SK Hynix Just Raised $26.5 Billion on Wall Street. It Makes More Profit in One Quarter Than Meta Spends on Electricity in a Year.

Twenty-six and a half billion dollars. That is what investors paid Thursday for shares of a South Korean memory company most Americans have never heard of, making SK Hynix's Nasdaq debut the largest U.S. listing by a foreign company in history. Demand exceeded supply by a factor of seven. Baillie Gifford, Coatue Management, and Situational Awareness Partners alone indicated interest in up to $7 billion worth of shares, according to Bloomberg. When trading begins Friday under the ticker SKHY, it will be the second-largest share sale of 2026, trailing only SpaceX's $85.7 billion IPO last month.

Nobody covering the listing has answered a basic question: what does SK Hynix's financial performance actually tell us about who captures the value in the AI economy, and the answer reveals an uncomfortable truth about who really profits from AI. Memory makers are extracting a larger share of every AI dollar than the companies building the AI models themselves, and the market structure makes that extraction nearly impossible to disrupt.

Numbers That Should Make Every AI CEO Uncomfortable

SK Hynix's first quarter of 2026 was not merely good but historically absurd: revenue tripled year-over-year to $34.5 billion, net income quintupled to $26.5 billion, and operating margin hit 72 percent, the highest in the company's history. For perspective, that single quarter of net profit, $26.5 billion, exceeds the $10.6 billion Meta will spend on electricity in all of 2027 to run its planned 14 gigawatts of computing infrastructure.

SK Hynix is not an outlier, because every major memory producer is printing money at a rate that dwarfs every other link in the AI supply chain. Samsung's operating profit rose 19-fold last quarter, and Micron's profit margins climbed to nearly 85 percent, up from 38 percent a year ago, according to Investopedia's analysis of the company's filings. All three memory stocks are up between 130 and 250 percent in 2026 alone, a synchronized rally that would be remarkable in any sector but is almost unheard of in an industry where overcapacity crises have destroyed shareholder value roughly once every five years for three decades.

These margins would be extraordinary in any industry. In semiconductors? Completely without precedent, a level of profitability that even Intel at its monopoly peak in the Pentium era never achieved. Nvidia, the company most associated with the AI boom, reported a 57 percent net margin in its most recent quarter. SK Hynix's 72 percent operating margin sits above even that, and unlike Nvidia, SK Hynix does not design the chips that define what AI can do. It never has. It makes the memory that those chips need to function, and that distinction is crucial: SK Hynix's profit comes not from innovation leadership but from controlling a physical bottleneck.

Memory Tax: 35 to 40 Cents of Every AI Dollar

Bank of America analysts estimate that memory will represent 35 to 40 percent of the $1.5 trillion that Big Tech companies are projected to spend on cloud and AI infrastructure in 2027, according to Barron's reporting on the firm's research. Apply that range to the total and the number becomes staggering: between $525 billion and $600 billion flowing to memory makers in a single year, more than the combined annual revenues of Boeing, Ford, and General Electric.

Three companies will divide that pile, and SK Hynix controls 56.4 percent of the high-bandwidth memory market, according to its own SEC filing. Samsung holds roughly 15 percent and Micron takes the rest. No other company manufactures HBM at commercial scale, and no fourth entrant on the horizon could begin shipping competitive product before 2029 at the earliest, because the capital expenditure required to build an advanced memory fab from scratch now exceeds $20 billion and the process takes four years from groundbreaking to first silicon.

Company	HBM Market Share	Q1 2026 Operating Margin	2026 Stock Performance
SK Hynix	56.4%	72%	+235%
Samsung	~15%	Not disclosed (profit up 19x)	+130%
Micron	~21%	~85% gross margin	+250%

At SK Hynix's 72 percent operating margin, its share of that $525-600 billion memory market translates to approximately $213 billion to $244 billion in annual operating profit from AI-related memory alone. Put differently: a single company that most people have never heard of could generate more operating profit from selling memory chips to AI companies than the entire U.S. airline industry generates in revenue.

This concentration makes the memory market function less like a competitive technology sector and more like a natural resource cartel. At least OPEC members can cheat on production quotas; building a new HBM fabrication facility costs $15-20 billion and takes three to four years. There is no cheating, only waiting.

$30.8 Billion in Memory IPOs in One Week

Five days after SK Hynix begins trading in New York, ChangXin Memory Technologies will open book building for a $4.3 billion IPO on the Shanghai Stock Exchange, according to Reuters data cited by CoinDesk. CXMT is China's largest DRAM manufacturer, and it is also on the Pentagon's list of Chinese military companies, making it ineligible for U.S. investment. Two exchanges. Two visions of who controls memory. One week.

What this juxtaposition reveals is stark: one memory company raises capital in New York, where it will use the proceeds to build an HBM packaging plant in West Lafayette, Indiana, subsidized by $458 million in CHIPS Act funding, $500 million in federal loans, and $700 million in Indiana state incentives. Its counterpart raises capital in Shanghai, where it will use the proceeds to upgrade production lines that posted first-quarter revenue of 50.8 billion yuan (up 700 percent year-over-year) while remaining barred from purchasing advanced lithography equipment from ASML under U.S. export controls.

Combined, $30.8 billion in memory chip offerings in a single week is the capital markets telling you something unambiguous: the memory supply chain is the choke point of the AI economy, and both superpowers know it. Washington is paying a Korean company $1.66 billion in subsidies to package memory chips on American soil. China is funding its own memory champion to manufacture DRAM without access to Western equipment. Mirror images. Neither approach addresses the fundamental problem, which is that three companies, and only three, can make the memory that AI requires.

Korea's Discount Is Real, and It Costs Billions

One of the more revealing data points in SK Hynix's listing is the valuation gap with Micron. SK Hynix trades at a 12-month forward price-to-earnings ratio of 5.5x. Micron trades at 6.66x. That 21 percent valuation discount persists despite SK Hynix holding 56.4 percent of the HBM market compared to Micron's approximately 21 percent, as MarketWatch detailed.

Fundamentals do not explain the gap, because the discount is about geography. Korean equities have traded at persistent discounts to U.S. peers for decades, a phenomenon analysts call the "Korea discount," driven by lower shareholder protections, conglomerate governance structures, and the inconvenience of trading in a timezone twelve hours ahead of New York. SK Hynix's U.S. listing is explicitly designed to close that gap. If SKHY trades at Micron's multiple, the company's market capitalization, already above $1.3 trillion in Seoul, would rise by roughly $280 billion on paper.

That is not a small number; it is larger than the annual GDP of Finland.

Memory's Price Is Everyone's Price

Contract prices for 16-gigabyte DDR5 DRAM chips quadrupled year-over-year and are expected to rise a further 55 to 60 percent this quarter, according to TrendForce. Supply is so tight that Apple, which normally dictates terms to its suppliers, has begun testing DRAM from ChangXin Memory Technologies, the same Pentagon-blacklisted Chinese firm raising capital in Shanghai, according to the Financial Times. Apple simultaneously raised MacBook and iPad prices by 20 percent in June, citing memory costs.

Ripple effects extend far beyond consumer electronics, reaching into every autonomous vehicle running a perception stack, every hospital deploying a diagnostic model, and every research lab training on datasets too large to fit in conventional DRAM. Every AI query, every chatbot response, every image generation, every autonomous vehicle perception cycle depends on memory bandwidth. HBM is not a commodity that can be substituted. It is stacked, bonded, and integrated into the processor package itself. Nvidia's next-generation Vera Rubin platform is designed around SK Hynix's HBM chips specifically. No HBM, no Vera Rubin. No Vera Rubin, no next-generation AI training. Absolute dependency. No substitutes exist.

Barron's analysts note that "little additional capacity is coming online until midway through 2027 at the earliest." SK Hynix's own management told investors that demand will outstrip supply "for at least three more years." That is a monopolist's timeline, not a commodity producer's.

Strongest Counterargument

Memory is the most cyclical sector in semiconductors, and anyone buying SKHY at a 7x-oversubscribed listing is betting that this time is different. Think it cannot crash? History offers little comfort: DRAM supercycles in 1993-95, 1999-2000, 2017-18, and 2020-22 all ended in crashes that wiped 50 to 80 percent of peak-to-trough value from memory stocks. Samsung has already cautioned that HBM3E supply may "briefly outpace demand," language that, in memory industry parlance, typically precedes a pricing correction.

Bulls argue that structural demand from AI is fundamentally different from PC and smartphone cycles. That may ultimately prove correct, and the structural argument deserves more respect than previous cycle-peak rationalizations received. But the companies making the bull case (SK Hynix, Samsung, and Micron) are the same companies that made the same structural demand argument in 2018 (cloud computing), 2020 (5G), and 2022 (the metaverse). Every cycle ends, and whether AI has genuinely broken the pattern or merely delayed the reckoning, and a 72 percent operating margin is the kind of signal that historically attracts the capital investment that eventually creates oversupply.

Limitations

The $213-244 billion operating profit estimate for SK Hynix's AI-related memory sales in 2027 assumes that BofA's $1.5 trillion cloud/AI infrastructure spending forecast is accurate, that the 35-40 percent memory share holds, that SK Hynix maintains its current 56.4 percent HBM share, and that operating margins remain near 72 percent. All four assumptions could prove wildly optimistic or wildly pessimistic, and the history of the memory industry suggests that analysts have been wrong about the direction of the cycle more often than they have been right, particularly at inflection points where consensus confidence is highest. Infrastructure spending could slow if AI models become more efficient. Memory's share of total spending could shift as custom silicon (like Meta's Iris chip) reduces reliance on general-purpose GPUs. Samsung's HBM production ramp could erode SK Hynix's share. And margins at 72 percent are, by any historical standard, unsustainable; the question is timing, not direction. CXMT's revenue figure (50.8 billion yuan, +700% YoY) is self-reported and has not been independently audited by Western accounting firms.

Bottom Line

AI's economy has a landlord problem. Rent is due. Three companies — SK Hynix, Samsung, and Micron — own the land, and every AI company in the world pays rent. SK Hynix's $26.5 billion Nasdaq listing is not just a capital raise. It is the moment when the memory monopoly's economics became visible to the largest pool of investors on earth, and those investors were so eager to buy in that they oversubscribed by 7x.

For investors, here is what to watch: SKHY's Friday debut will likely trade at a premium to its Seoul-listed shares, but the real signal is what happens to Micron. If the listing compresses Micron's valuation premium rather than expanding SK Hynix's, it means U.S. investors are reallocating within the memory sector rather than growing it, a sign that the smart money considers peak memory pricing already in the rearview mirror. If both rise, the supercycle has legs. Watch the spread.

For the tech industry at large, the lesson is simpler. Every dollar spent on AI infrastructure is, in significant part, a dollar paid to three companies in a supply chain that nobody can replicate in less than four years. AI's boom is real. But the biggest winners may not be the companies building the intelligence, because the ones selling the memory it runs on have figured out something that Silicon Valley, for all its disruption mythology, has never managed to disrupt: a three-company oligopoly with 72 percent margins and a four-year moat.