TSMC goes all-in on America

The day's biggest number belongs to TSMC. The world's largest contract chipmaker has swelled its total planned U.S. investment to $265 billion, adding four new plants — a figure driven by a fresh $100 billion commitment to expand its American operations. The Arizona buildout now anchors what is becoming the most expensive foreign manufacturing bet in U.S. history.

The spending is a direct response to demand it simply can't keep up with. TSMC has lifted its 2026 outlook, telling investors that appetite for AI hardware is running ahead of what its factories can currently build. When the company that makes the chips says it can't build fast enough, that's the story of the year in one sentence.

Nvidia fights on every front

Nvidia is everywhere today — and fielding questions from all directions. It's deepening its AI push in Japan through a new factory partnership aimed at building out the country's computing capacity. It's also positioning itself at the front of the line for HBM4, the next generation of high-bandwidth memory that feeds data to AI processors, where one supplier is reportedly grabbing a striking 70% of orders.

Looking ahead, Nvidia is framing its next platform, Vera Rubin, around a metric it calls "intelligence per dollar" — the value an AI system delivers relative to what it costs to run — a pitch tuned for the emerging agentic AI era. But not everything is smooth: Nvidia is juggling delays on its consumer RTX 5000 Super graphics cards, questions about its flagship AI systems, and a fresh competitive threat carrying a $20 billion price tag.

That $20 billion rival has a name

The threat is Etched, the AI chip startup now in talks to raise money at a valuation of roughly $20 billion, according to an exclusive Wall Street Journal report. A specialized challenger reaching that scale signals just how much money is chasing alternatives to Nvidia's dominance.

Etched isn't the only one going in-house. Meta is pushing its own AI chip, codenamed Iris, into production, with the goal of doubling its computing power. Mark Zuckerberg's company making more of its own silicon is part of a broader hyperscaler drift away from buying every processor off the shelf.

The machines behind the machines

Upstream of all this sits ASML, the Dutch firm with a near-monopoly on the lithography machines used to print advanced chips. ASML wants to raise its prices — and its biggest customer, TSMC, is pushing back hard. It's a rare public friction between two links in a supply chain that usually moves in lockstep, and a reminder that even in a boom, someone is arguing over who captures the margin.

Chips as collateral

Finally, a sign of how deep AI has burrowed into finance: an AI inference startup, General Compute, has borrowed $400 million against an unusual form of collateral — the specialized chips it uses to run AI, as reported by TechCrunch's Tim Fernholz. When GPUs become the asset you pledge to Wall Street, the hardware isn't just powering the AI economy anymore. It's becoming its currency.

The through-line

Every thread today points the same way: demand for AI compute is outrunning supply, capital, and even the factories being built to catch up. TSMC can't build fast enough, Nvidia is racing to lock up memory and fend off well-funded rivals, hyperscalers are baking their own silicon, and lenders are treating chips like gold. The scramble is on — and no one in the supply chain is sitting still.