AMD's nearly 30% September run briefly lifted its market value past $1 trillion

Cramer's case for the agent-driven leg was about CPUs. "Earlier this year, the focus for investors shifted from AI model training to inference and agentics, and it became clear that demand for CPUs was going to explode, because you need more CPUs to run useful AI agents," he said ^[8]. CPUs are the market where, in his words, AMD "remains the leader" ^[7].

A gain of almost 30% in September ^[5] that ends near $1 trillion ^[4] began from about $770 billion, so the month added roughly $230 billion of market value ^[1]. Cramer acknowledged that a run this large leaves the stock open to pullbacks ^[15].

On accelerators, Cramer's case rests partly on the leader's shortfall. "Nvidia literally can't make their own chips fast enough to satisfy the demand. Of course there was room for AMD here," he said ^[10]. He also said the market is large enough to support multiple winners ^[11]. The two claims fail under different conditions. The first stops holding once Nvidia's supply catches up with demand; the second does not depend on Nvidia at all. CNBC did not report AMD's revenue, its backlog, or the size and timing of the Meta and OpenAI commitments ^[9].

The long-run figures hold a smaller puzzle. At its brief peak, AMD's market value was about 400 times what it was when Su took over ^[2], while the shares are up more than 18,000%, or about 181 times [c13, d3]. Market value is price times share count, so the count grew roughly 2.2 times over those years. A holder from October 2014 now owns a bit under half the fraction of the company they started with ^[4].

Neither commentator discloses a position in AMD [c3, c14]. Cramer's Charitable Trust owns Intel, Meta and Nvidia ^[14]. The author of the Seeking Alpha piece, headlined "AMD: The Scary $1 Trillion Question" ^[2], is long Nvidia, TSMC, Alphabet and Meta ^[3]. Both own Meta, AMD's GPU customer and the company behind the agent launch [c6, c9].

I think the September buying paid in advance for agent-driven CPU demand, and the evidence here supports that view with moderate confidence. The counter-thesis is Cramer's, and it wins in either of two ways: server-CPU sales pick up fast enough that the $230 billion looks early, or the Meta and OpenAI orders convert on a schedule that carries the price on accelerators alone. If Nvidia's supply catches up while agent demand stays thin, the September gain is what the stock stands to lose. A quarter of data-center CPU growth that tracks agent adoption would prove this view wrong.