Friday, September 11, 2026
AI 인프라 · 뉴스 & 분석
반도체·하드웨어리포트
반도체·하드웨어 · 리포트

마이크론은 차세대 HBM4가 빠르게 성장하는 시장 점유율을 차지할 것으로 예상됨에 따라 월간 HBM 생산 능력을 약 10만 웨이퍼로 두 배 확대할 계획이다.

HBM 생산 확대는 차세대 AI 가속기 처리량을 제약하는 핵심 병목 현상을 직접 해소하여 더 높은 밀도의 랙 배포를 가능하게 하고, SK Hynix와 같은 경쟁사의 가격 압박을 안정화한다.
업계 전문지Slicast · September 7, 2026 · 미국 · 출처: igor´sLAB
중요도 85

Micron is now shipping HBM4 in volume, generating billions in revenue, and plans to significantly expand production. According to South Korea’s Electronic Times, citing industry sources, the U.S. memory manufacturer aims to increase its total HBM capacity to approximately 100,000 wafer starts per month by the end of 2026. Last year, output stood at roughly 40,000 to 50,000 wafers monthly, meaning this target would effectively double capacity in a short timeframe. While the 100,000-wafer figure remains unconfirmed by Micron as an official production target, the company has officially verified a rapid HBM4 ramp-up, ongoing volume shipments, and cumulative HBM4 revenue exceeding $1 billion. Given these confirmed metrics, the reported capacity expansion is highly plausible, though the exact figures remain based on industry reporting.

The Electronic Times report indicates that Micron intends to add capacity for up to 60,000 additional HBM wafer starts per month by year-end, with equipment suppliers reportedly receiving substantial orders to support the initiative. Primary manufacturing and packaging will be concentrated in Taiwan and Singapore, with preparatory investments also underway in Hiroshima, Japan. This aggressive scaling underscores Micron’s efforts to close the gap with rivals SK hynix and Samsung, which each reportedly operate around 150,000 to 200,000 HBM wafer starts monthly. Even at 100,000 wafers, Micron would not surpass its competitors, but the competitive divide would narrow considerably.

The shifting product mix warrants particular attention. Industry reports suggest that HBM4 12-Hi accounted for only 20% to 30% of Micron’s HBM output at the start of 2026, with that share projected to reach up to 50% by year-end. Though unverified officially, Micron’s aggressive HBM4 push is well-documented. The company is already in high-volume production of the 36-GB HBM4 stack, which features twelve DRAM layers, a 2,048-bit-wide interface, per-pin speeds exceeding 11 Gbit/s, and aggregate bandwidth surpassing 2.8 TB/s per stack. Compared to HBM3E, Micron states HBM4 delivers 2.3 times the bandwidth and over 20% greater energy efficiency. Additionally, the company has shipped 48-GB 16-Hi HBM4 samples to customers, boosting per-stack capacity by one-third relative to the 12-Hi variant. In June, Micron noted that the HBM4 12-Hi ramp-up is progressing roughly twice as fast as the previous HBM3E 12-Hi cycle, coinciding with the aforementioned $1 billion+ in revenue.

Micron’s HBM4 architecture was explicitly engineered for NVIDIA’s Vera-Rubin platform. Transitioning from HBM3E to HBM4 intensifies demand for manufacturing capacity, as HBM requires substantially more wafer area per bit than conventional DRAM and involves complex stacking and advanced packaging processes. Consequently, 100,000 HBM wafer starts do not equate to the same memory yield as 100,000 standard DDR5 wafers. Scaling HBM production disproportionately consumes resources from the broader DRAM ecosystem, a dynamic that is currently exacerbating global memory shortages. Despite these constraints, the investment is strategically sound. AI memory has emerged as one of Micron’s highest-margin segments, with demand for next-generation accelerator memory consistently outpacing supply. Citing Counterpoint Research, the Electronic Times notes that Micron held an estimated 18% global HBM market share in Q2 2026, trailing SK hynix (~50%) and Samsung (~32%), but firmly establishing itself beyond a marginal player.

Expanding to roughly 100,000 wafer starts monthly would noticeably strengthen Micron’s market position. However, capacity alone will not dictate success. Final outcomes will depend on fabrication yields, packaging efficiency, qualification cycles with NVIDIA and other key clients, and the pace of transition to HBM4E. These factors will ultimately determine how much saleable memory materializes from the new infrastructure. The report aligns closely with Micron’s official disclosures regarding its HBM4 trajectory. Doubling capacity to ~100,000 wafers monthly is ambitious, yet entirely credible given the $1 billion+ in HBM4 revenue and accelerated production timelines. Nevertheless, the 100,000-wafer target and potential 50% HBM4 product share should still be treated as industry estimates rather than confirmed corporate guidance. What remains certain is that Micron is producing HBM4 at scale and ramping faster than with HBM3E. The actual production volume by year-end remains a critical metric to watch.

원문 보기