Google’s Chip Push, SK Hynix’s $29 Billion Buyback

Ninety minutes in, tech was soft as Treasury support lowered yields. Marvell stood out, giving Google a $12.2 billion warrant and deeper chip-design ties.

The product list is supporting cast; the AI accelerator is Nvidia turf. Google is getting serious about silicon, with Marvell turning its IP into TSMC-made chips.

Two readings of the warrant: backing a valuable customer can grow business, but if the partnership is attractive, why hand value back?

For Alphabet, this is about enough supply. The test is whether AI funding creates capacity or recirculates capital; cash flow and ROIC will tell us. I favor a barbell: incumbents plus edge beneficiaries, especially industrial implementation. Salesforce is embedded and sticky; we avoid Micron because free cash flow has been uneven.

SK Hynix plans a rapid $29 billion repurchase by mid-November, cancelling shares. Memory pricing could support bigger returns; US ADRs let it sell high and repurchase cheaper Korean stock.

Anthropic’s revolving facility could exceed $10 billion as banks seek a role before an IPO. It is liquidity, not borrowed cash: a reserve for surprises.

It is dangerous when one person thinks they know better than everyone else. We need collective governance, not a CEO God complex—but AI can help disease, education, and aging.

Local resistance to data centers crosses party lines over power, water, and AI; 75 projects were delayed in three months. Oracle’s Project Jupiter is sponsoring rodeos, food kitchens, and parks to look neighborly.

British PC prices alone are not chipflation. AI costs may show up in services; hardware is small and quality-sensitive, while July inflation was household energy. Model prices vary widely, and Europe lacks US-scale energy and water pressure.

Fuse is commercializing fusion from national-lab and security investment. Our generators produce radiation for electronics and materials testing; this $100 million builds more units and hosts more customers in Albuquerque. Fusion is now a commercialization race: we need generators, not PowerPoints today.