#0成本拿2股SK海力士 A garbage stock, endured 13 years on the cold bench, today its market value surpasses Bitcoin!

$13.5 trillion. On June 22, South Korea’s chip manufacturer SK Hynix’s intraday market cap hit this number, pushing Bitcoin (about $1.29 trillion) down from the 16th spot on the global asset list—and also temporarily knocking Samsung Electronics off Korea’s top market cap position. Reading the news headlines, you might think: just another asset ranking change. But if you extend the timeline a bit, this story becomes absurd.

More than twenty years ago, this company was called "Hynix Semiconductor," with over $6 billion in debt, its stock price collapsed into worthless paper, even failing to sell to Micron after negotiations. It should have died in the aftershocks of the 2001 dot-com bubble. "Unwanted child," betting on an unwanted technology at the time, Hynix was a burden dumped by the Modern Group—annual sales over $13.5k, nearly $3 billion in losses, creditors took over, bouncing on the brink of bankruptcy. Everyone advised: sell quickly, recover as much as you can. But during those most humiliating years, the engineering team stubbornly focused on one thing: HBM, high-bandwidth memory.

What is this stuff used for? At the time—almost nothing. Stacking multiple layers of DRAM chips, using silicon vias for vertical interconnects, the process was insanely difficult, costly, and the market demand was zero. The entire industry saw it like watching a group of people building a dock in the desert. In 2012, a real turning point arrived. SK Group’s leader, Chey Tae-won, overruled the entire board, spent about $3 billion to buy the "garbage stock" Hynix, renamed it SK Hynix, and continued pouring money into that niche technology. The board probably thought the boss was crazy. His gamble was: AI will come someday, and AI not only needs computing power but also "data input/output speed"—that is, memory bandwidth. From the first generation of HBM to today’s HBM3E, this gamble has lasted nearly 13 years.

Until ChatGPT exploded, NVIDIA’s GPUs flooded the market, and HBM suddenly transformed from a "scholarly toy" into the critical component for AI servers. 72% operating profit margin—this isn’t hype, just look at the numbers, no fluff: SK Hynix Q1 revenue was 52.58 trillion won, operating profit 37.61 trillion won—profit margin of 72%. Analysts’ consensus for Q2 operating profit has been revised upward from around 50 trillion won in early April to 62–65 trillion, with optimistic predictions even exceeding 68 trillion. It is NVIDIA’s main HBM supplier, with over 60% global market share; Samsung and Samsung’s HBM3E yield rates are stuck at the certification stage, unable to keep up, Micron is behind but with a much smaller share. This isn’t "market sentiment," it’s real money on the order sheet. Management’s own words are straightforward: the structural memory shortage driven by AI will last for years, and the company must significantly increase capital expenditure to expand capacity. And the capacity expansion cycle for HBM? 2 to 3 years. Three companies share this cake—locked by technological barriers, physical production lines, and time. This scarcity cannot be replaced by white papers, tokenomics, or DAO votes.

So, who is the market betting on today?

This is the real meaning behind SK Hynix’s market cap surpassing Bitcoin’s—not a victory for Korean stocks, not a sensational headline about "chips beating coins," but a straightforward arithmetic problem the market has solved: which part of the AI value chain do you occupy? Do you have real orders? Is your supply bottleneck physical and irreversible? Are your profits tangible and visible? SK Hynix’s answer is four "yes"es. As for the Crypto AI track… frankly, most projects’ answers are still on PPT slides. The IC3 report jointly released by 13 universities including Cornell is polite but deadly: the integration of crypto and AI is still in early stages, decentralized computing power, data markets, governance—most are still just ideas. The representative project Bittensor’s TAO has fallen 20% over the past three months; its co-founder admitted on X that the underlying economic incentives are still dominated by the team, sacrificing decentralization to ensure iteration speed, and the core mechanism needs another year and a half of refinement. Closer to hardware, crypto mining companies aren’t doing much better—Bitcoin miners are entering a "surrender phase," with mining difficulty down over 20% from historic highs, and those claiming to pivot to AI computing face a short-term funding gap of about $50 billion. Currently, only about a quarter of the leased AI capacity has been delivered. In short: SK Hynix is selling the real shovels needed in the AI era, and the production of these shovels is locked by physical laws; what Crypto AI is selling, often, is still a "map of future mines." Maps can be very valuable—provided someone believes there’s gold in that mountain, and you’ve got the entry point.

Don’t rush to pick a side
I know some will argue after reading this: "Are you bearish on Bitcoin?"
No. Bitcoin’s narrative has never been about "72% operating profit"—it’s digital gold, a parallel liquidity system, a macro hedge tool. It shouldn’t be judged by "PE valuation models." The real question is: in this round, which has drained about $1.5 trillion in new debt issuance since 2022—roughly matching the same period’s dollar M2 growth—AI has almost consumed all the new liquidity, and Bitcoin hasn’t even had a sip. Arthur Hayes’ article "Reality Test" makes this point: it’s not just "AI’s money flowing back into crypto when it crashes," but that the IPOs of Anthropic and OpenAI will continue to siphon off liquidity. If the AI bubble bursts and credit tightens, Bitcoin will be sold off too. The more accurate interpretation of what’s happening now is—markets are pricing two types of scarcity: one is "code-level scarcity" (21 million coins, hardcoded in the protocol), the other is "physical scarcity" (HBM production lines taking three years to build, yield rates stuck, only three companies worldwide can produce). The former relies on faith; the latter relies on cement, lithography machines, and engineers working night shifts. When liquidity is no longer flooding the market, "certainty premiums" will outweigh "imagination premiums."

Looking back, where was the most brutal blow in this story?
It wasn’t SK Hynix defeating anyone. It was a company that should have gone bankrupt, surviving because of one thing: doing the hardest, most physical, no-shortcuts kind of work when no one believed in it. Chey Tae-won’s 2012 acquisition, opposed by the entire board, he pushed through regardless. From 2009 to 2022, HBM was pushed forward 13 years—countless times it could have been cut from budgets or declared "no commercial prospects," but it never stopped. That’s what makes "narrative-driven assets" uncomfortable—the ability to produce white papers, hold AMA sessions, and tweet about "decentralized computing power about to revolutionize everything" is easy. But if you don’t spend 13 years on the production line, confronting those dirty yield issues, packaging processes, and collaboration with TSMC’s substrates—you can’t hold anyone’s neck, and you’ll always be waiting for the next "hot spot," rather than becoming the reason for the hot spot.

SK Hynix’s market cap surpassing Bitcoin is a stark reminder: the AI infrastructure dividend is flowing to those who turn bottlenecks into moats. As for everyone else—including crypto AI and various "next-gen narratives"—it’s not that there’s no opportunity, but that you first need to answer this simple question: which part of the value chain are you stuck on?
Answer it, and market value will come to you.
Fail to answer, and you’ll just keep talking and waiting.