3 AI infrastructure stocks trading at bargain valuations

The tech bull market has been largely driven by artificial intelligence growth stocks, particularly in infrastructure, yet several AI-focused companies still trade at bargain valuations despite strong gains in the sector.

Nvidia, the poster child of the AI infrastructure boom and one of the stock market's biggest drivers over the past five years, carries a forward P/E ratio of just 16 times analyst earnings consensus for fiscal 2028 (ending January 2028) while continuing to grow rapidly. The company is the dominant semiconductor stock for AI model training, and with most foundational AI code written on its CUDA software platform for its graphics processing units (GPUs), that position appears secure. Nvidia is also well positioned as inference and agentic AI become more important. Earlier this year, Nvidia acquired Groq for $20 billion（約3.2兆円）. Groq's language processing units (LPUs) are ideal for the decode phase of inference to speed up response times when answering queries. Nvidia incorporated them into its CUDA ecosystem, so it can now offer complete systems that combine its GPUs, LPUs, central processing units (CPUs), and networking gear into servers designed specifically for inference, while also offering systems for training, agentic AI, and AI storage.

Micron Technology has been one of the best growth stories over the past year, with revenue more than quadrupling and gross margin exploding from 37.7% to 84.6%, yet its stock trades at a forward P/E of just above 6 times analyst estimates for fiscal 2027 (ending August 2027). The reason for Micron's low P/E is that the memory market has historically been very cyclical. However, there are good reasons to believe the current DRAM supercycle has legs. Volume growth in the DRAM (dynamic random access memory) market is largely being powered by high-bandwidth memory (HBM), which gets packaged with GPUs and other AI chips. With demand soaring, the big three DRAM makers are scrambling to increase capacity, but there are limitations. The biggest reason is that HBM, GPUs, and other advanced chips are all manufactured using EUV (extreme ultraviolet lithography) machines, and there is only one company in the world, ASML, that has this technology, so supply is limited. On top of that, HBM requires upward of 3 times the wafer capacity of regular DRAM. Micron has locked in long-term agreements, adding more predictability to its business.

Taiwan Semiconductor Manufacturing (TSMC) is at the heart of the AI infrastructure boom, as it has proven to be the only foundry capable of manufacturing advanced logic chips like GPUs with few defects at scale. As competitors have struggled with yields, TSMC has been an integral part of the semiconductor ecosystem and established a virtual monopoly on advanced chip manufacturing, giving it strong pricing power. TSMC is benefiting from the surge in demand for all kinds of logic chips, including GPUs, AI ASICs (application-specific integrated circuits), and CPUs. The company is also aggressively increasing its own capital expenditure to boost capacity to meet future demand. Despite its strong position and growth prospects, TSMC trades at a forward P/E of below 20 times 2027 analyst estimates.

The article frames three semiconductor and memory companies as undervalued participants in the AI infrastructure boom. Nvidia, despite being the most visible beneficiary of AI growth over the past five years, trades at a forward P/E of just 16× for fiscal 2028 — a multiple the author considers attractive for a company growing rapidly. The company's dominance in AI model training through its CUDA platform and its recent $20 billion（約3.2兆円） acquisition of Groq signal an effort to consolidate control across both training and inference workloads.

Micron presents a different but related story: its revenue more than quadrupled and gross margin jumped from 37.7% to 84.6%, yet it trades at a forward P/E just above 6× for fiscal 2027. The author attributes the low valuation to investor memory of memory market cyclicality, but argues the current cycle has structural support. High-bandwidth memory (HBM) demand is surging because it pairs with GPUs and other AI chips, yet supply is constrained by a bottleneck in EUV lithography — only ASML manufactures these machines — and HBM requires upward of 3 times the wafer capacity of regular DRAM.

TSMC rounds out the trio, trading below 20× forward P/E for 2027 despite holding a near-monopoly on advanced chip manufacturing. The article notes that competitors have struggled with yields while TSMC has maintained close partnerships and pricing power, and is now aggressively increasing capital spending to meet future demand. All three stocks benefit from complementary roles in the AI supply chain: Nvidia designs and sells the chips; Micron supplies the memory attached to them; TSMC manufactures them at scale.

No comments yet. Be the first to share your thoughts!