Sunday, October 11, 2026
AI Infrastructure · News & Analysis
Home › Chips & Hardware › Report
Chips & Hardware · Report

Micron's HBM production line in Taichung has fired up, with AI demand seen as steadying the stock.

Adding high-bandwidth memory capacity in Taichung expands HBM supply for AI accelerators, easing a memory bottleneck that constrains GPU and custom-chip shipments.
Trade pressSlicast · October 10, 2026 at 16:31 UTC · US · Source: AD HOC NEWS
importance 60

The most consequential bottleneck in artificial intelligence isn't written in code. It's etched into silicon inside climate-controlled cleanrooms. While investors fixate on software models and valuation, the physical supply of high-bandwidth memory (HBM) determines how fast the AI buildout can proceed. Micron Technology is betting billions that this hardware layer, not the algorithms, will set the pace.

That bet is taking concrete shape in Taiwan. According to media reports, Micron Taiwan chairman Donghui Lu put the company's cumulative investment on the island at NT$1.6 trillion through June, roughly $50.2 billion. The second advanced packaging fab in Taichung is scheduled to begin HBM production in mid-October, a milestone that directly addresses the supply constraint at the heart of modern AI server architecture.

Additional capacity will not arrive quickly. New lines in Taoyuan and Tongluo are not expected to be operational until 2027 and 2028, leaving the market structurally tight in the meantime. Micron itself expects supply-side scarcity to intensify across fiscal 2027 and 2028.

**A Record Year and a Bullish Guide**

The numbers behind that expansion are striking. Micron closed fiscal 2026 with record revenue of $133.19 billion. In the fourth quarter alone, the company reported record sales of $54.23 billion and GAAP net income of $37.70 billion.

Management's outlook for the first quarter of fiscal 2027 points to revenue of $61.5 billion, with a variance of $1.5 billion, a range that brackets the $60 billion to $63 billion guidance cited elsewhere. CEO Sanjay Mehrotra has said memory market conditions in calendar 2027 and 2028 should be noticeably tighter than in the current year, with no rebalancing point yet in sight. According to media reports, the company has already raised HBM contract prices for 2027.

Sentiment around the stock has been volatile, driven less by Micron's own results than by conflicting signals about AI spending. On Thursday, a report that OpenAI's annualized revenue was running at only about $50 billion rattled the sector, and Micron shares fell 4.8% that day as traders worried about a pullback in high-performance memory orders. The following day brought a counterweight: Bloomberg reported that OpenAI expects annualized revenue of $70 billion or more by year-end, easing fears of an abrupt data-center spending freeze.

Micron stock ended Friday's European session at €919.60, down a modest 0.4%. Since the start of the year, the shares have climbed 265%, though they remain roughly 17% below their 52-week high.

**Labor Friction and Legal Clarity**

Rapid expansion brings its own complications. At the Taoyuan site, union members voted 1,994 to 2,012 in favor of authorizing a strike. No date for any walkout has been set, leaving the practical impact on output uncertain.

On the legal front, Micron resolved a longstanding patent dispute with Netlist through a settlement. The agreement covers a five-year license running from the fourth quarter of fiscal 2026 through the third quarter of fiscal 2031, at $30 million per quarter, or $600 million in total.

Shareholders also received a signal of confidence in the form of capital allocation. The board authorized expanding the share buyback program to as much as $35.16 billion, with repurchases permitted from December 9, 2026, subject to conditions tied to US subsidies.

The picture that emerges is of a company with strong margins, a sold-out product line, and a multi-year growth runway, whose fortunes remain tightly bound to the investment cycles of leading AI developers. The infrastructure for artificial intelligence is built over years, not quarters. In this cycle, whoever controls the memory controls the tempo.

Read the original
Micron's HBM production line in Taichung has… · Slicast