---
title: "Micron's $50 Billion Quarter Meets an AI Stack That Moves Its Cache"
description: "Micron reports Wednesday on a $50 billion revenue guide at an 86% gross margin, and even the Overweight camp expects smaller estimate hikes. New serving measurements show frontier models relocating cache from GPU HBM to host DRAM instead of shrinking it, a shift this week's earnings previews do not touch."
publisher: "The Inference"
section: "Chips"
published: 2026-09-29T22:51:05.746Z
modified: 2026-09-29T22:51:05.746Z
canonical: https://theinference.org/article/micron-s-50-billion-quarter-meets-an-ai-stack-that-moves-its-cache
language: en
keywords: "Micron, Semiconductors, Memory, DRAM, HBM, AI Inference"
---

# Micron's $50 Billion Quarter Meets an AI Stack That Moves Its Cache

> Micron reports Wednesday on a $50 billion revenue guide at an 86% gross margin, and even the Overweight camp expects smaller estimate hikes. New serving measurements show frontier models relocating cache from GPU HBM to host DRAM instead of shrinking it, a shift this week's earnings previews do not touch.

[Micron](https://theinference.org/markets/companies/micron) reports fiscal Q4 after the close on Wednesday, 30 September 2026, [guided to $50 billion of revenue](https://finance.yahoo.com/markets/stocks/articles/micron-mu-lands-100-billion-050641920.html) plus or minus $1 billion at an 86% gross margin [1][2][3]. Last quarter printed $41.46 billion at an [84.6% gross margin](https://theinference.org/article/gm-pays-interest-to-wait-in-line-for-chips-while-micron-books-an-84-6-margin) [2][7]. [The stock has more than tripled this year](https://theinference.org/article/burry-adds-to-micron-and-nebius-shorts-backing-acer-s-ceo-against-the-memory-kings), up over 230% [1], a run priced for another round of big estimate hikes.

[Consensus sits above the guide at $51.24 billion](https://finance.yahoo.com/markets/stocks/articles/micron-earnings-preview-analysts-see-022234081.html), up 353%, the fastest on record [4]. [Morgan Stanley](https://theinference.org/markets/companies/morgan-stanley) stays Overweight at a $1,200 target and still expects estimates to rise, just "[less so than prior quarters](https://finance.yahoo.com/markets/stocks/articles/micron-post-earnings-estimate-hikes-162711418.html)" [1][5]. [Fiscal 2027 profit estimates jumped](https://www.fool.com/investing/2026/08/09/micron-has-surged-207-this-year-brace-for-a-steep/) from $95.80 to $154.70 in three months, then rose just 1.2% in the month through early August [6]. For holders, the size of the next hike is the trade.

*Chart: **From $4 billion to a guided $50 billion in three years** Line chart of Micron quarterly revenue from fiscal Q4 2023 through fiscal Q3 2026, rising from $4.0 billion to $41.5 billion, with the fiscal Q4 2026 guide of about $50 billion marked as guidance*

*Micron quarterly revenue by fiscal quarter, USD billions, from SEC filings via Sharadar; the final point is company guidance for the quarter reported 30 September 2026. [7][1][6]*

## Frontier models push cache into host memory

None of that math prices what the newest frontier serving stack does to memory. Z.ai's GLM-5.x models, 744 billion parameters with 40 billion active per token, use sparse attention to read less of the cache at each step [8]. The catch: picking which tokens to read requires the full context to sit in HBM first, so the capacity bottleneck survives [8].

SGLang's HiSparse layer offloads cache entries to host DRAM, and Z.ai says GLM-5.3 adds local storage as a further tier [8]. [Measured on a B200 serving GLM-5.2](https://newsletter.semianalysis.com/p/sparse-savings-persistent-demand-inside-glm53), doubling concurrency from 8 to 16 requests cut GPU-memory token reuse from 90.3% to 54.8% while host-memory reuse rose from 6.0% to 40.3%, with the overall hit rate above 95% [8]. Agents, which recycle long histories and write little, are the workload doing this [8].

Sparse attention does not shrink the memory bill. It forwards the bill to a cheaper tier.

*Chart: **Doubling concurrency moved B200 cache reuse from GPU memory to host DRAM** Column chart showing GPU memory reuse falling from 90.3% to 54.8% and host DRAM reuse rising from 6.0% to 40.3% while overall cache hit rate stays above 95% as serving concurrency doubles*

*Share of prompt tokens reused from each memory tier while serving GLM-5.2 on a B200 system at 8 versus 16 concurrent requests, with overall hit rate above 95%. Source: SemiAnalysis / InferenceX measurements, 28 September 2026. [8]*

## Conventional DRAM delivers the volume while HBM keeps the margin

Micron sells both ends of that ladder, and its growth already lives at the bottom. Its [June-quarter DRAM revenue rose 65.5%](https://www.msn.com/en-us/money/general/micron-posted-fastest-dram-revenue-growth-in-q2-2026-sk-hynix-s-hbm-lead-was-its-liability/ar-AA2bKJJe) from the prior quarter to $36.0 billion, the fastest of the big three, while its share of the HBM market slid from 21% to 18% [9].

The premium stays with HBM. [SK Hynix](https://theinference.org/markets/companies/sk-hynix) holds about half that market and runs a 76% operating margin, and each HBM4 wafer eats roughly three DDR5 wafers of capacity, part of why conventional DRAM prices jumped 58% to 63% in a quarter [9].

![SK Hynix DDR5 server memory modules on display](https://storage.googleapis.com/theinference-public/img/203675f0bb1e0268.jpg)

*Server DDR5 memory modules: conventional DRAM pricing jumped over 50% in a single quarter as capacity was diverted toward HBM. (4300streetcar)*

Micron's own investor site [pitches agentic AI under a white paper](https://investors.micron.com/overview/default.aspx) titled "Agentic AI in Data Centers: The Critical Role of High-Bandwidth Memory" [10]; the serving measurements put agentic traffic one tier down.

## Long-term purchase pacts cannot resolve where future margins settle

The bull case has receipts: [16 strategic customer agreements worth about $100 billion](https://www.forbes.com/sites/jimosman/2026/09/28/micron-stock-earnings-may-show-why-100-billion-wont-save-the-rally/) through 2030, with roughly $22 billion of customer cash and other commitments [2][3]. It also has arithmetic: $100 billion is about two quarters of revenue at the guided pace [2]. Synovus sees supply [tight through fiscal 2027](https://theinference.org/article/cxmt-starts-mass-producing-cheaper-dram-as-rivals-bet-the-shortage-lasts-to-2030) [1], and SK Hynix is [weighing a Solidigm IPO](https://www.reuters.com/world/sk-hynixs-solidigm-weighs-ipo-that-could-value-the-unit-up-150-billion-sources-2026-09-25/) that values its NAND unit at up to $150 billion [11].

Even the Overweight camp concedes there is "no real way to prove that near term" [1]. The beat will take care of itself; at an 86% margin, scarcity is not the question. Which memory the agentic era pays a premium for is. The test on Wednesday's call: whether Micron's HBM share keeps sliding while conventional DRAM does the growing, and whether management's DRAM pricing talk credits scarcity or substitution [9].

## Takeaway

While Wall Street expects Micron to report another blowout quarter guided at $50 billion, frontier serving architectures are shifting cache offloading from premium HBM into host DRAM, moving the center of memory consumption down the margin stack.

## Sources

1. [StockTwits via Yahoo Finance, "Micron's Post-Earnings Estimate Hikes May Lag Previous Quarters, Flags Morgan Stanley," 28 September 2026](https://finance.yahoo.com/markets/stocks/articles/micron-post-earnings-estimate-hikes-162711418.html)
2. [Forbes (Jim Osman), "Micron Stock: Earnings May Show Why $100 Billion Won't Save The Rally," 28 September 2026](https://www.forbes.com/sites/jimosman/2026/09/28/micron-stock-earnings-may-show-why-100-billion-wont-save-the-rally/)
3. [Simply Wall St via Yahoo Finance, "Micron (MU) Lands $100 Billion In Customer Deals After Blowout Q3," 27 June 2026](https://finance.yahoo.com/markets/stocks/articles/micron-mu-lands-100-billion-050641920.html)
4. [StockTwits via Yahoo Finance, "Micron Earnings Preview: Analysts See Over 350% Revenue Jump As Retail Bulls Bet On Fresh Buyback," 27 September 2026](https://finance.yahoo.com/markets/stocks/articles/micron-earnings-preview-analysts-see-022234081.html)
5. [MT Newswires via Yahoo Finance, "Micron's 'Conditions' Likely Remain Strong, Morgan Stanley Says," 28 September 2026](https://finance.yahoo.com/markets/stocks/articles/micron-apos-apos-conditions-apos-172222362.html)
6. [The Motley Fool (Manali Pradhan), "Micron Has Surged 207% This Year. Brace for a Steep Pullback.," 9 August 2026](https://www.fool.com/investing/2026/08/09/micron-has-surged-207-this-year-brace-for-a-steep/)
7. [Sharadar quarterly fundamentals for Micron, from SEC filings, retrieved 29 September 2026](https://www.sec.gov/cgi-bin/browse-edgar?action=getcompany&CIK=0000723125)
8. [SemiAnalysis, "How GLM5.3 Sparse Attention Affects HBM Memory Usage," 28 September 2026](https://newsletter.semianalysis.com/p/sparse-savings-persistent-demand-inside-glm53)
9. [Tech Times (Oscar Lazenby), "Micron posted fastest DRAM revenue growth in Q2 2026; SK Hynix's HBM lead was its liability," 7 September 2026](https://www.msn.com/en-us/money/general/micron-posted-fastest-dram-revenue-growth-in-q2-2026-sk-hynix-s-hbm-lead-was-its-liability/ar-AA2bKJJe)
10. [Micron Technology Investor Relations, company overview and white papers, accessed 29 September 2026](https://investors.micron.com/overview/default.aspx)
11. [Reuters (Milana Vinn and Echo Wang), "SK Hynix's Solidigm weighs IPO that could value the unit at up to $150 billion, sources say," 25 September 2026](https://www.reuters.com/world/sk-hynixs-solidigm-weighs-ipo-that-could-value-the-unit-up-150-billion-sources-2026-09-25/)


---

Micron's $50 Billion Quarter Meets an AI Stack That Moves Its Cache — The Inference. Canonical: https://theinference.org/article/micron-s-50-billion-quarter-meets-an-ai-stack-that-moves-its-cache