先进微型器件 AMD ARK的股票故事 ARK Invest

00:00:00 Technology capex as a percent of GDP is now greater than spending during the Tech andtelecom bubble. The big four hypers scalealers alone are expected to spend more than 700dollars billion dollars in 2026 and that number is only expected to move higher in 2027,today Nvidia owns The AI accelerator market but competition is coming. AMD has alreadyshown it can take share in data center compute steadily gaining ground in server cpus since2017, now it's bringing AI accelerators to market that could challenge NVIDIA's dominancein one of the most important technology markets in the world. We think the TAM for AIsystems is going to triple by 2030 reaching nearly 1.5 trillion dollars in annual sales. AMDdesigns two of the most critical chips in modern data centers cpus that run generalworkloads and gpus that accelerate AI models. AMD's customers include all of the majorclouds, and internet giants like meta AMD owns. 00:01:00 No factories TSMC handles their manufacturing so the company really competes on chip andsystem design AMD's held almost zero server market share in 2017, when it reent enteredthe server CPU market and by q one two thousand and twenty six their share of server CPUrevenue grew to 46% of the market. AMD shipped faster, more power efficient processorson an annual cadence, while Intel kept missing deadlines. Each cloud instance family thatbuilt around am m d's epic processor line represents a multiy year procurement cycle.Agentic AI is a tailwind for AMD's core business. Cpus are becoming more relevant in The AIstory, as we have highly capable AI Agent s that command more traditional computeresources to get work done than the world of AI chatbots. AMD is running the sameplaybook they used to compete against Intel in the CPU market to try to compete againstNvidia in AI accelerators. They're coming to market with a rack scale solution called hel iOS. 00:02:00 In the second half of this year, that will compete directly with NVIDIA's offerings in themarket. Our research shows that AMD chips can already be more performant than Nvidiachips per dollar on some workloads, and we're optimistic on the demand for Helios givencustomer announcements with OpenAI, meta, and oracle. 00:02:21 We expect the competition in the accelerator space to put up a better fight than Intel hasover the last five years. Nvidia' has demonstrated execution quality that's equal to andequally as impressive as AMD's over that time period, and some of their largest targetcustomers like Google and Amazon are rapidly pursuing their own custom chip projects.With their breadth of portfolio across CPU and GPU architectures and the demonstrated execution of the AMD team, we think that they can capture a reasonable share of the futureAI compute TAM going off a very small base that they have today.