마이크론의 수익성 전망, 이제 경계해야 할 때
Micron Earnings Have Never Looked Better. Start Worrying.
AI 수요에 따른 수익성 전망과 반도체 시장 사이클적 리스크가 혼재되어 있어, 당분간은 중립적인 입장을 유지하는 것이 타당합니다.
핵심 요약
마이크론 주가 9배 상승해 시가총액 1.2조 달러 돌파; AI 인프라 7000억 달러 투자 전망.
핵심요약
- 마이크론 주가 12개월 간 9배 상승, 시가총액 1.2조 달러 돌파
- AI 인프라 투자 7000억 달러 예상, 고대역폭 메모리 수요 급증
- EPS 올해 61달러(7배 성장), 내년 118달러 전망
- 메모리 산업 사이클적 특성, 공급 과잉 가능성 경고
도입
마이크론의 급격한 성장과 높은 시장 평가에 투자자들은 주목해야 합니다. 반도체 메모리 시장의 특수성과 AI 인프라 수요의 변동성을 고려할 때, 현재 주가 상승이 지속 가능한지 검토할 필요가 있습니다. 특히, 메모리 산업의 역사적인 사이클적 특성을 분석하는 것은 중요한 포인트입니다.
본문 1: AI 인프라 수요가 만든 마이크론의 성장 동력
마이크론의 주가 상승은 AI 인프라에 대한 수요 증가가 주요 원인입니다. 마이크로소프트, 알파벳, 메타 등 주요 기업들은 올해 AI 인프라에 약 7000억 달러를 투자할 예정이며, 이는 마이크론의 고대역폭 메모리 수요를 급증시키고 있습니다. EPS는 올해 61달러로 7배 증가할 전망이며, 내년 118달러로 더 성장할 것으로 예상됩니다. 이는 마이크론의 수익성 개선이 AI 수요와 직접적으로 연결되어 있음을 보여줍니다.
본문 2: 메모리 산업의 사이클적 특성과 리스크
메모리 산업은 역사적으로 사이클적 변동성이 심한 특성을 보입니다. 2022-2023년에는 재고 과잉으로 인해 마이크론은 최대 23억 달러의 손실을 기록했고, 주가도 50% 급락했습니다. 2018-2019년에도 클라우드 운영자의 과다 주문이 재고 쌓임으로 이어져 DRAM 가격은 40%, NAND 가격은 60% 하락했습니다. 이러한 과거 사례들은 현재 마이크론의 수익성도 사이클적 정점일 수 있음을 시사합니다.
본문 3: 공급 과잉의 가능성과 향후 전망
현재 마이크론은 HBM(High Bandwidth Memory) 생산 능력을 2026년까지 완전 판매할 수 있는 상황입니다. 그러나 산업 전체적으로 새로운 생산 설비에 대한 투자 확대되고 있어, 향후 공급 과잉 가능성이 있습니다. 새로운 팹 건설에는 2~3년과 수십억 달러의 비용이 소요되며, 이는 시장 수요가 감소할 경우 가격과 수익률에 부정적인 영향을 미칠 수 있습니다. 따라서 투자자들은 마이크론의 현재 수익성 평가에 대해 신중한 접근이 필요합니다.
결론
마이크론의 현재 성장세는 AI 인프라 수요 증가에 기반하고 있지만, 메모리 산업의 사이클적 특성과 공급 과잉 가능성을 고려할 때 지속 가능한지 검토해야 합니다. 향후 마이크론의 수익성 평가와 주가 동향을 주시하는 것이 중요합니다. 특히, 신규 생산 설비의 진척 상황과 시장 수요의 변동성을 면밀히 관찰해야 합니다.
Original Article
Micron Earnings Have Never Looked Better. Start Worrying.
Micron Technology’s stock has risen by almost 9x over the past twelve months, taking its market capitalization above $1.2 trillion – among the largest single-year gains in the company’s four-decade trading history. The rally is driven by surging demand and a shortage of high-bandwidth memory that is used alongside AI accelerators in data centers. Hyperscalers, including Microsoft (MSFT) , Alphabet (GOOG) , and Meta (META) , are projected to spend roughly $700 billion on AI infrastructure this year, and memory is a critical input at every stage of that build-out.
EPS is on track to grow roughly 7x this year to $61, with consensus pointing to $118 next year. The stock trades at just 9.6x that forward figure. On the surface, that looks like a bargain. However, memory is one of the most cyclical segments of the semiconductor industry, and low P/E ratios can be deceptive. Earnings are often strongest when supply is tight and pricing is elevated. While Micron has sold out its HBM capacity through 2026, the industry is investing heavily in new capacity. If supply catches up with demand, pricing and margins could come under pressure, making today’s earnings closer to a cyclical peak than a sustainable run rate. See MU valuation metrics
How Memory Booms Turn Into Busts
The memory industry has produced multiple severe crashes sharing the same cause. Building a new fab takes two to three years and costs tens of billions. Once built, the economics favor running it at full capacity regardless of price. Demand surges, manufacturers commit to new capacity, and by the time it arrives, the market it was built for may have cooled considerably.
In 2022-2023, post-pandemic demand evaporated, inventories hit 31 weeks, and Micron posted its largest ever quarterly loss of $2.31 billion. The stock fell roughly 50%. In 2018-2019, cloud operators over-ordered, then cut purchases as inventories swelled. DRAM fell 40%, NAND 60%, and Micron dropped 57% from peak to trough. In 2014-2016, capacity expanded ahead of PC demand that never arrived. The stock fell 70%.
Why Investors Should Still Worry About Oversupply
None of this overrides the economics of semiconductor manufacturing. Micron has guided fiscal 2026 capex above $25 billion, while SK Hynix is expected to spend roughly $27 billion. Samsung is also expanding aggressively. Collectively, the industry’s largest players are committing more than $75 billion annually to new capacity. Given the typical two- to three-year lag between investment and production, 2027-2028 is shaping up as the period when oversupply risk becomes most acute.
The scale of equipment spending reinforces that concern. Wafer fab equipment spending is projected to reach roughly $145 billion in 2026 and $156 billion in 2027. These tools are long-lead-time investments that cannot be easily canceled once ordered. In other words, the industry is not just planning for more capacity; it is already locking it in. That raises the odds that supply growth will remain robust through 2027 and 2028, even if demand growth slows.
That would be less concerning if demand were guaranteed. However, the current upcycle still depends on sustained AI investment by hyperscalers such as Alphabet (GOOG) and Amazon (AMZN) who are together expected to spend over $600 billion in capex this year. If hyperscalers face pressure to deliver stronger returns on their massive capex investments, spending could slow, feeding directly back into memory demand.
The bullish case for Micron rests on the idea that HBM is not behaving like traditional memory markets.
HBM demand is growing geometrically. Nvidia (NVDA) used 80GB of HBM. The Rubin GPU, shipping in the second half of 2026, carries 288GB. Rubin Ultra, due in 2027, targets 1TB. That is a 12x increase in HBM per GPU in roughly three years. The shift from AI training to inference is also expanding the addressable market, as applications such as AI agents and real-time video generation require large amounts of high-speed memory.
HBM is also increasingly sold through multi-year agreements rather than on the spot market, reducing the abrupt order cancellations that have historically amplified downturns. In March, Micron signed the industry’s first five-year HBM supply agreement covering both volume and pricing. The 2026 contracted volumes are well documented. The 2027 picture is far less defined, and that matters because 2027 is precisely when new capacity from all three major suppliers comes online.
These differences suggest the current cycle could prove more durable than previous memory booms.
However, the key question is whether those advantages are sufficient to absorb the massive wave of capacity now being built. At under 10x next year’s earnings, Micron may look inexpensive. But if those earnings represent peak-cycle profits rather than a new baseline, the stock may be far less of a bargain than the headline multiple suggests.
Opportunities like Micron highlight how individual semiconductor stocks can surge dramatically during technology transitions, but they also carry concentrated exposure to industry cycles, capacity expansion, and execution risk. A disciplined portfolio approach helps smooth these risks while still participating in long-term growth themes. The Trefis High Quality (HQ) Portfolio has consistently outperformed its market benchmark since inception, delivering cumulative returns of over 105%.