▲ All images in this article are sourced from the internet. Please contact us for removal if any infringement occurs.

This Is Not a Cycle—It's a Restructuring

This article was first published in Shadow Memo. Written by Mo Yingsheng.

This is not a cycle—it's a restructuring. Nor is it a binary choice between bubble and myth, as it is both simultaneously.

In June 2026, the consumer electronics industry witnessed an unprecedented wave of price hikes.

Apple raised global prices for MacBooks and iPads by up to $500. Microsoft announced a third price hike for Xbox starting August 1, with some models increasing by $150. In the domestic market, major manufacturers like Xiaomi, OPPO, Vivo, and Honor followed suit with price adjustments.

The capital market's reaction mirrored the true flow of this wealth redistribution. Apple's stock price dropped 6.12%, wiping out $263.5 billion in market value overnight. Meanwhile, memory chip manufacturer Micron Technology saw its stock surge over 15%...

What Lies Behind the Data?

In Q1 2026, contract prices for commodity DRAM surged 93-98% quarter-over-quarter, while NAND flash memory rose 55-60%.

By Q2, DRAM prices climbed another 58-63%, and NAND flash memory outpaced them with 70-75% gains—both marking record quarterly increases for the industry. The spot market was even more extreme, with DDR5 16GB modules soaring up to 627% in a year.

The macroeconomic impact was staggering. The World Semiconductor Trade Statistics organization projected the global semiconductor market would exceed $1.51 trillion in 2026, up 90% year-over-year. Memory chips alone would surge 250% to over $800 billion in value.

But these figures only show 'how much' prices rose, not 'why this time is different.'

Past memory price hikes were driven by inventory cycles in smartphones and PCs, with rallies typically lasting 3-6 months before collapsing after inventory replenishment.

This cycle, however, represents a fundamental shift. It is no longer an inventory correction but a long-term structural shortage triggered by explosive AI computing power demand.

The key to understanding everything lies in the word 'structural.'

Wang Junjie, Executive Secretary-General of the China Semiconductor Industry Association, bluntly stated at the 2026 Shenzhen Integrated Circuit Summit that the semiconductor industry had entered a 'super cycle.' But this super cycle is not a prosperity 'where everyone gets a piece of the pie.' He posed a stark question: 'Is this truly a healthy boom, or a structural illusion amplified by AI?'

The term 'structural illusion' merits reflection.

Who Is Creating the 'Structural Shortage'?

HBM Is 'Devouring' the Entire Memory World.

AI servers' demand for HBM (High Bandwidth Memory) resembles a ravenous beast.

According to SemiAnalysis, DRAM supply will fall short of demand by ~7% in 2026, with HBM facing a 6% deficit—widening to 9% in 2027.

TrendForce revealed at a June semiconductor summit that HBM would remain in short supply until 2027.

The issue is that HBM consumes several times more wafers than traditional DRAM.

Faced with nearly unlimited purchasing willingness from AI chip firms, Samsung, SK Hynix, and Micron—the three memory giants—have allocated 70% of their new or reallocatable capacity to HBM.

The result: a sudden shortage of consumer-grade memory chips for smartphones, PCs, and gaming consoles.

But this is just the first layer. The deeper issue is that memory is transforming from a 'standardized commodity' into a 'customized strategic asset.'

A China Merchants Securities report noted that starting with HBM4, base dies will adopt logic processes, while new architectures like CXL, SOC AMM, and 3D DRAM emerge continuously. Memory products are accelerating their shift from standardization to customization.

This means the industry's commoditization and cyclical volatility are weakening. Price hikes are no longer just temporary supply-demand outcomes but inevitable products of structural industry changes.

Capacity Locked by 'Long-Term Agreements': The Free Market Is Disappearing.

Traditionally, the memory industry relied on quarterly short-term orders. Any market fluctuation allowed end customers to cancel orders directly, often triggering price collapses.

Not anymore.

Micron has secured 16 'take-or-pay' contracts spanning 3-5 years, locking in tens of billions in guaranteed revenue and receiving $18 billion in cash deposits upfront. These agreements cover 20% of its DRAM and one-third of its NAND capacity. Analysts predict this locked proportion could rise to 70%.

Samsung and SK Hynix are adopting a 'customer selection' model—prioritizing long-term supply agreements with PC industry giants.

This means substantial consumer-grade capacity has been 'pre-purchased.' Even if prices have room to decline, locked capacities won't return to the spot market. Legal contracts are replacing the free market's price discovery mechanism.

The Bottleneck Extends Beyond Memory: A 'Full-Chain Shortage.'

Reducing this AI hardware price surge to 'just memory hikes' oversimplifies the issue.

AI shortages have evolved from initial high-end GPU constraints into full-supply-chain tensions. At the chip manufacturing level, TSMC's 5/4nm advanced nodes now operate beyond 120% utilization—overloaded.

Advanced packaging has become the most glaring bottleneck. Mainstream processes like CoWoS may face 20-30% supply-demand gaps in 2026, leaving many AI chips undeliverable due to packaging capacity shortages.

Surprisingly, electricity has emerged as a new constraint. Overseas data centers face a 45GW power deficit from 2025-2028. Even with sufficient chip and packaging capacity, computing clusters cannot launch.

From wafer fabrication to advanced packaging, from HBM to power infrastructure, this is a systemic, industry-wide bottleneck—not a temporary shortage in a single segment.

Who's Celebrating and Who's Paying the Price?

The Winners' Revelry!

Memory manufacturers' performance has reached astonishing levels.

Micron's Q3 FY2026 revenue hit $41.46 billion, up 346% YoY—a fifth consecutive record quarter. Net profit surged nearly 14-fold to $28.2 billion, with an 84.9% gross margin—an industry record surpassing NVIDIA's.

The entire memory chip industry saw 2026 sales jump 250% YoY to $803.9 billion. Yangtze Memory's NAND global share reached 13%, while ChangXin Memory's DRAM share hit 7.7-8%—both achieving mass profitability.

At NVIDIA's 2026 annual shareholder meeting, Jensen Huang declared the 'Age of Intelligent Agents' had arrived, defining this computational paradigm shift as the 'largest industry reset in 60 years.'

In his view, 'useful AI' is here, with tokens becoming billable, profitable production units. AI infrastructure building will enter multi-decade long cycles.

The Losers' Dilemma!

The picture on the other side is starkly different.

Apple began manufacturing A-series chips at TSMC in 2014 and became its largest customer in 2015, holding that top spot for over a decade.

By 2025, NVIDIA had replaced Apple as TSMC's largest customer. Now, among TSMC's top five clients—Apple and Qualcomm aside—NVIDIA, Broadcom, and AMD are all 'computing power arms dealers.'

Losing the top customer position means losing priority for foundry discounts and incremental capacity.

Tim Cook mentioned capacity constraints twice during earnings calls: 'We are indeed supply-limited due to advanced node availability for SoCs.' The former 'supply chain master' is now throttled by AI.

Lei Jun and Yu Chengdong also complained publicly. Xiaomi's Redmi K90 base model priced $15 higher than its predecessor—a significant hike amid intense domestic smartphone competition.

But this isn't the cruelest part.

The Crueler Truth: Growth Is 'Price-Driven,' Not 'Demand-Driven.'

Experts recently highlighted a critical signal: Memory shipments grew just 13.4% (vs. industry average 10.5%). The growth appears healthy but stems from price hikes, not real demand.

HBM prices have skyrocketed, and original manufacturers' profits have surged, but this doesn't indicate overall industry prosperity.

AI is 'devouring everything': Advanced nodes are occupied by AI chips, HBM capacity locked by large models, advanced packaging prioritized for computing power—resulting in traditional markets being 'squeezed out.'

Typical signs: Weak smartphone and PC demand, lower-than-expected CIS demand, and pressure on consumer electronics supply chains—even Top manufacturers (leading manufacturers) are not spared.

On one side, prices explode. On the other, demand shrinks. This 'shrinking volume, rising prices' combination has an unflattering economic term: stagflation.

Gartner projects DRAM and SSD prices will surge 130% by late 2026, raising PC prices by 17% and smartphones by 13%. However, global PC shipments are expected to drop 10.4%, and smartphones by 8.4%.

IDC is even more pessimistic, revising smartphone shipments down to a 12.9% annual decline and PCs to 11.3%.

Price hikes suppress demand, but demand weakness can't force price cuts since capacity is locked by long-term agreements.

This creates an unprecedented deadlock.

Bubble or Myth?

The market has split into two camps over this question.

The Bubble Camp's Warnings!

Nobel laureate Paul Krugman recently warned: 'We're in a state where the AI 'quasi-bubble' is about to burst.'

Bank of America strategist Michael Hartnett noted in a report that AI chip stock rallies now approach or exceed extremes seen during the 2000 dot-com bubble and 1720 French Mississippi Bubble.

Wharton finance professor Itay Goldstein pointed out that many signs indicate overvaluation, with Wall Street's top five tech giants collectively worth $18 trillion.

Yann LeCun, one of the 'AI Godfathers,' bluntly stated: 'The AI bubble could burst sooner than imagined. It could pop any time.'

The Bank for International Settlements also sounded the alarm: The top five hyperscale cloud providers expect capital expenditures exceeding $1 trillion from 2025-2026, but 'poor returns could trigger sudden financing withdrawals, turning the capital spending boom into a prolonged investment bust.'

Capital market signals are equally alarming. NVIDIA's stock dropped 14.19% in June, underperforming the S&P 500 year-to-date. On June 10, its market cap fell below $5 trillion. The Philadelphia Semiconductor Index plunged 10.3% in a single day—its worst performance since March 2020.

The Myth Camp's Faith!

The opposing voice is equally loud.

Ren Zeping fiercely responded on social media: 'Most people still don't understand the AI tech bull market. They think tech stock valuations are inflated and all hype because they fundamentally don't grasp the industry's endgame.' He called discussions of a bubble peak 'shortsighted and narrow-minded—such cognition deserves poverty.'

At NVIDIA's shareholder meeting, Jensen Huang argued that computing power is transforming from a cost center into a revenue generator. He emphasized that AI capital spending still has vast room to grow, as 'useful AI' is already here and profitable.

Goldman Sachs raised its global server market forecast to $1.1 trillion by 2028. Top U.S. cloud providers' cumulative capital expenditures from 2026-2028 will reach $1.145 trillion. Morgan Stanley predicts global cloud capital spending will exceed $1 trillion in 2026.

CITIC Securities offers a middle-ground perspective: This AI rally resembles the 2021 new energy cycle more than the 2000 dot-com bubble.

The industry trend is real, but capital market pricing may have outpaced fundamentals.

Choosing sides between these camps might miss a more fundamental truth.

Bubbles and myths are not mutually exclusive. They can both be true.

Historically, every true technological revolution has been accompanied by massive asset bubbles. The 19th-century railway mania, the 20th-century dot-com bubble—after most companies vanished post-bubble, railways and the internet reshaped the world.

Today's AI hardware frenzy is following the same script.

On the one hand, industrial transformation is real. AI is evolving from 'generative question-and-answer' to intelligent agents with autonomous planning capabilities, with memory demands shifting from linear growth to exponential leaps.

Storage is no longer a common accessory that follows the market trend but has been upgraded to the core infrastructure of AI computing power. Analysts from Goldman Sachs point out that high memory costs and the continuous growth of AI workloads are the core factors driving the upward revision of server market forecasts.

On the other hand, 'structural hallucination' is also real. Growth is driven by prices rather than real demand, production capacity is locked in by long-term agreements rather than market regulation, and winners take all while losers bear the pressure.

When Jensen Huang predicts that 'the annual investment in the future AI industry could reach $20 trillion,' should we pause and think: Is this number reasonable? Who will foot the bill?

What is even more alarming is whether AI can smoothly transition from the 'training' phase to large-scale 'inference' applications.

If inference demand explodes, hardware demand will gain a second growth curve; if inference deployment falls short of expectations, current hardware investments will become a huge surplus capacity.

But we all know: The future is not about telling stories but about delivery capabilities. Any company whose performance falls short of expectations will be swiftly corrected.

In Closing

Three years ago, Altman brought large models into the public eye. Three years later, this 'AI shockwave' has swept through data centers, memory factories, and end-user supply chains, impacting the entire consumer hardware industry.

There are no bystanders in this shockwave.

For ordinary consumers, buying a phone or computer in 2026 will cost several hundred or even over a thousand yuan more than in previous years. Economists from Columbia University estimate that by the end of 2032, total global AI construction expenditures could reach as high as $8 trillion, and 'silicon-based inflation' may have just begun.

For end-device manufacturers, the once-dominant 'supply chain bargaining power' is being stripped away by AI giants. From Apple to Xiaomi, from Huawei to Honor, all hardware companies without an 'AI label' are under dual pressure from cost squeezes and loss of bargaining power.

For investors, AI hardware represents both an opportunity and a trap. The industrial trend is clear, but valuation bubbles are equally real.

AI's impact on hardware is neither a pure bubble nor a perfect myth.

This is an ongoing industrial restructuring that is far from over. It is supported by real demand, chased by surplus capital, constrained by technological bottlenecks, and carries the risk of bubble bursting.

The only certainty is that this restructuring will not stop and wait for anyone.

Whether you like it or not, AI has already entered your wallet.

And this 'structural inflation' has only just begun.

I know you're watching.