What TradingKey 美股財經 said in this video:
- AI Capexneutrallong-termThe host pours cold water: ten times more AI agents does not mean ten times more CPUs, since software gets leaner and chips faster; demand durability depends on how many tasks have real paying customers behind them.Source
- ARMneutralmid-termThe host notes Arm's shift from licensing to selling finished chips carries real costs — manufacturing, packaging and inventory risk, plus competing with its own licensees; the two businesses should be judged separately until chip revenue and margins materialize.Source
- METAneutrallong-termThe host cites Meta as the cloud giants' hedging playbook — co-developing with Arm, being the first large standalone NVIDIA Grace customer, and Qualcomm's first public server CPU customer — more suppliers means more pricing leverage.Source
- QCOMneutralThe host mentions in passing that Meta is the first public major customer of Qualcomm's server CPU, as an example of multi-sourcing, without stating a directional view.Source
- AI Capexneutralmid-termThe host's own call: short-term follow IDC — supply stays tight for one to two years and whoever has capacity earns first; long-term follow Gartner — once capacity catches up, new AI-agent workloads get carved away by new architectures, while decade-old core databases stay put.Source
- NVDAneutrallong-termThe host argues NVIDIA's CPU push mainly raises revenue per AI factory rather than chasing standalone CPU profit; with quarterly revenue near $100B, CPU alone cannot drive the growth story, and the key metric is standalone Vera adoption outside NVIDIA racks.Source
- AMDneutralmid-termThe host advises watching AMD's revenue and gross margin together: share loss does not equal revenue loss in a growing market, but defending orders with big price cuts may not please shareholders; its competitive set has widened from Intel alone to the whole Arm camp.Source