinguish, amid market noise, which logic is disproven and which trend is strengthened.
Five key follow-up signals that will determine AI storage trends
To grasp the AI Phase 2 rally, keep a close eye on five key signals:
First, track NVIDIA's platform iteration progress. Don't just look at stock price; focus on the rollout pace of next-gen platforms like Blackwell and Vera Rubin, as well as personal AI PCs and robot computing platforms, to confirm whether long-term advanced memory demand continues to expand.
Second, track SK Hynix's capacity and orders. Continuously watch whether it repeatedly releases messages of HBM supply shortage, continues capacity expansion, and signs long-term locked-price orders. A leading manufacturer daring to make large capital expenditures is the best proof of real long-term demand.
Third, track Micron's earnings data. Skip the simple revenue numbers; focus on HBM shipment share, cloud storage business growth rate, gross margin sustainability, and free cash flow. Sustained high gross margin indicates the company has successfully transformed from a traditional cyclical stock to a core AI infrastructure player.
Fourth, track Samsung's technology catch-up progress. If Samsung achieves a technology breakthrough and HBM4 mass production, it will significantly ease supply tightness and intensify competition; if it continues to lag, the scarcity and pricing power of SK Hynix and Micron will keep rising.
Fifth, track the memory price structure. Reasonable HBM price increases benefit suppliers, but if ordinary DRAM, NAND, and consumer electronics memory prices rise too fast, they will suppress end demand and plant seeds for a cyclical reversal. The core of AI storage investing is to judge whether price increases are supported by real demand and can persist, not to blindly follow the price hike theme.
Industry divergence intensifies: AI storage and ordinary storage are two completely different stories
In the long run, the memory industry will see extreme divergence. Not all memory products will enjoy the AI dividend. HBM, data center DRAM, enterprise SSDs, and high-capacity AI-specific NAND are completely separate tracks from traditional consumer electronics memory.
A boom in AI does not mean all memory companies profit. Those that can secure a valuation premium are the high-end memory products that can enter AI servers, AI accelerators, large data centers, and next-gen intelligent computing platforms. Ordinary memory for phones and PCs, even with short-term price rebounds, is unlikely to achieve high valuations.
Therefore, when analyzing SK Hynix, Micron, and Samsung, don't just look at the word "memory"; focus on product mix: HBM shipment volume, share of top tech customers, long-term locked order quantity, technology packaging strength, and the ability to maintain stable profitability and cash flow through cyclical swings. This is the core investment logic of the AI storage track.
AI industry enters the "counting pennies" stage, earnings delivery is king
This collaboration between NVIDIA and SK Hynix marks the official transition of the AI industry from Phase 1 ("imagination and expectation hype") to Phase 2 ("counting earnings, profits, and cash flow").
In the past, the market was willing to pay for AI's future vision, valuing long-term growth space. Now, capital is more pragmatic, caring only about real landing issues: Can the GPU run at full load after purchase? Can data center power and cooling support it? Can model inference costs keep falling? Can cloud providers' AI investments generate positive returns? Can the supply chain deliver stably?
This also means that the ultimate winners of AI Phase 2 are no longer the companies that best tell stories or paint a rosy picture, but those that can convert industry demand into real revenue, stable profits, and ample cash flow.
Micron's value is not in empty bullish talk on AI, but in consecutively beating earnings expectations, proving that AI storage demand has truly materialized and is being delivered. The NVIDIA-SK Hynix collaboration is not an empty optimistic view on the industry, but an early lock-in of the technology roadmap and supply system for the next several years.
The biggest taboo in investing is emotional hype without earnings support. The earnings delivery path for advanced AI storage is now clearly visible, but risks cannot be ignored.
Five core risks that must always be watched
First, valuation bubble risk. The long-term industry logic holds, but the overall semiconductor sector has high valuations, making it susceptible to sharp corrections from interest rate fluctuations, capital sentiment, and single-quarter earnings reports.
Second, AI investment return risk. If cloud giants like Microsoft, Amazon, and Google find AI spending hard to recoup and cut capital expenditure, the entire AI hardware supply chain will cool down.
Third, supply-demand reversal cycle risk. The current HBM supply shortage is real, but with SK Hynix and Micron continuing to expand capacity and Samsung accelerating, a few years later capacity may be released en masse. If demand growth falls short of expectations, the memory industry will again sink into a cyclical trough.
Fourth, customer concentration risk. The global HBM core large customers are highly concentrated on NVIDIA. While suppliers have scarcity, they are constrained by core customers in technology validation, product roadmap, and pricing power, posing a risk to bargaining power.
Fifth, technology iteration risk. From HBM3E to HBM4 to future generations, the pace of technology iteration is extremely fast. Current technology leadership cannot be sustained long-term and may be overturned by new technologies or new players at any time.
Conclusion: understanding the core competition of AI Phase 2
Overall, the core logic of the AI rally has completely turned:
In the past, you only needed to ask, "Will NVIDIA rise?" Now you must ask, "Where are the core bottlenecks of AI factories?"
In the past, only the GPU single chip was watched; now the focus is on the entire system behind the GPU: memory, packaging, interconnection, power, and cooling.
In the past, memory was seen as a high-volatility cyclical stock; now we must clearly distinguish between "ordinary consumer memory" and "AI advanced memory"—they follow completely different valuation logic.
The long-term collaboration between NVIDIA and SK Hynix is on the surface a supply chain cooperation news, but in essence, it is a major signal of a phased upgrade in the AI industry: the next stage of AI competition is centered on system efficiency, bottleneck breakthroughs, and infrastructure construction.
In the Phase 2 industry landscape, SK Hynix benefits first with its technology and embedded advantages; Micron stays firmly in the core track with real earnings delivery; and Samsung, as the biggest variable, determines the intensity of industry competition.
For ordinary investors, the most prudent strategy is: don't chase short-term emotions, don't fear short-term pullbacks; keep a close eye on five indicators—earnings proof, order visibility, gross margin trends, capacity pace, and price transmission—to find balance between long-term industry trends and short-term risks.
In the future, once Micron and SK Hynix consistently beat earnings, Samsung lags in technology progress, and NVIDIA's next-gen AI platform demand fully explodes, AI advanced memory will officially take over from GPU as the core theme of AI Phase 2. Conversely, if AI capital expenditure cools or memory prices rise too fast to suppress demand, be cautious in position sizing to avoid short-term correction risks.
