May 25 Members' Morning Brief: Trump Faces "Iran Dilemma" Over the Weekend, AMD CEO Bullish on CPUs

1. [Trump Faces "Iran Dilemma" Over the Weekend] On Sunday, Trump stated that the agreement with Iran is "not even fully negotiated yet," retracting his Saturday remarks that the deal was "basically agreed upon," and dismissed critics as "people who are meddling in things they know nothing about." Trump said the U.S. blockade of the Strait of Hormuz would remain in place until the agreement is reached.

Commentary: With the blockade entering its sixth week, domestic inflation pressure in the U.S. continues to build, and every fluctuation in oil prices is reflected through CPI in the political ledger of the midterm elections. The more Trump needs a "victory moment" to conclude this conflict, the fewer bargaining chips he holds at the negotiation table. Iran fully understands this. Tehran is in no rush, so it can be very firm on "sovereignty is non-negotiable," and is willing to accept a preliminary memorandum that defers the nuclear issue for another 60 days—it wanted to drag things out in the first place: every extra day delayed adds more pressure on Trump domestically, and reduces the room for concessions Iran might need to make.

2. [AMD CEO Bullish on CPU] At an AI forum in Taipei, AMD CEO Lisa Su delivered a signal that could prompt the entire semiconductor industry to recalibrate its expectations: Over the past 3-4 years, the CPU market has seen annual growth rates of only about 3%–4%, while almost all industry focus has been on GPU; but as AI shifts toward inference and agentic systems, the CPU sector will get on the "fast track," with annual growth averaging over 35% for the next five years. 

Commentary: During the training phase, GPUs are irreplaceable—this hasn't changed. But once workloads tilt toward inference—especially agentic AI—the system's bottleneck is no longer just matrix multiplication throughput. Agentic AI requires task orchestration, data transfer, tool invocation, context management, and parallel scheduling—these are precisely the strengths of CPUs and the weaknesses of GPUs.