In the AI infrastructure boom, Nvidia (NASDAQ: NVDA) sells the brains of the AI factory while Micron (NASDAQ: MU) supplies the memory that keeps those brains fed with data, and that difference shapes which stock will benefit more from the current phase of this historic spending wave. In my view, Nvidia is the clearer winner because a greater fraction of every dollar of hyperscaler capex is spent on its accelerators than goes toward memory chips of the type that Micron manufactures. Micron still looks like a powerful second-derivative play, since AI servers can't be built without the high-bandwidth memory it supplies.
The money flow this year is wildly high. The hyperscalers themselves say they plan to spend hundreds of billions of dollars in 2026 alone to expand AI data centers, GPU clusters, networking, and power infrastructure, a sharp jump from already elevated 2025 levels. One estimate puts combined capex for Amazon, Microsoft, Alphabet, and Meta Platforms at around $700 billion, with roughly two-thirds of that directed toward AI infrastructure rather than traditional cloud. Within that budget, the largest line item is the AI server stack itself, where accelerated servers built around high-end GPUs drive most of the component revenue growth. And of course, the hyperscalers are not the only tech players building data centers now.
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »
Nvidia sits directly in the center of this buying spree. Its data center business now revolves around entire racks of AI computing power, not just single chips. Systems like the GB200 Grace Blackwell Superchip and GB200 NVL72 tie together dozens of CPUs and GPUs into logical accelerators that can train and serve trillion-parameter models more efficiently than the prior-generation Hopper platforms. Hyperscalers are lining up to deploy these systems in their AI clouds, with massive companies committing to offer GB200 NVL72 instances to customers who want to run large language models at scale. All this sounds dense, but the basic point is that Nvidia products are in steady demand.
Nvidia's roadmap also continues to push the limits of performance and memory. Architectures like Blackwell and its new Vera Rubin processors combine vast computing throughput with enormous pools of high bandwidth memory (HBM), turning racks into "AI factories." That keeps Nvidia at the absolute center of procurement decisions when cloud providers are calculating how many clusters they will need to handle their training and inferencing workloads in 2026 and beyond.
#memory #high #signal
The money flow this year is wildly high. The hyperscalers themselves say they plan to spend hundreds of billions of dollars in 2026 alone to expand AI data centers, GPU clusters, networking, and power infrastructure, a sharp jump from already elevated 2025 levels. One estimate puts combined capex for Amazon, Microsoft, Alphabet, and Meta Platforms at around $700 billion, with roughly two-thirds of that directed toward AI infrastructure rather than traditional cloud. Within that budget, the largest line item is the AI server stack itself, where accelerated servers built around high-end GPUs drive most of the component revenue growth. And of course, the hyperscalers are not the only tech players building data centers now.
Missed Nvidia in 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" signal flashed for a little-known chipmaker called Nvidia. For the first time in years, that same "Total Conviction" signal is flashing for a company 1/100th the size of Nvidia. Continue »
Nvidia sits directly in the center of this buying spree. Its data center business now revolves around entire racks of AI computing power, not just single chips. Systems like the GB200 Grace Blackwell Superchip and GB200 NVL72 tie together dozens of CPUs and GPUs into logical accelerators that can train and serve trillion-parameter models more efficiently than the prior-generation Hopper platforms. Hyperscalers are lining up to deploy these systems in their AI clouds, with massive companies committing to offer GB200 NVL72 instances to customers who want to run large language models at scale. All this sounds dense, but the basic point is that Nvidia products are in steady demand.
Nvidia's roadmap also continues to push the limits of performance and memory. Architectures like Blackwell and its new Vera Rubin processors combine vast computing throughput with enormous pools of high bandwidth memory (HBM), turning racks into "AI factories." That keeps Nvidia at the absolute center of procurement decisions when cloud providers are calculating how many clusters they will need to handle their training and inferencing workloads in 2026 and beyond.
#memory #high #signal
5 hours ago
