AI Buys the Substrate: SK Hynix Lists, Anthropic Leases a Power Plant
In one July week, SK Hynix raised about $26.5 billion in the largest US listing ever by a foreign company, and Anthropic signed a $19 billion, 20-year lease for a 401 MW data center on a former Kentucky aluminum smelter. The scarcity in AI is moving from models to memory and megawatts.
Two deals landed within four days of each other in July 2026, and reported separately they look like unrelated financial news. Read together, they mark a shift in where the value and scarcity of artificial intelligence actually live.
The memory maker goes to Wall Street
On July 10, SK Hynix listed on the Nasdaq under the symbol SKHY and raised roughly 26.5 billion dollars, selling about 177.9 million American depositary shares at 149 dollars each. It was the largest US listing ever by a foreign company and the second-largest US share sale in history, behind only SpaceX. The shares closed their first day up about 12.8 percent, near 168 dollars, putting the company at a valuation around 1.27 trillion dollars.
SK Hynix is a memory company. It reported 2025 revenue of about 64.1 billion dollars and net income near 28.3 billion dollars — a net margin around 44 percent — and it controls roughly 60 percent of the global market for high-bandwidth memory by revenue. HBM is the stacked DRAM that sits beside the logic die inside almost every AI accelerator, and it is in acute shortage. The listing gives US investors direct exposure to what has become the tightest choke point in the AI supply chain.
SK Hynix Nasdaq listing
$26.5B raised
Largest US listing ever by a foreign company, second-largest US share sale after SpaceX; closed up about 12.8 percent at a valuation near $1.27 trillion
The AI lab becomes a twenty-year tenant
On July 6, Anthropic signed a 20-year lease worth about 19 billion dollars with TeraWulf — until recently a bitcoin miner — for the "Justified Data" campus in Hawesville, Kentucky. The 401-megawatt facility is built on a decommissioned aluminum smelter, with initial capacity expected in the second half of 2027 and full capacity by early 2028. The lease is expected to generate about 19 billion dollars of contracted revenue over its initial term, backed by investment-grade credit, and it follows Anthropic's roughly 65-billion-dollar funding round. On the same day, TeraWulf sold its majority stake in a Texas joint venture for 450 million dollars to redirect capital into wholly owned AI hosting.
The site choice is the tell. Aluminum smelting is one of the most electricity-intensive industrial processes in existence, so the smelter came with grid interconnection and a power envelope that would take a decade to build from scratch. AI is now resurrecting dead heavy industry for its electrical bones.
Anthropic and TeraWulf lease
$19B / 20 years
A 401 MW data center on a former Kentucky aluminum smelter; initial capacity H2 2027, full by early 2028, backed by investment-grade credit
Why they are one story
Both deals price the physical layer of AI, not the model layer. One sells direct access to the memory bottleneck at a trillion-dollar valuation; the other locks up two decades of electricity because power, not talent, now gates growth. The move comes in the same week that multiple labs shipped competitive frontier models within days of one another — a reminder that the model is increasingly the abundant stage while the substrate that runs it is the scarce one.
The bitcoin-miner angle is not incidental. Miners spent years mastering exactly what AI now needs: siting dense, power-hungry compute next to cheap, abundant electricity and running it around the clock. TeraWulf and Marathon, which acquired a two-gigawatt Texas site the same week, are converting that skill into a far more valuable output. A new intermediary tier — the neocloud — is forming between the labs that have the demand and the physical world that has the power.
For the full analysis of what this shift means for anyone building on AI, see the companion feature, The Physical Layer Turn: AI Scarcity Moves to Memory and Megawatts.