The week ahead offers a rare pause. With earnings and conferences thin, it’s a good time to step off the calendar and look directly at companies. A week like this reminds us that research matters most when it isn’t dictated by an earnings date. The noise is low; the signal stands out.

I think you’ll dig our Micron note below — it was a mountain of research, but paints a picture of technological and deep cultural change at the Boise giant.

Visit the archive of past notes and our searchable research database at epistrophy.beehiiv.com.

As always, I’m focused on three things:
1) Technology-driven change;
2) the latest in innovation and startup trends, and;
3) stock fraud.

Companies Discussed

In This Note:

• Companies Discussed

• Not Your Father’s Micron

  • Micron FY2025 Q4 Earnings & A Historic Study of Mi …

    • Micron’s 22 Years Of Hiring Shifts: A Study

      • 2013 — Fab-Centric Micron

      • 2022 — Straddling Old and New

      • 2025 — The Split Personality Micron

• Tweet O’ The Week

• Epistrophy In The News

• 📆 of Epistrophy Events

• Availability This Week

Not Your Father’s Micron

Micron FY2025 Q4 Earnings &
A Historic Study of Micron Job Postings

Micron’s (MU:NYSE) Q4 results showed us the fundamental change underway at the company as it goes through its biggest metamorphosis since founder J.R. Simplot was driving around potato fields in his cream collared Lincoln.

It’s all about the embrace of AI. You can see it in the numbers reported on Sept. 23, and you can see it in our Epistrophy Capital Research study of twenty years of Micron job postings.

Fiscal Q4 2025 results showed exceptional $11.2 billion in sales up 45% year-over-year. Gross margins, which were negative two years ago, hit 44.7%. The company generated $17.5 billion in operating cash flow. Its Cloud Memory unit –  40% of sales –  had $4.5 billion in very profitable quarterly revenues – bigger than the entire company's revenues two years ago in the same quarter. 

Memory has always been cyclical. This cycle is different, and the financials make that plain.

High Bandwidth Memory was the star of the quarter, driving record data center revenues and pulling Micron into conversations once reserved for Nvidia (NVDA:NASDAQ)and AMD (AMD:NYSE). HBM is not just stacked DRAM; it’s a vertically integrated architecture where eight or twelve DRAM dies are connected by through-silicon vias (TSVs) and mounted on a silicon interposer that sits next to a GPU or accelerator die. The result is bandwidth measured in terabytes per second with far lower power per bit than conventional DIMMs.

Micron’s product line makes that clear. The company has moved from its HBM2E offerings to HBM3, with HBM3E samples now shipping, delivering speeds of up to 9.2 Gbps per pin. Each stack can push more than 1.2 TB/s of bandwidth at capacities up to 24 GB per stack, a configuration designed explicitly for Nvidia’s Blackwell GPUs and AMD’s MI300 accelerators. The company has emphasized that HBM is being built not just in Boise fabs but with advanced packaging and validation hubs in Texas and California.

The contrast with Micron’s legacy businesses is sharp. Commodity NAND flash remains oversupplied, pricing continues to fall, and Micron itself has slowed NAND R&D investment, focusing instead on cost discipline. Even DDR5 DRAM, while technically advanced, is destined for servers and PCs where margins are thinner and performance needs are lower. HBM is different: it commands a premium dollar-per-bit, is tightly coupled to GPU roadmaps, and cannot be substituted by cheaper memory without crippling performance.

HBM is taking off now because AI workloads — transformers, LLMs, recommendation engines — saturate conventional DDR5 and even GDDR6. Training a frontier-scale model can require tens of terabytes per second of memory bandwidth across hundreds of GPUs. That makes HBM not a niche but a potential bottleneck. And Micron, after years of lagging Samsung and SK Hynix in graphics and specialty DRAM, is finally positioned to compete in the highest-value corner of memory. “We definitely continue to see strong long-term growth and are very excited about all these various announcements of massive data center infrastructure spend,” said CEO Sanjay Mehrota. “We have talked about trillions of dollars of spend over the next several years. And of course, memory is very much at the heart of this AI revolution. This means a tremendous opportunity for memory and certainly a tremendous opportunity for HBM.”

Scale alone is no longer enough. Micron has built fabs before, but AI requires more than capacity. Chip design, advanced packaging and co-optimization with customers are the new frontiers. Packaging is no longer an afterthought — it’s where value is captured. Software workloads — PyTorch, transformer inference, large-model training — now shape how memory is designed. Partnerships with hyperscalers and GPU vendors are table stakes. Micron cannot remain a commodity DRAM maker if it wants to stay relevant.

Micron’s 22 Years Of Hiring Shifts: A Study

We examined 22 years of Micron hiring data, and above all it paints a picture of the sudden technical and cultural  evolution led by Mehrota. The shift in Micron’s business shows up not just in press releases and investor calls but in who Micron is trying to hire and where those jobs are based. 

Our methodology was straightforward: scrape, archive, and code job postings from Micron’s careers pages, job boards, and the Wayback Machine. We sorted postings by geography and role type — fab/process, packaging/HBM, software-aware systems performance, and customer-facing business development. We then tracked the mix by quarter from 2020 through 2025 and cross-checked against historical postings from 2013 and 2014.

We’ve identified three distinct periods for Micron, the fab-centric yield obsessed RAM maker, the cross over period and then the new Micron that has emerged with high-bandwidth memory.

2013 — Fab-Centric Micron

Micron’s careers page in 2013 and 2014 was unmistakably the world of a DRAM foundry. The listings were dominated by roles like CDSEM Metrology Process Development Engineer in Boise, Process Control Systems Engineer in Boise and Intern NAND Memory Cell Engineer in Boise — all focused on wafer-level precision, yield optimization, and process reliability. Even the entry-level jobs, such as Intern-Shift Reticle Engineer in Boise, Idaho, pointed to the same thing: a culture centered on the tools and techniques of the fab. The one outlier — a System Performance Modeling Engineer in Folsom, California — was a sliver of what was to come, but at the time it was buried among postings about metrology and reticle shifts. Customers, workloads, and software were nowhere to be found. Micron’s engineers lived inside the fab, competing on cost per bit and incremental yield gains.

2022 — Straddling Old and New

Micron’s careers page in 2022 showed a company caught between eras. Boise still dominated, with postings for Water Engineers, Site Engineers, and DRAM Reliability Test Engineers — roles tied to maintaining fabs, facilities, and process stability. These were classic Micron jobs, extensions of the fab-centric culture that defined the company for decades. Yet mixed in were signals of a new direction. In Taichung City, Taiwan, a posting for Compute Validation Team Leader reflected a different mandate: validating memory performance not just in isolation, but in the context of compute workloads. Another Taichung-based role, Technical Assistant, SMART MFG & AI, Data Engineering, hinted at AI-driven process control and data-centric manufacturing. The language was shifting from wafers and reticles to compute and AI. Micron was still largely old-school in 2022, but the seeds of its transformation were visible in the kinds of engineers it had just begun to hire.

2025 — The Split Personality Micron

By 2025 the split in Micron’s hiring is unmistakable. Boise still anchors the company with roles like Facilities Electrical Systems Senior Engineer, the kind of job that has kept fabs humming for decades — overseeing voltage distribution, ensuring OSHA compliance, and troubleshooting equipment failures. But alongside these are postings Micron would hardly have imagined ten years earlier. In Folsom, California, a Principal Systems Design Engineer, HBM is tasked with co-designing memory interfaces for the highest-bandwidth AI accelerators. In Richardson, Texas, a Systems Design Engineer points to system-level optimization work that places Micron engineers directly alongside GPU vendors and hyperscalers. And in Longmont, Colorado, an SSD Customer Validation Engineer is dedicated to testing Micron products in the real-world environments of its largest customers. These jobs speak a new language: not wafers and yield, but bandwidth, validation, and customer workloads. In 2025, Micron is still a fab operator in Boise — but it is also something else entirely: a systems partner embedded in the AI ecosystem.

This juxtaposition makes the transformation visible. Boise is still there, with its legacy roles. But the strategic weight is shifting toward new jobs in new geographies, jobs that exist only because Micron is trying to become a system-level partner in AI.

Micron’s U.S. hiring outside Idaho is small in absolute numbers, but it’s where the cultural shift shows up. Roles in Silicon Valley, Austin/Richardson, and Denver/Longmont track directly to advanced packaging, HBM, and system-performance engineering. Just as important, these hubs place Micron’s engineers in proximity to the operating software developers, GPU designers, and hyperscalers whose workloads now dictate how memory must be built. These jobs are the thin but strategic layer that explains how Micron is repositioning itself for AI.

To quantify these shifts, we constructed a longitudinal index of Micron job postings from 2020 through 2025, augmented by archived listings from earlier years. Because raw counts are inconsistent — postings appear and disappear, and the overwhelming majority are concentrated in Boise — we normalized the data to a fixed baseline. Boise was assigned a constant value of 90 to represent the stable bulk of fab- and process-oriented roles, while postings in other U.S. hubs were indexed relative to that benchmark. The resulting series does not represent absolute headcounts, but rather directional changes in geography and function. Exhibits omit Boise to remove noise from the dominant baseline and to highlight relative growth in hubs such as Silicon Valley, Austin, and Denver.

Just as revealing is the kind of jobs being filled. The role mix shows Micron moving away from fab-bound process engineers toward advanced packaging, software-aware performance, and customer-facing positions. This shift in titles explains why the company must hire in new hubs: these jobs only make sense in places where engineers can collaborate directly with hyperscalers and GPU designers.

To be sure, this transformation carries risk. Building fabs in Idaho is one thing; competing in advanced packaging and systems engineering is another. Fixed costs rise, the balance sheet bears more leverage, and R&D cycles stretch. Missing a node in HBM or falling behind on packaging yield could be as damaging as missing a CPU generation was for Intel. The financial position can support the change, but execution risk is higher than ever.

Micron’s next test will be whether it can turn this cultural pivot into durable advantage. In an industry where memory makers have long been price-takers, AI may create an opening for Micron to become a price-setter — if it can keep pace with Nvidia and AMD’s roadmaps and match packaging and software to their cycles. For a company that has spent decades surviving cycles, the question now is whether it can define one.

This transformation is also a measure of leadership. Sanjay Mehrotra has steered Micron through downturns before, but the AI pivot is different: it is his bet that memory must be co-designed with the system, not bolted on afterward. Under his watch, Micron has begun hiring where its customers are, chasing packaging and software awareness, and positioning itself not as a commodity supplier but as a partner in the AI build-out. If he is right, Mehrotra will be remembered as the CEO who turned Micron from a survivor of cycles into a shaper of them.

The hiring evolution tells that story as clearly as the income statement. This is the metamorphosis — and it will define whether Micron remains a commodity survivor or becomes a strategic player in the AI age.

Tweet O’ The Week

Epistrophy In The News

On Schwab Network, I argued that the next phase of artificial intelligence will be less about chips and more about plumbing: “connecting one data center to the next.” Alibaba’s (BABA: NYSE) commitment of more than $53 billion to AI infrastructure shows the scale of capital required, but I remain skeptical of China’s AI capabilities compared to U.S. firms.

On NewsNation, the discussion centered on the proposed sale of TikTok USA from ByteDance. The sticking point isn’t just the mechanics of control but the valuation—Vice President JD Vance floated a price of only $14 billion. That number, if accurate, is a steal and practically a gift to Dell (DELL:NASDAQ), Oracle (ORCL:NASDAQ) (full disclosure: a stock I own), Andressen Horowitz, Silver Lake and whoever else is involved in this deal — when and if it actually closes.

📆 of Epistrophy Events

Availability This Week

I’m in New York City Monday and Tuesday, back in San Francisco after that. A quieter week, which makes for a good one to dig into questions that don’t fit into a broadcast segment or a quarterly slide deck.

Written reports are available to clients, with video summaries on YouTube, and of course our popular summaries of the summaries on Instagram, TikTok, and YouTube Shorts.

I’d love to hear from you and work with you to work ahead!. As always, comments, questions and ideas are appreciated.