Micron faces chip shortage through 2028 as memory supercycle deepens

Memory supply will remain constrained through at least 2028, extending a pricing cycle that has already lifted Micron's revenue nearly 5x in one year.

Micron Technology's memory chips will remain in short supply through at least 2028, CEO Sanjay Mehrotra told analysts last month, as AI infrastructure demand consumes fabrication capacity faster than the industry can build new plants.

"Demand continues to significantly exceed industry supply, and we expect tight conditions to persist beyond calendar 2027," Mehrotra said on the company's June 24 earnings call.

The results backing that forecast are extraordinary by any historical measure. Micron posted $41.5 billion in fiscal third-quarter revenue, up roughly 4.5 times from a year earlier and beating the $35 billion consensus by 18.6%. Earnings per share of $25.11 surpassed the $20.20 estimate, while net income reached $28.2 billion. Operating margins expanded to 68% from 20% in the year-ago period. The company guided for approximately $50 billion in current-quarter revenue, up from $11.3 billion in the same quarter last year.

The supply crunch traces to a structural mismatch. Three companies — Micron, Samsung Electronics and SK Hynix — control roughly 90% to 95% of global DRAM production, and all three have redirected fabrication capacity toward high-bandwidth memory (HBM), the 3D-stacked chips that Nvidia's AI accelerators require. Manufacturing one gigabyte of HBM consumes about three times the wafer capacity of standard DDR5, meaning every wafer committed to AI memory is a wafer taken from the conventional DRAM that goes into PCs, laptops and enterprise servers.

DRAM contract prices surged 80% to 95% quarter over quarter in early 2026, according to TrendForce, with further increases of 40% to 50% expected in the third quarter, per Jefferies. IDC projects 2026 DRAM supply growth of only about 16% year over year, well below the historical 20% to 30% range, because new fabrication plants require two to three years to reach meaningful output.

The capacity pipeline is visible but distant. Micron is building a $250 billion fabrication complex in Clay, New York, targeting operations around 2028. SK Hynix raised $26.5 billion in a Nasdaq listing on July 10 — the largest-ever foreign IPO on a US exchange — to fund its Yongin Semiconductor Cluster in South Korea and a $4 billion advanced-packaging plant in Indiana. Samsung is expanding at Pyeongtaek P5. All three projects are expected online between 2027 and 2028, which most analysts expect will bring the first meaningful supply relief.

HBM Is the Bottleneck That Cannot Be Bypassed

High-bandwidth memory is a 3D-stacked DRAM architecture in which up to 16 memory dies are stacked vertically and connected through thousands of microscopic copper pillars called through-silicon vias (TSVs). The stack sits directly adjacent to the GPU on a shared silicon interposer, creating a data bus roughly 32 times wider than conventional DDR5. HBM3E, the current generation used in Nvidia's Blackwell GPUs, delivers 1.2 to 1.33 terabytes per second of bandwidth per stack. HBM4, which SK Hynix began mass-producing in February, doubles the I/O width to 2,048 bits and exceeds 2 terabytes per second per stack.

SK Hynix controls roughly 56.4% of global HBM revenue and is Nvidia's primary supplier. UBS projects it will capture about 70% of HBM4 orders for Nvidia's next-generation Rubin platform. Micron has signed a multi-year deal to become a primary memory supplier for Anthropic's AI infrastructure, diversifying its customer base beyond the Nvidia-centric HBM supply chain.

The Consumer Cost of the AI Memory Shift

The same dynamics that have propelled Micron's stock — up more than 800% over the past year to a $1.23 trillion market cap — carry a less visible cost. DRAM prices have risen roughly 700% over four years, according to a federal antitrust class action filed June 25 in the Northern District of California. The suit, Garciaguirre v. Samsung Electronics, alleges the three DRAM manufacturers coordinated supply restrictions beginning around October 2022. The allegations are unproven; all three companies have said they intend to defend the case.

For investors, the question is whether this cycle differs from the boom-and-bust pattern that has defined the memory industry for three decades. Micron trades at 23.4 times trailing earnings, a multiple that looks reasonable against 450% revenue growth but leaves little room for error if AI infrastructure spending slows. The Roundhill Memory ETF (DRAM), which holds Micron as its top weighting at 27.6%, is up 176% year to date.

Mehrotra's forecast that tight supply extends beyond 2027 suggests pricing power will persist through at least two more earnings cycles. The risk is that all three manufacturers' new fabs come online around the same time — 2027 to 2028 — recreating the oversupply that has punished memory stocks after every previous boom. For now, the data supports the bull case: every wafer is spoken for, and the next wave of capacity is still two years away.

This article is for informational purposes only and does not constitute investment advice.