엔비디아도, 마이크론도 아니다. 2030년까지 보유할 AI 반도체 주식 1개라면?

Every GPU that powers a model training cluster has to be built somewhere, and that somewhere is almost always a fab operated by Taiwan Semiconductor Manufacturing (TSM -0.48%). Taiwan Semi does not design the artificial intelligence (AI) chips that get the headlines. Instead, the company manufactures them at nodes no one else can reliably hit at scale. That quiet competitive advantage has turned TSMC into the physical backbone of the AI infrastructure buildout. Nevertheless, the conversation still seems to start and end with Nvidia (NVDA +1.10%), Advanced Micro Devices (AMD +4.00%), or Micron Technology (MU +0.12%). The foundry that is actually printing the silicon rarely gets the same amount of airtime. This gap is strange once you understand how the AI chip stack works. While designers continue to fight over architecture and software, the constraint that decides how many of those designs actually ship is advanced-node capacity -- and TSMC owns most of that market. The foundry actually supporting the AI chip stack Nvidia and AMD design GPUs while Micron builds memory solutions layered alongside these chips. None of these companies do all of the manufacturing work to bring their products to life, however. TSMC holds roughly 73% of the pure-play foundry market as measured by revenue and reportedly closer to 90% of the market for the most advanced nodes. For years, Apple was credited as TSMC's largest customer. However, Nvidia took over that title earlier this year. This shift directly tracks the AI infrastructure cycle. High-performance computing, the area that covers AI accelerators and related server silicon, has become Taiwan Semi's core growth engine. The company's management points to AI-accelerator revenue compounding in the mid-to-high 50% range through the end of the decade, and lifted its longer-term company revenue growth outlook toward at least 25%. While competition from Samsung and Intel exist, both companies are spending heavily on scaling foundry operations and neither company is yet to close the gap on the yield and volume at nodes Nvidia and AMD actually need for current and next-generation accelerators. NYSE: TSM Key Data Points Taiwan Semi's growth is structural TSMC's financial growth supports the idea that the AI infrastructure supercycle is more than a temporary surge. Full-year 2025 revenue reached about $122 billion, up nearly 36% year over year. Meanwhile, gross margin climbed to 60% from 56% percent the year before. Through the first half of 2026, Taiwan Semi's revenue grew 37% year over year to $76.1 billion while while gross margin expanded into the mid-60% range, underscoring the level of pricing power the company commands right now. The company is also spending heavily on capital expenditures (capex) to help stay ahead of demand. Rising infrastructure spend might look expensive in any single year, but over a multi-year AI buildout it is what keeps the company's moat intact. Customers cannot easily dual-source manufacturing for leading edge nodes. This makes Taiwan Semi's revenue and profit tailwinds tied to the duration of the infrastructure supercycle itself rather than to one specific product launch. Taiwan Semi stock could be poised for a run The chart below illustrates how investors are treating TSMC as a picks-and-shovels name that should trade at a discount to some of the primary chip designers. Trailing price-to-earnings (P/E) puts the stock near 34 times, which looks rich upon first glance. But on a forward P/E basis, Taiwan Semi is a different story. While the company's forward P/E of 28 is not a deep-value multiple per se, it is also not stretched against a company still expected to grow revenue by a high double-digit percent range and compound earnings as margins hold. Nvidia and AMD capture more of the narrative premium because their products are visible. For the time being, Micron will ride the memory upcycle that AI also requires. Taiwan Semi sits underneath all of them. While geopolitical risk around Taiwan is an obvious factor to consider, in addition to the cost of building fabs here in the U.S., they do not erase a foundry position that remains the hardest part of the AI supply chain to replicate. On a forward earnings valuation basis, I think TSMC stock still looks like a strong candidate to buy and hold throughout the AI infrastructure era.