Micron vs SK Hynix: HBM4 race for Nvidia Rubin, September 2026

Micron closed at $1,096.16, up 5% on September 22, as the company pursues HBM4 qualification for Nvidia's Rubin platform — where SK Hynix is already shipping — backed by $15.86B in FY2025 capital expenditure, the highest intensity among chip peers.

Micron's 5% gain on September 22 to close at $1,096.16 has no single publicly reported catalyst, but it arrives in a period when the structural case for AI memory exposure has rarely looked more concrete. High-bandwidth memory is reported to have sold out through 2027, with customers facing severe shortages. TrendForce data cited in coverage shows Micron's conventional DRAM revenue grew 65.5% quarter-on-quarter in Q2 2026 — the fastest clip among major memory suppliers — as AI workloads drove a broad-based server infrastructure upgrade. Dell's strong quarterly results, flagged in separate coverage as confirming demand well beyond HBM, underscore that the server memory story is not a one-product narrative.

Slicast's compilation of SEC XBRL data shows the company's capital expenditure reached $15.86B in FY2025, up 89.1% from $8.39B in FY2024 and equal to 42% of $37.38B in revenue — ranking first among 15 chip peers on capital intensity and first among 14 on absolute outlay. Reports from September indicate the company is targeting a doubling of HBM output to roughly 100,000 wafers per month, with HBM4 expected to take a sharply larger share of that mix. The logic is clear: Micron is deploying capital aggressively to compress its technology gap with SK Hynix and Samsung before the current HBM pricing cycle turns.

That gap is real and consequential. According to coverage, SK Hynix is already shipping 16-layer HBM4 for Nvidia's Rubin GPU platform, while Micron and Samsung are still in an evaluation phase with a Q4 2026 decision point and no confirmed orders. The management reshuffle disclosed in a Form 8-K filed August 26, 2026 — in which Manish Bhatia was named President and Chief Operating Officer — reflects that urgency; reporting also cites a concurrent CTO appointment, both executives tasked with resolving persistent HBM production bottlenecks. On the broader server memory front, Micron's September announcement of 512GB DDR5-9200 modules with a 16W power draw and a claimed 60% energy advantage over competing configurations points to a parallel effort to win the RDIMM market where its competitive exposure is more limited.

The risks are structural and compounding. China's CXMT has reached mass production in DRAM; a Nikkei Asia report claims CXMT's profit margin now surpasses both SK Hynix and Micron — a single-source assertion that warrants caution, but one directionally consistent with the broader pattern of rapidly ascending Chinese capacity documented across multiple items. In Taiwan, Micron settled a labor dispute with a package that reportedly included up to 68 months of severance pay, removing near-term supply disruption risk while adding to cost pressure. The algorithmic side carries its own uncertainty: reports note that DeepSeek's latest model reduced KV-cache HBM requirements by 75% and SSD needs by 87.5%, a demonstration that software efficiency can deflate hardware demand curves. And according to Wccftech, Nvidia's NVHBM architecture — which retains control of the base die, reported to cost three to four times the core die — structurally captures most of the HBM value chain, progressively narrowing the margin space available to memory suppliers.

Three concrete signals will determine whether Micron converts its current capital cycle into durable earnings power. The first is the outcome of HBM4 qualification for Nvidia Rubin in Q4 2026: success would close the shipping gap with SK Hynix and anchor the return on the $15.86B FY2025 investment; a further delay would deepen reliance on conventional DRAM revenues. The second is the pace and ceiling of CXMT's ramp: if Chinese capacity reaches a scale that compresses conventional DRAM pricing, the 65.5% Q2 2026 sequential growth figure becomes difficult to sustain. The third is the trajectory of HBM contract pricing into 2027; forecasts cited in coverage project a 140% rise, but the combination of industry-wide capacity expansion and algorithmic efficiency gains introduces real uncertainty into that number. Micron enters this period with a stronger technology portfolio and heavier capital commitment than it has brought to any prior industry upcycle. Whether that is enough depends on how memory-intensive the AI buildout remains — and on execution milestones the company has yet to hit.