(00:00:00) Trillion-Dollar Selloff: Broadcom's Miss, SpaceX GPUs & HBM4E Race
(00:01:10) SpaceX GPU Rental Deal Formalized
(00:02:03) HBM4E Memory Race Accelerates
(00:02:53) Nvidia Bonds and Buyback Signal
(00:03:44) Qualcomm Tenstorrent and Marvell Moves
(00:04:32) Hyperscaler Capex Arms Race

A single guidance miss from Broadcom erased $1.2 trillion in semiconductor market cap in one session — Nvidia alone shed $280 billion, falling below a $5 trillion valuation. But the real story isn't the selloff. It's the six structural developments unfolding beneath it that are quietly redrawing the AI hardware landscape.

SpaceX has formalized a deal to rent 110,000 Nvidia GPUs to Google for $920 million per month through June 2029, cementing its role as a third infrastructure layer between chip manufacturers and hyperscalers. SK Hynix pulled HBM4E sample shipments forward by six months to compete with Samsung for qualification on Nvidia's Rubin Ultra — a GPU requiring 384GB of HBM capacity, a radical spec escalation. Nvidia raised $20 billion in its first bond issuance since 2021 despite generating $96.6 billion in free cash flow last fiscal year, signalling an undisclosed capex commitment or acquisition.

Qualcomm is in talks to acquire Tenstorrent for $8–10 billion in a hard pivot toward data centre AI chip design, while Jensen Huang publicly endorsed Marvell as the next trillion-dollar company — backed by a $2 billion Nvidia investment. And four hyperscalers — Meta, Amazon, Google, and Microsoft — are on track to spend $720 billion on chips and data centres in 2026 alone.

This episode breaks down what the Broadcom miss actually signals, whether SpaceX can deliver on its September GPU timeline, and what HBM4E yield data in H2 2026 will determine about the Rubin production ramp.

This episode includes AI-generated content.