AMD (AMD) has become the fourth US chip maker to reach trillion-dollar status, after Nvidia (NVDA), Broadcom (AVGO), and Micron (MU). These four companies show where demand sits in the AI market.
Two drivers behind AMD's rise
AMD (AMD) is being pushed up by two things. First, CPUs. Second, AMD is now seen as the second source for AI factories. This is more than chips - it is a full system. Companies like HPE (HPE) and Lenovo are building these systems now and will ship them to customers at the end of this year and start of next. The market recognizes AMD as the second source for hyperscalers and frontier AI companies building out AI factories.
Much of AMD's move this week tied to Muse and Meta (META) as a big customer. Muse is a huge driver. Agentic AI - where the AI does tasks for you instead of just talking - becomes as normal as turning on a light once you use it daily. When people see AI actually get things done, usage climbs.
Why usage drives CPU demand
That usage puts pressure on CPUs. CPUs are the conductor of the orchestra - the chip that tells all other chips what to do. GPUs (AI chips) run very fast, but need CPUs to control them and give instructions. Cisco (CSCO) describes it this way: humans click but agents swarm. One person can run a hundred agents at once - going through old emails, reaching out to old contacts, doing many tasks. That pattern drives the need for more CPUs.
What has agentic AI replaced in daily use? It does not replace things. It makes possible things you would not have done before. Example: Amazon Quick (AMZN) does not book flights, it plans a two-week trip. Before agentic AI, no one would look at 15 companies across three cities and all the options, and an assistant's time would be better spent elsewhere. It makes possible what was not worth the time or effort before, and it raises the bar on what a good trip, a good deck, or a good email looks like.
CPU supply and pricing power
Intel (INTC) says it can only fill a certain share of its orders, especially for CPUs. This shortage was expected for Intel (INTC), AMD (AMD), and all companies that license from ARM (ARM). The key change now: software is catching up to what the chips can do. Recent tests of current-generation chips show software in some cases fully using the chip. This matters because Nvidia (NVDA) moved to a new generation every year, whereas AI technology used to get a new generation every couple of years, and all of it is now running.
Among the trillion-dollar members, much of what AMD (AMD) announced in the last few weeks of evaluations uses a type of computer-to-computer communication that relies heavily on Broadcom (AVGO). Broadcom (AVGO) enables much of the communication between these chips and also designs many of the custom chips the big hyperscalers use. Watch AMD (AMD) and Broadcom (AVGO) together.
Chipflation and pricing
A report, not confirmed by AMD (AMD), says it may need to raise prices. On pricing power: scarcity is the usual argument for raising prices. Memory costs have run up, and AI factories run on huge amounts of memory, so prices there rose from scarcity. With CPU demand so strong and supply short, AMD (AMD), Intel (INTC), and others now have pricing power over CPUs they did not have a couple of years ago. This is chipflation.
The data center is the customer
For AMD (AMD), the data center is the customer. These giant systems do not go to enterprise customers first. When AMD's Helios launches, one of the first customers will likely be Oracle (ORCL), which is essentially a hyperscaler. The buyers are the big hyperscalers, the frontier labs, and then large enterprises.
The business is building an end-to-end solution, not just selling chips to put in servers. When Nvidia's (NVDA) Vera Rubin platform launched, the first units were delivered by DeLorean a couple of months ago. The plumbing in these systems is enormous because they must be cooled, so companies have to plan cooling across their data centers. It becomes a large infrastructure story rather than a product-delivery story.
