AI

The Rise of China’s Hottest New Commodity: AI Tokens

Published

3 months ago

March 26, 2026

Imagine a new global commodity traded not in barrels or bushels, but in trillions of invisible computational units — weightless, borderless, and already reshaping the architecture of economic power. In the summer of 1858, a copper-core cable crossed the Atlantic seabed and rewired who controlled the flow of value across empires. In the spring of 2026, something structurally similar is happening, only the cable is digital, the commodity is China’s AI tokens, and the empire building is happening in plain sight.

The numbers are now difficult to ignore. China’s daily consumption of tokens — the tiny data units processed by AI models — has surpassed 140 trillion as of March 2026, a more than 1,000-fold increase from the 100 billion recorded at the beginning of 2024, and over 40 percent higher than the 100 trillion logged at the end of last year. China.org.cn Liu Liehong, administrator of China’s National Data Administration, announced the figure publicly and framed it not as a technical milestone but as a strategic one. The surge, he said, signals China’s AI industry “evolving from basic chat functions to more sophisticated systems capable of decision-making and task execution.” This is bureaucratic language with a geopolitical subtext: China is no longer catching up in artificial intelligence. It is setting the pace in the metric that matters most — actual usage, at scale, in the real economy.

From OpenRouter to the World: How China’s AI Tokens Surpassed the US

The clearest empirical signal of this shift has come from an unexpected source: OpenRouter, a San Francisco-based API aggregation platform that functions as a kind of global stock exchange for large language models. OpenRouter data published on February 24, 2026, shows that models built in China account for 61% of total token consumption among the platform’s top ten most-used models, with aggregate consumption reaching 5.3 trillion tokens out of a combined 8.7 trillion. Dataconomy The three most-consumed models that week were all Chinese. MiniMax M2.5 claimed the top position with 2.45 trillion tokens consumed in a single week — a 197% increase from the prior week. Moonshot AI’s Kimi K2.5 followed with 1.21 trillion tokens, and Zhipu AI’s GLM-5 placed third with 780 billion tokens, itself up 158%. TechBriefly

The historical reversal was swift and decisive. In the first week of February 2026, the weekly call volume of Chinese models had jumped to 2.27 trillion tokens, sending a strong signal of pursuit. Just one week later, Chinese models officially surpassed their US counterparts with 4.12 trillion tokens versus 2.94 trillion. By the week of February 16th, Chinese models had soared to 5.16 trillion tokens — a 127% increase in three weeks. 36Kr The growth is structural, not episodic, and it has been observed at the highest levels of the American venture capital industry. Andreessen Horowitz partner Martin Casado estimated that roughly 80% of startups using open-source AI stacks are running Chinese models. TechBriefly OpenRouter COO Chris Clark put the dynamic plainly: Chinese open-weight models have gained large market share because they are “disproportionately heavy in agentic flows run by U.S. firms.”

Ciyuan: When a Nation Brands Its Commodity

Beijing has never been content to let economic transformations arrive without a conceptual framework to accompany them. At the 2026 China Development Forum, Liu Liehong used the term ciyuan as the official Chinese translation for “token” during a speech on AI development, effectively resolving a debate within China over how the term should be rendered. South China Morning Post The naming is deliberate and worth examining. In Chinese, ci translates to “word,” while yuan carries double meaning: it is the basic unit of Chinese currency, and the suffix used when naming most foreign currencies in Mandarin. Liu said the token, or ciyuan, was not only a value anchor for the intelligent era but also a “settlement unit” linking technological supply with commercial demand, thereby allowing business models to be quantified. South China Morning Post

The People’s Daily had introduced the concept in January, describing ciyuans as the smallest unit of information processed by large models — possessing characteristics “emergent in the intelligent era” of being quantifiable, priceable, and tradable, with a new value system centered on their invocation, distribution, and settlement rapidly taking shape. TechFlow The semantic move is not accidental. China is not simply producing more AI tokens than the United States. It is trying to name, define, and ultimately govern the unit of account for the next phase of the global technology economy. Jensen Huang arrived at the same conceptual destination independently. At Nvidia’s GTC developer conference last week in San Jose, clad in his trademark leather jacket, Huang told the audience that “tokens are the new commodity,” declaring that Nvidia should no longer be seen mainly as a chip maker but as a builder of what he calls “AI factories” that produce tokens in large numbers. South China Morning Post Two of the world’s most consequential technology figures, one American and one Chinese, are now converging on the same metaphor — which suggests the metaphor is correct.

The Structural Edge: Electricity, Architecture, and the Token Economy

China’s dominance in China’s AI tokens is not a speculative narrative driven by state media hype or a single viral product launch. It rests on compounding structural advantages that are difficult to reverse quickly through policy alone.

The most fundamental is energy. China’s total electricity costs are approximately 40% lower than in the United States — a physical cost advantage that competitors cannot easily replicate. China Academy When a developer anywhere in the world calls a Chinese AI model’s API, the request is processed in a Chinese data center powered by the Chinese grid. The economic value of that electricity is exported globally as a high-margin digital service — one that bypasses customs, evades tariffs, and barely registers in conventional trade statistics. Industry estimates suggest that converting raw electricity into AI processing services can increase its value by up to 22 times compared to simply exporting electricity at the grid rate. China.org.cn China’s western regions — Xinjiang, Inner Mongolia, Yunnan — provide abundant, low-cost renewable energy at scale. The country has also built a vertically integrated supply chain spanning ultra-high-voltage transmission equipment, liquid-cooled data centers, and server assembly that few rivals can match.

The second advantage is architectural. Chinese AI laboratories have pioneered efficiency-first model design under the pressure of US chip export restrictions. DeepSeek V3’s Mixture-of-Experts architecture activates only a fraction of the model’s parameters during inference, with independent tests showing its inference cost is roughly 36 times lower than GPT-4o. MiniMax M2.5, despite having 229 billion total parameters, activates only 10 billion during inference. China Academy These are not merely clever engineering choices. They are the product of operating under genuine resource constraints — constraints that have paradoxically made Chinese models leaner, cheaper, and more deployable at global scale.

The third advantage is price. MiniMax M2.5 charges $0.30 per million input tokens and $1.10 per million output tokens. By comparison, Claude Opus 4.6 costs $5 per million input tokens and $25 per million output tokens — roughly 10 to 20 times more expensive. TechBriefly In the new agentic AI era, where a single automated workflow can consume millions of tokens in a matter of hours, this price differential is not a marginal consideration. It is frequently the deciding factor. A Silicon Valley developer who once tested workflows with GPT-4 at tens of dollars a day has little rational reason not to switch when a Chinese alternative delivers comparable benchmark performance at a tenth of the cost.

Alibaba Token Hub and the Industrialization of Ciyuan

Corporate China has received the signal and reorganized accordingly. Alibaba has established a new internal division called the Alibaba Token Hub, directly overseen by Chief Executive Eddie Wu, moving the research team that develops its flagship Qwen models, the consumer-facing app division, and major AI-related products under a single unified structure. Bloomberg The unit will focus on creating, distributing, and applying tokens — the basic computing units used by AI models — while integrating several internal teams to cover the full AI stack, from foundation model development to enterprise-level AI applications. TechNode The naming of the division after the commodity it produces is itself a statement of intent. Alibaba is not building an AI company. It is building a token factory.

The reorganization lands against a backdrop of surging Chinese AI cloud pricing that reflects genuine demand pressure. Alibaba Cloud announced price increases on select services effective April 18, 2026, citing global AI demand, rising supply-chain costs, and sharp increases in token call volume. Baidu Smart Cloud made an identical announcement the same day. Zhipu launched a new agent-optimized model and simultaneously raised its API price by 20% on March 16th. Tencent Cloud adjusted billing strategies for its intelligent agent development platform starting March 13th. 36Kr When Chinese AI providers raise prices in unison, it is not a cartel behavior — it is a market clearing mechanism. The supply of ciyuans is being consumed faster than it can be provisioned, and the price signal is propagating through the ecosystem.

A report jointly released by Andreessen Horowitz and OpenRouter shows that the total token call volume of Alibaba’s Qwen series ranks second globally at 5.59 trillion, second only to DeepSeek’s 14.37 trillion. 36Kr These are not vanity metrics: they represent real developer adoption, real API revenue, and real geopolitical influence embedded in the codebases of companies that may scale into tomorrow’s global technology infrastructure.

The Counterpoints: Profitability, Chip Constraints, and Sovereign Risk

Honest analysis demands acknowledgment of what the token volume data does not tell us. Market share on OpenRouter — a platform beloved by independent developers and AI hobbyists rather than large enterprise procurement departments — does not translate automatically into enterprise dominance. The main battleground for corporate AI workloads remains, for now, in the hands of American providers offering the accountability, compliance tooling, and integration depth that large institutions require. OpenRouter represents a thin slice of the global AI market; its developer-skewed demographics mean the 61% figure overstates Chinese penetration of the full economy.

The profitability question is equally live. Aggressive token pricing is partly a land-grab strategy — buying market share at margins that may not be sustainable. The simultaneous wave of Chinese cloud price increases in March 2026 suggests the economics are tightening. DeepSeek’s inference costs may be radically lower than GPT-4o’s, but training costs, talent costs, and the escalating expense of acquiring increasingly scarce advanced chips under US export restrictions are real. Washington’s ongoing efforts to tighten the chip embargo — extending restrictions to additional Nvidia architectures and closing loopholes used to route chips through third-country entities — represent a genuine long-run constraint on China’s ability to scale inference capacity. And sovereign risk is not zero. Developers in regulated industries and allied governments face real legal and reputational exposure from routing sensitive workloads through Chinese infrastructure, regardless of how cheap or fast those tokens may be.

Token Exports as a New Form of Digital Soft Power

Yet the strategic logic of China’s position is more durable than its critics typically concede. Tokens are intangible, bypass customs, evade tariffs, and don’t appear in official trade statistics. China exports massive compute and electricity services, yet it remains virtually invisible in trade data. China Academy This invisibility is a feature, not a bug. Token exports occupy a legal and regulatory grey zone that trade hawks find difficult to target. You cannot sanction a token. You cannot put a tariff on an API call. The infrastructure that produces the tokens — the data centers, the power grid, the model weights — sits firmly within Chinese sovereignty and beyond the reach of extraterritorial enforcement.

Beijing appears to understand this clearly. China has named 2026 the “Year of Data Element Value Release,” is building a single national data market with unified property rights, and by end of 2025 had compiled over 100,000 high-quality datasets totaling more than 890 petabytes — roughly 310 times the digital collection of the National Library of China. MEXC The scale of data assembly, combined with cheap inference, low-cost energy, and rapid model iteration cycles, constitutes a vertically integrated token economy that took China’s industrial sector decades to assemble in steel or semiconductors — and that is being assembled in AI in a matter of years.

Chinese artificial intelligence service stocks rallied this week after state media highlighted a sharp increase in domestic AI model adoption and a surge in the token usage they generate. Bloomberg The market’s reaction is rational. Investors are pricing in what economists have been slow to formally model: that the token, like oil before it, will become a commodity whose production geography matters enormously to the distribution of global wealth. The country that most cheaply produces what the world most needs will, history suggests, extract durable rents. In the oil era, that was the Persian Gulf. In the token era, the early evidence points unmistakably toward the Yangtze River Delta, the Pearl River Delta, and the data centers of Guizhou province humming with renewable hydropower.

The British Empire laid the cables. The rest, as they say, was history. The question now is who controls the flow — and at what price per million tokens.

Discover more from The Economy

Subscribe to get the latest posts sent to your email.

Up Next

PSX Sheds Nearly 3,500 Points as Iran Rejects US-Backed Ceasefire: Geopolitical Shockwaves Hit Pakistan’s Markets

Don't Miss

Scott Bessent’s Fed Overhaul: How the Bank of England Blueprint Is Reshaping U.S. Central Bank Independence Under Trump

Click to comment

AI

AI Memory Chip Shortage 2026: Nvidia, Apple & What Comes Next

Published

12 hours ago

June 27, 2026

Abdul Rahman

A global memory chip shortage is hitting AI hyperscalers, tanking Nvidia and Apple shares, and triggering a Wall Street rotation. Here’s what the AI sector’s supply crisis means for investors.The artificial intelligence boom that has driven Wall Street’s most extraordinary bull run in a generation is running headlong into a physical constraint: the world cannot produce memory chips fast enough to feed it.

On Friday, June 26, 2026, technology stocks extended a brutal weekly decline even as the broader market stabilized and advancing shares outnumbered declining ones. Nvidia slipped another 1% in early trading and was on pace for an 8% weekly loss—its worst five-day stretch in more than a year. Apple dived after announcing price increases for several iPad and Mac models, citing higher costs from memory chip shortages. Oracle and CoreWeave fell after the New York Times reported that OpenAI was considering delaying its initial public offering to as late as 2027.

What the headlines share is a single underlying cause: the cost of the memory chips that power AI infrastructure is rising faster than even the most aggressive hyperscaler budgets assumed, and the shortage driving that cost increase is not expected to ease before 2028.

The Architecture of the Crisis

Memory chips—specifically the high-bandwidth memory, or HBM, used in AI accelerators—are produced by a small number of manufacturers: SK Hynix, Micron, and Samsung. Demand for HBM has exploded because each new generation of Nvidia’s AI chips requires substantially more of it. As Nvidia pushes its product cycle faster to maintain competitive advantage, each cycle pulls forward enormous new demand for chips that take 18 to 24 months to ramp in production.

Micron reported strong quarterly earnings—its results have been spectacular—but the very strength of those results is the problem for the rest of the tech sector. Micron’s margins are rising because memory is scarce and expensive. The companies buying that memory—Microsoft, Amazon, Alphabet, Meta, and the rest of the hyperscaler complex—are absorbing higher input costs on a scale that is beginning to show up in margin guidance.

Analysts at Charles Schwab noted a “growing wedge” in the technology sector between memory producers like Micron—which is posting massive gains—and the hyperscaler stocks that are watching their AI infrastructure economics deteriorate. The latter group includes names like Microsoft, Amazon, and Alphabet, which are collectively projected to spend between $660 billion and $700 billion on AI infrastructure in 2026, according to research from Fair Observer.

Nvidia’s Problem Is a Market Concentration Problem

Nvidia entered 2026 having crossed a $5 trillion market capitalization—larger by GDP comparison than all but four national economies. That concentration made the stock not merely a bet on AI but a systemic weight in the S&P 500. Nvidia and its mega-cap technology peers now account for roughly 30% of the entire index—the highest concentration in half a century.

When Nvidia corrects, it does not correct in isolation. It reprices the risk premium of every fund manager with an S&P 500 benchmark, which is nearly every institutional investor in the world. The 8% weekly decline in late June—attributed to a combination of rising memory costs, margin anxiety among hyperscaler customers, and a broader rotation away from high-multiple AI stocks—had ripple effects across semiconductor infrastructure names including Lumentum, Marvell Technology, and Corning.

Apple Raises Prices—and Reveals the Exposure

Apple’s announcement of price increases for iPad and Mac models was notable for two reasons. First, Apple’s supply chain is among the most sophisticated on earth; if Apple could not absorb memory cost increases without raising consumer prices, the margin pressure is acute. Second, Apple’s pricing decision revealed an exposure that consumer electronics companies had managed to keep largely invisible through inventory buffers.

Those buffers, built up when memory was cheap, are now depleted. The shortage is forecast to persist through 2027 and potentially into 2028, driven by Nvidia’s accelerated chip release cadence and the insatiable demand of AI data centers for high-bandwidth memory. Analysts at Briefing.com noted that higher memory costs are seen “persisting throughout 2027 and perhaps into 2028, driven by increasing data center demand and Nvidia’s rapid introduction of updated AI chips.”

OpenAI Delays Its IPO—Absorbing the Lesson From SpaceX

The reported delay in OpenAI’s public offering is a direct consequence of two market developments: the broader tech weakness driven by the memory supply crisis, and the troubled IPO debut of SpaceX earlier in June, whose shares suffered heavy losses in the days following listing as global markets repriced risk.

OpenAI executives, who had targeted 2026 for a public offering, are now said to be evaluating a 2027 launch—giving markets time to stabilize and giving the company time to demonstrate that its AI infrastructure economics are sustainable at the scale that a public market valuation would demand.

The Rotation That May Define the Rest of 2026

The most significant market dynamic emerging from the memory chip crisis is not the decline in any single stock but the rotation it is enabling. As the mega-cap AI trade faces margin headwinds, investors are moving into financial and industrial companies, healthcare, and energy—sectors that had been overshadowed for years by the AI growth narrative. The Dow, weighted toward those steadier names, was holding up even as the Nasdaq declined through the final week of June.

That divergence—Dow up, Nasdaq down—is a familiar pattern in sector rotation cycles. It does not necessarily signal a bear market. It may signal the beginning of a more broadly distributed bull market, one less concentrated in five or seven names. The memory supply crisis, in that reading, is not the end of the AI boom—it is the first serious test of whether the boom’s economics are durable enough to survive contact with physical constraints.

Discover more from The Economy

Subscribe to get the latest posts sent to your email.

AI

AI Energy Demand 2026: Data Centres, Power Grids & the $725B Infrastructure Boom

Published

21 hours ago

June 27, 2026

Abdul Rahman

Hyperscalers are spending $725 billion on AI infrastructure in 2026. The energy demands of this buildout are reshaping global power markets, utility valuations, and electricity costs. Here’s the full picture.

Behind every AI-generated image, every chatbot response, and every earnings forecast produced by a large language model is a data centre consuming electricity at a scale that is quietly reshaping global energy markets.

Microsoft, Google, Meta, and Amazon — the four hyperscaler giants powering the AI economy — are collectively spending more than $725 billion on AI infrastructure in 2026. This unprecedented wave of capital expenditure is building data centres that require power at a scale that has fundamentally changed the conversation around energy security, grid stability, electricity pricing, and the commercial viability of every power generation technology from natural gas to nuclear.

The AI energy story is not a footnote to the technology boom. It is one of the most consequential investment themes of the decade.

The Scale of the Demand Shock

To understand the magnitude of AI’s energy appetite, consider the trajectory. A single large AI training run — the computational process that creates a frontier model like those produced by OpenAI, Anthropic, or Google DeepMind — can consume more electricity than a medium-sized city uses in a month. Inference — the ongoing process of serving queries to users — multiplies that consumption across millions of simultaneous interactions.

OpenAI’s inference compute costs are projected at $14.1 billion for 2026. Inference compute is largely an energy and chip cost. The company’s gross margin of approximately 33% reflects how significant this load has become.

Across the hyperscalers, the $725 billion AI infrastructure budget funds:

Data centre construction — new campuses in the US, Europe, Southeast Asia, and the Middle East
Nvidia GPU procurement — the primary compute engine for AI workloads
Network infrastructure — high-speed interconnects between training clusters and inference nodes
Power infrastructure — substations, backup generation, and energy contracts

The power requirement for a modern AI training cluster can exceed 100 megawatts — enough to power approximately 80,000 US homes. Planned hyperscaler buildouts in 2026 will require gigawatts of additional generating capacity, much of which does not yet exist.

The Grid Cannot Keep Up

The fundamental constraint in the AI energy build is not capital or technology — it is the pace at which electrical grids can be upgraded to deliver power at the scale and reliability that data centres require.

In the United States, utilities are reporting data centre interconnection queues that extend three to five years into the future. The permitting and construction timelines for new transmission lines — often the binding constraint for connecting new power generation to load centres — have not accelerated at the pace of data centre demand.

In Northern Virginia — home to the world’s largest concentration of data centres — the PJM Interconnection grid has been grappling with the challenge of meeting rapidly growing load from AI campuses while maintaining reliability across the broader regional grid. Similar dynamics are playing out in Ireland, Singapore, and Texas.

The consequence: electricity prices in AI-intensive regions are rising as demand competes with existing industrial and residential load. This is not a temporary phenomenon — it reflects a structural demand shift that will persist for years as AI infrastructure deployment continues.

Who Wins in the AI Energy Build

The AI energy story is generating a distinct set of investment winners that extend well beyond the semiconductor and software sectors.

Utilities

Electric utilities with significant exposure to data centre load — particularly in Virginia, Texas, Georgia, and Ohio — are seeing accelerated earnings growth as hyperscalers sign long-term power purchase agreements. These agreements provide utilities with revenue visibility that justifies capital investment in generation and transmission capacity.

Dominion Energy (Virginia), AEP (Ohio and Texas), and Duke Energy (Georgia) are among the utilities that have flagged data centre load as a material driver of near-term demand growth.

Data Centre REITs

Real estate investment trusts focused on data centre infrastructure are trading at premium valuations as institutional capital seeks AI infrastructure exposure without the technology risk of individual semiconductor or AI software companies.

Equinix, Digital Realty, and Iron Mountain have seen significant demand from hyperscalers seeking colocation capacity. The constraint on their growth is increasingly power availability rather than capital.

Nuclear Energy Operators

Nuclear power has emerged as the preferred baseload generation technology for hyperscalers seeking 24/7 carbon-free electricity. Microsoft has signed a deal with Constellation Energy to restart the Three Mile Island nuclear plant in Pennsylvania specifically for data centre power. Amazon and Google have made direct investments in nuclear start-ups building small modular reactors.

Nuclear’s appeal for data centres is straightforward: it provides continuous, dispatchable power without the intermittency of solar and wind — a critical feature for high-reliability compute workloads.

Natural Gas Operators

In the near term — before new nuclear capacity comes online and before renewable build catches up with demand — natural gas is filling the gap. Gas-fired generation is being commissioned specifically to serve data centre load in multiple US markets. This has created demand for both gas generation capacity and for the pipeline infrastructure that delivers fuel to these plants.

The Geopolitical Dimension: AI Data Centres as Strategic Infrastructure

Governments increasingly view AI data centre capacity as strategic national infrastructure — comparable to port facilities, road networks, or military installations. The race to host hyperscaler AI infrastructure is shaping foreign investment policy, grid modernisation plans, and energy procurement strategies across Asia, Europe, and the Middle East.

Singapore, navigating its role as ASEAN chair in 2026, has positioned its AI infrastructure capacity as a key element of its regional leadership agenda. The city-state has approved new data centre construction after a moratorium, tying approvals to energy efficiency standards and renewable power commitments.

Saudi Arabia and the UAE have made massive commitments to attract AI infrastructure investment as part of their post-oil economic diversification strategies, offering land, regulatory expediting, and preferential power arrangements to major hyperscalers.

India is building AI data centre capacity at scale in Hyderabad, Mumbai, and Chennai, positioning itself as the primary alternative to Chinese AI infrastructure for global enterprises seeking supply chain diversification.

The Cost Pass-Through: Who Pays for AI’s Energy Appetite

The $725 billion AI infrastructure buildout is not self-contained. Its costs ripple through the economy in several ways:

Electricity price pressure: Rising data centre demand in grid-constrained markets pushes up wholesale power prices, increasing costs for all electricity consumers — industrial, commercial, and residential.

Enterprise AI licensing costs: The compute costs embedded in AI services translate directly into licensing fees for enterprise customers. Companies that have deployed AI copilots, coding assistants, and customer service automation are reporting costs that exceed initial projections — creating a “sticker shock” dynamic that is beginning to slow enterprise AI adoption.

Carbon accounting complexity: As hyperscalers procure renewable energy to offset data centre consumption, they are absorbing significant portions of new renewable generation capacity that might otherwise reduce costs for the broader grid. The interaction between data centre power procurement, renewable energy credits, and carbon markets is creating new complexities for corporate sustainability accounting.

The Investment Implications

The AI energy infrastructure theme represents one of the most durable and under-appreciated investment opportunities in the current cycle. While the market has priced AI enthusiasm into semiconductor and software valuations extensively, the downstream infrastructure beneficiaries — utilities, data centre REITs, nuclear operators, and gas pipeline companies — remain relatively less valued for the structural demand shift they are absorbing.

Key investment considerations:

Data centre REITs offer exposure to AI demand without the valuation risk of pure-play AI companies, with dividend income providing a return buffer
Regulated utilities in high-growth data centre markets offer earnings visibility supported by long-term power purchase agreements with investment-grade counterparties
Nuclear energy operators benefit from a structural shift in hyperscaler procurement strategy that is likely to persist for a decade
Grid infrastructure companies — transmission equipment manufacturers and engineering firms — are positioned for multi-year demand as utilities upgrade capacity to serve AI load

The Bottom Line

The $725 billion AI infrastructure buildout is not just an investment theme — it is a structural transformation of global energy markets. The data centres being built today will consume power for decades. The grid upgrades required to serve them will reshape electricity pricing, generation mix, and geopolitical energy strategy across the world’s major economies.

Investors who understand the energy dimension of the AI boom — not just the semiconductor and software dimensions — have access to investment opportunities that carry less valuation risk, more earnings visibility, and more durable competitive positions than the high-profile AI pure-plays currently commanding headlines.

FAQ

Q: How much energy do AI data centres use?
A: A single large AI training cluster can exceed 100 megawatts of power consumption. Across all hyperscalers, the collective AI infrastructure buildout of $725 billion in 2026 will add gigawatts of new demand to global electricity grids.

Q: What companies are building AI infrastructure in 2026?
A: Microsoft, Google, Meta, and Amazon are the four primary hyperscalers collectively spending over $725 billion on AI infrastructure. Nvidia supplies the primary GPU compute hardware. Data centre REITs including Equinix and Digital Realty provide co-location capacity.

Q: How is AI affecting electricity prices?
A: In grid-constrained regions with high data centre concentrations — particularly Northern Virginia, Texas, and Singapore — AI data centre demand is contributing to rising wholesale electricity prices. This affects all electricity consumers in these markets.

Q: Why are hyperscalers investing in nuclear energy for AI data centres?
A: Nuclear power provides continuous, dispatchable, carbon-free electricity — the ideal power source for high-reliability AI compute workloads that cannot tolerate intermittency. Microsoft, Amazon, and Google have all made commitments to nuclear generation specifically for data centre power.

Discover more from The Economy

Subscribe to get the latest posts sent to your email.

AI

AI Semiconductor Selloff 2026: Micron Crash, Nasdaq Pullback & What Comes Next

Published

2 days ago

June 26, 2026

Abdul Rahman

On June 24, 2026, Micron Technology shares fell 13% in a single session — the stock’s worst single-day performance since June 5. The memory chipmaker had become a proxy for AI infrastructure demand, a stock that had ridden the AI enthusiasm wave to gains that justified its premium valuation. When it fell, the signal it sent through technology markets was unmistakable: the AI trade is not a one-way bet.

The Micron crash was not an isolated event. It was the latest episode in a pattern of volatility that has characterised the Nasdaq Composite throughout 2026 — a market that has delivered extraordinary returns over the past three years while simultaneously exhibiting the kind of volatility that characterises late-stage speculative cycles.

Understanding what Micron’s collapse reveals — and what it doesn’t — is essential for investors navigating the most complex technology market environment since 1999.

What Actually Happened: The Micron Story

Micron reported fiscal third-quarter results after the close on June 25, 2026. The earnings release came after a session in which the stock had already declined sharply on what appeared to be pre-announcement anxiety. The 13% single-day drop on June 24 — before the results — reflected a combination of factors:

High expectations were embedded in the valuation. Micron had been one of the primary beneficiaries of the AI-driven memory boom, as high-bandwidth memory (HBM) — the type of memory chip most important for AI compute workloads — commands significant pricing premiums and rapid volume growth. A stock priced for perfection leaves no margin for disappointment.

South Korean technology stocks had already broken. The Kospi — South Korea’s benchmark index, heavily weighted toward semiconductor companies including Samsung and SK Hynix — had plunged approximately 10% in the period leading up to the Micron selloff. Given the integrated nature of the global memory supply chain, this was a significant signal.

The SpaceX IPO absorbed market attention and capital. With the SPCX listing consuming enormous institutional bandwidth — and with some evidence of portfolio rebalancing as money rotated into the new AI pure-play listing — technology sector positioning was unsettled heading into the Micron earnings window.

Wedbush Securities’ Dan Ives was among the bulls holding the line. Following his channel checks across Asia and enterprise AI demand trends, Ives saw “no cracks in the armor,” arguing that the South Korean selloff was more likely a pause after a near-100% Kospi rally in 2026 rather than a signal of weakening AI fundamentals. His view: “The selloff in South Korean technology stocks was more likely a pause after a near-100% rally in the Kospi this year, rather than a sign of weakening fundamentals.”

The distinction Ives draws — between valuation-driven volatility and fundamental deterioration — is the central analytical question for investors in AI semiconductors.

The Broader Tech Picture: Nasdaq in a Choppy Range

The Nasdaq Composite closed at 25,476.64 on June 24 — down 0.43% on the day — as the Micron selloff pulled the tech-heavy index lower. The S&P 500 declined 0.10% to 7,358.22, while the Dow Jones Industrial Average — dominated by financials and industrials rather than technology — actually gained 182 points, advancing 0.35%.

This divergence is important. It reflects the continued rotation dynamic that has characterised 2026 markets: investors moving from high-multiple technology and AI stocks into more stable financials, industrials, and defensive sectors. The Dow rising while the Nasdaq falls is a classic late-cycle rotation signal — not necessarily a precursor to a market crash, but a sign that the consensus AI enthusiasm is being repriced.

The Nasdaq’s trajectory in 2026 has been shaped by three conflicting forces:

Bull case: AI capex is real and accelerating ($725 billion from hyperscalers in 2026), enterprise adoption is proceeding even if slowly, and the SpaceX/OpenAI IPO wave is bringing new capital into AI-adjacent public markets.

Bear case: Valuations remain extended relative to earnings, the AI bubble concern is growing (the CEPR launched its AI Bubble Monitor in June), and earnings multiples across the semiconductor sector leave no margin for guidance disappointment.

Wild card: The Federal Reserve’s hawkish turn under Kevin Warsh. Higher-for-longer rates are unequivocally negative for high-multiple growth stocks — the precise companies that dominate the Nasdaq. If BofA’s forecast of three rate hikes materialises, the discount rate applied to future earnings rises, compressing multiples across technology.

Memory Chips Specifically: The Supply-Demand Calculus

Micron’s situation reflects a supply-demand dynamic in memory chips that is more complex than the simple “AI = buy semiconductors” narrative suggests.

High-bandwidth memory (HBM) for AI training and inference is in strong demand, with supply constrained by the technical complexity of the manufacturing process. This segment is performing well for Micron, Samsung, and SK Hynix.

Standard DRAM and NAND flash — the memory types used in conventional computing, consumer electronics, and data storage — remain in a more normalised supply-demand balance. Consumer electronics demand has not recovered to the peaks of the 2021–2022 pandemic era. PC refresh cycles are extending. Mobile upgrade rates are slowing.

The result is a bifurcated memory market where AI-specific products command premium pricing but represent a smaller share of overall revenue, while conventional memory faces ongoing pricing pressure. Investors who extrapolate AI demand across the entire semiconductor industry are making an analytical error.

The South Korea Kospi: A Canary or a Correction?

South Korea’s Kospi is among the most AI-intensive equity markets in the world, with Samsung Electronics and SK Hynix representing major index weights. The 100% Kospi rally in 2026 — before the recent pullback — was one of the most dramatic performances of any major market globally.

A near-100% rally in under a year, in a market concentrated in semiconductor names, followed by a 10% correction is — by historical standards — a healthy pause, not a fundamental reversal. But it deserves scrutiny.

The Kospi’s AI sensitivity cuts both ways. If AI infrastructure demand continues to accelerate, the South Korean memory supply chain is among the primary structural beneficiaries. If AI capital expenditure decelerates — whether from a bubble correction, enterprise budget fatigue, or recession — the Kospi would likely underperform global markets significantly.

Wedbush’s Ives is probably right that the 10% Kospi pullback is a pause, not a peak. But the risk scenario — where AI demand disappointment triggers a more serious Kospi correction — is the kind of fat tail that position sizing should account for.

Oil Prices and Tech: An Overlooked Correlation

One underappreciated dynamic in June 2026 tech markets is the negative correlation between oil price relief and technology performance. As Brent crude fell from elevated levels — reflecting Strait of Hormuz reopening optimism — energy sector stocks declined, while the capital freed from energy inflation concerns did not flow uniformly into technology.

Instead, falling oil prices reduced the inflation urgency that had been supporting gold and energy stocks, while simultaneously creating space for the Fed’s hawkish pivot to dominate the market narrative. The net effect on the Nasdaq was mildly negative, as rate-hike expectations offset the energy relief.

This interconnection illustrates a key feature of 2026 markets: macro factors are more dominant than sector fundamentals in driving short-term price action across equities. A portfolio manager who correctly identified Micron as a fundamentally sound business still lost 13% in a single session because macro sentiment — Fed hawkishness, oil-driven inflation dynamics, and South Korean contagion — overwhelmed the fundamental picture.

The Investment Outlook for AI Semiconductors

Despite the volatility, the long-term structural case for AI semiconductor demand remains intact. The $725 billion hyperscaler AI infrastructure buildout generates genuine and sustained demand for compute hardware. Nvidia’s GPU dominance in AI training is real. HBM demand from data centres will grow as AI models scale.

The relevant question is not whether to own AI semiconductors, but at what price and with what risk management.

The risk-adjusted approach for investors:

Avoid concentration in single names that are priced for perfect execution — a 13% single-day decline on pre-announcement anxiety illustrates the asymmetry of high-expectation positioning.

Consider broader index exposure through semiconductor ETFs (SOXX, SMH) rather than individual stock concentration, allowing participation in structural AI demand without maximum idiosyncratic risk.

Monitor HBM-specific positioning — the AI-specific memory segment that genuinely benefits from training demand — versus conventional memory exposure, which faces different supply-demand dynamics.

Watch the Fed. Three rate hikes by year-end would put meaningful pressure on Nasdaq multiples. The tech sector’s performance in 2H 2026 is as much a function of monetary policy as it is of AI earnings delivery.

Micron’s 13% crash is not the beginning of an AI semiconductor collapse. It is a reminder that valuation matters, expectations matter, and late-cycle technology markets are not immune to gravity.

The South Korean Kospi correction, the SPCX post-IPO decline of 17%, and the Nasdaq’s choppy performance in June 2026 are all consistent with a market that has priced AI excellence aggressively and is now requiring proof of delivery.

The AI semiconductor thesis is intact. The trade needs to earn its valuation — and the process of earning it will involve more of the volatility that June 2026 has delivered.

FAQ

Q: Why did Micron stock drop 13% in June 2026?
A: Micron fell 13% on June 24, 2026 — its worst session since June 5 — amid high earnings expectations, a broader AI semiconductor selloff that followed South Korean technology stock declines, and pre-announcement anxiety ahead of its quarterly results.

Q: Is the Nasdaq in a correction in 2026?
A: The Nasdaq has been volatile in 2026, with multiple single-session declines and a rotation dynamic away from high-multiple technology stocks. As of late June, the index has not entered formal correction territory (a 10% decline from highs), but valuations remain stretched relative to earnings.

Q: Should I buy semiconductor stocks in 2026?
A: The structural case for AI semiconductor demand remains intact, but individual stock selection and entry point matter significantly. Broad-based ETF exposure (SOXX, SMH) reduces idiosyncratic risk compared to single-name concentration. The Federal Reserve’s rate trajectory is a key near-term risk to watch.

Q: What happened to South Korean tech stocks in June 2026?
A: The South Korean Kospi fell approximately 10% from recent highs, with semiconductor-heavy names including Samsung and SK Hynix leading the decline. Most analysts characterised the move as a valuation-driven pause after a near-100% 2026 rally rather than a sign of fundamental AI demand deterioration.