Micron Is the Third-Best Performing S&P 500 Stock in 2026. Here Is Why High-Bandwidth Memory Changed Everything.

The Number That Defines Micron in 2026

Micron Technology closed at $1,211.38 on June 22 up 324.61 percent so far this year. That makes it the third-best performing stock in the entire S&P 500 this year behind only SanDisk and Western Digital. I grew up thinking of Micron as the classic example of a boring cyclical chipmaker. Something changed and the one-word answer is hbmMicron's HBM revenue surpassed $1 billion in a single quarter for the first time in the company's history earlier this year. That milestone more than anything else in the filing is what changed the price of the entire company

Micron reports fiscal third-quarter 2026 earnings on June 24 after the market close. Wall Street expects roughly $35 billion in revenue and adjusted earnings per share of $20.57. Going back to 2023 those numbers would have sounded like a typo. Micron was posting a loss that year in the depths of the memory crisis. Between then and now two things happened. The price of Micron recovered.memory which is the cyclic part. AI computer architecture began to require HBM which is the structural part. Only three companies in the world can build HBM at a significant scale and Micron is one of them

What High-Bandwidth Memory Actually Is

Standard DRAM the memory that's been inside computers for decades is fast enough for almost everything you do on a laptop. It's not fast enough for training a large language model. Moving the volume of data that AI workloads require between the processor and memory at the speed those workloads demand is a genuinely different engineering problem

HBM solves this by stacking DRAM chips vertically and connecting the layers with via silicon vias tiny electrical channels drilled directly through the silicon. That stack is then located directly in the same package as the GPU connected with thousands of microscopic connections called microstrokesThe result is bandwidth about an order of magnitude greater than the DRAM that sits on the side on a regular circuit board. It's memory designed to live within an inch of the processor rather than across the motherboard

Every NVIDIA H100 H200 and GB200 ships with HBM in the package. NVIDIA's upcoming Blackwell Ultra chips require even more per GPU. AMD's MI300X uses it. Google's TPU v5 uses it. Amazon's Trainium2 uses it. I can't find a credible way to train a frontier model or run full-scale inference that completely bypasses HBM. Only three companies build it at scale.significant: Samsung SK Hynix and Micron. A three-company oligopoly that finds itself under a lawsuit that no one can avoid is the investment argument in a sentence

Micron's Position in the HBM Market

Micron wasn't the first here. For most of 2024 it was the smallest and least advanced of the three HBM suppliers behind Samsung and SK Hynix in both track record and customer relationships. Its current-generation product HBM3E received NVIDIA qualification during 2024 and Micron has ramped up production considerably since then

Watch the June 24 release for one thing above all else: HBM's revenue growth and any guidance management provides on supply commitments for 2026 and 2027. If Micron shows its stake in HBM continues to expand with prices holding up even as supply increases the structural bullish case remains intact. If growth disappoints or management hints that Samsung's capacity additions are pressuring the price the stock will movestrongly at the news. There is not much space in the middle

The broader memory market is also important and it's worth separating the two stories entangled in Micron's results. Standard DRAM prices have recovered from the 2022-2023 low and NAND prices have also improved. Micron is experiencing a cyclical recovery in commodity memory and a structural shift toward HBM at the same time and that's not the same thing. HBM's growth is structural. The DRAM and NAND recovery iscyclical and cyclical things reverse. Micron's net income is projected to rank second only to NVIDIA in the entire PHLX Semiconductor Index by 2026 and 2027 which is the market betting that the HBM story is real enough to keep profitability elevated after the tailwind fades

Why Memory Has Always Been a Brutal Commodity Cycle

Before I explain why HBM might be different I want to be honest about what a corporate memory it has always been. That story is the only reason anyone is skeptical of this story in the first place

DRAM and NAND are fungible. A gigabyte of standard Samsung DRAM does roughly the same job as a gigabyte of SK Hynix or Micron. When a product is fungible price is the only lever that matters and the price in a commodity market is set by whoever has the most supply not whoever has the best engineering

The second problem is that memory factories are extremely capital-intensive and take years to build. A company adds capacity when demand seems strong and prices are high but by the time that factory comes online two or three years later demand may have cooled and two competitors may have made the same decision at the same time. The result is chronic oversupply exactly when no one wants it followed by underinvestment that manifests as shortages a few years later. It's a textbook capital cycle with an unusually long and relentless lag

The third problem actually shows up on the bottom line. Factories cost billions of dollars and that cost doesn't go away when demand declines. The factory depreciates whether it runs at 95 percent or 60 percent. When the market has too much supply the marginal producer keeps selling below cost just to cover part of that fixed cost because an idle factory wastes cash faster than a factory selling at a loss. That's the dynamic that pushed Micron into losses during the recession.2022 to 2023 and has been repeated throughout this industry for decades

Does HBM Actually Change the Physics of the Cycle?

So here's the real question behind this whole article. Is HBM just DRAM with better margins for now taking advantage of the same commodity cycle before it inevitably pulls it back? Or is there something about HBM's structure that actually breaks the pattern?

My honest answer is partially and I want to be precise about which part

HBM is not a commodity product like standard DRAM is. Qualifying a new HBM product with a customer like NVIDIA is a long and expensive process often over a year because the memory must be validated with a specific GPU package thermal wrap and reliability bar. That qualification cost is a real barrier that standard DRAM does not have where a buyer can often switch suppliers with comparatively little friction

The biggest structural change is contracting. A significant part of HBM's production is now sold through multi-year supply agreements negotiated directly with a handful of buyers who need it;The Anthropic deal mentioned above is a visible example of that pattern. This is a fundamentally different demand curve than spot market DRAM where the price is set daily by whoever needs inventory that week. A multi-year deal locks in volume and to varying degrees price insulating some of a supplier's revenue from the exact oversupply dynamics I just described

I don't think this makes HBM immune to the cycle. It reduces the portion of Micron's business that is exposed to it. Standard DRAM and NAND which still make up a large portion of Micron's revenue remain fully commoditized and fully cyclical. HBM is the part of the business that behaves differently not the entire company

A Worked Example: Fixed Costs, Utilization, and Gross Margin

Let me make the above mechanism concrete with a stylized example. None of these are actual figures from Micron. I built a simplified model to show why the swings get so violent

Suppose a factory costs $10 billion a year to operate once depreciation facilities and staffing are included and that cost is fixed no matter how much the factory actually produces. At full utilization let's say it produces 500 million memory units a year. The fixed cost per unit is 10 billion divided by 500 million or $20. Add $5 of variable cost per unit for materials and energy and the costTotal per unit in full use is $25

In a strong market let's say the average selling price or ASP is $35 per unit. Gross margin is 35 minus 25 divided by 35 or about 28.6 percent. Healthy normal business

Now the crisis is underway. Demand falls and the factory ships 400 million units instead of 500 million but the $10 billion in fixed costs are not reduced by this because the equipment still depreciates whether you use it or not. The fixed cost per unit increases to 10 billion divided by 400 million or $25. Add the same variable cost of $5 and the total cost per unit rises to $30. Meanwhile the pricealso falls because everyone in a commodity market is fighting for the same shrinking pool of buyers say 40 percent down to $21 per unit. The gross margin is now 21 minus 30 divided by 21 which is negative 42.9 percent. The company is losing money on every unit it ships and it is still the rational choice because an idle factory loses even more

That's the arithmetic behind Micron's actual losses in 2023

Now repeat the crisis for a hypothetical HBM product sold under a multi-year supply agreement. Suppose the contract sets a price of $80 per unit for a committed volume regardless of what the DRAM spot market is doing that quarter.43.75 percent and that number doesn't move even when the price of the commodity outside the contract is plunging as it did in the previous scenario. The contract not the spot market is what sets the price for that portion of revenue

Case Study: The 2017 to 2019 DRAM Supercycle and Bust

If you want to see why I'm not completely convinced that HBM will permanently break the cycle look at what happened the last time people said memory had changed structurally

During 2017 and 2018 DRAM prices rose sharply driven by the growth of smartphones PC demand and the first wave of data center and cloud construction. Samsung SK Hynix and Micron all posted some of their best results ever. A version of the exact argument I'm making about HBM was also largely made about DRAM back then. Cloud and hyperscale demand were supposed to be structurally differentFrom the old PC cycle buyers were bigger and tighter and memory makers had become more disciplined about capacity. Micron stock rallied strongly during that stretch along with the rest of the sector

Then demand growth slowed. Smartphone unit growth stagnated and some large cloud buyers had placed excessive orders and needed to work through inventory instead of continuing to buy. At the same time the capacity that Samsung and SK Hynix had committed to building during the boom began to work because factories take years to build and construction does not stop just because the market has cooled since the decision was made. Prices fell sharply during 2019 and continued to fallfor the better part of two years. Micron's profits and stock took a hit

The "this time is different" argument proved wrong again or at least early. I mention this not to delve into HBM's current thesis but because it is an honest precedent. The same industry many of the same executives made a similar structural argument seven or eight years ago and the commodity cycle reasserted itself anyway

Where This Breaks: The Counterargument

Let me properly address the skeptical case because it deserves more than a paragraph

The multi-year contracts I described are real but not infinite. Most last a few years not a decade. When it comes to renewal prices are renegotiated and if HBM's supply has grown faster than the number of companies actually training frontier models that renegotiation could favor the buyer. NVIDIA Google Amazon and Anthropic will no longer voluntarily pay cycle-resistant premium prices if memory makers have added enough capacity by then thatHBM starts to look like a normal product with additional steps

Capacity is the thing to watch out for. Samsung and SK Hynix are investing heavily in catching up and this is a business where a company that falls behind in one generation can catch up in the next because the process moves through generations not gradually. If all three suppliers plus potentially newer entrants such as Chinese memory makers building domestic capacity end up with roughly comparable HBM products at the same time the qualification barrier I described above becomes much less useful asmoat because customers will have rated all three anyway

There is also a risk on the demand side and I think it is underweight. The entire HBM bullish argument assumes that demand for AI inference and training continues to grow indefinitely at a rate close to its current rate. If AI capex growth slows even temporarily because some hyperscalers decide they built ahead of actual use demand for HBM could decline at the exact moment supply from three well-funded competitors increases. That's more or less the setupwhich preceded the 2019 crash I described above only AI accelerators replaced smartphones

My honest opinion is that HBM's contractual structure and rating barriers make the down cycle shallower and shorter than a pure commodity crash but not eliminate it. Anyone who writes off Micron as a cycle-proof perpetually repriced company is betting against the entire history of this industry

How I'd Actually Use This

I'm not going to tell you to buy or sell Micron. If this article sounds like stock advice I've spelled it wrong. What I'm really doing with a story like this is using it as a lens through which to read the next earnings report and the ones after that

This is the order in which I would ask the questions for the June 24 and subsequent printing. First is HBM's revenue still growing as a percentage of the total and is management willing to give real numbers on supply commitments for 2026 and 2027 instead of vague language which often indicates that commitments are softer than the headline suggests? Second what is happening to gross margin and specifically whether management can attribute any weakness to the changeof the mix towards a lower margin standard memory rather than pricing pressure within HBM itself because these are very different problems. Third I would track the capex guidance of all three HBM suppliers not just Micron because the supply side of the next drop is always built quietly years before anyone notices

I admit that I find the multi-year contract argument for HBM more convincing than I expected.I also think it's easy to convince yourself to believe that a cyclical business has become structurally correct before being proven wrong and the 2017 to 2019 stretch is recent enough that you don't want to write it off. If I were building a model of this business I would treat the HBM segment as durable but not permanent something closer to a long-term contract with real if imperfect downside protection sitting on top of a standard memory business that remains exactly ascyclical as it always was

The Bottom Line

Micron's 2026 stock move is real and HBM's story behind it is no exaggeration. HBM's revenue surpasses $1 billion in a quarter for the first time a multi-year Anthropic investment and supply deal and a net income ranking behind only NVIDIA in the entire PHLX semiconductor index all point in the same direction. But the memory has looked structurally different before most recently in 2017 and 2018and it wasn't different enough. The honest argument for appropriating this story is that HBM's qualification barriers and multi-year contracts should make the next crisis less profound than the last not that they have repealed the cycle entirely. Look at the contracts as renewal approaches look at the capacity that all three suppliers are building at once and remember that the factories that are ramping up today because demand seems unstoppable are exactly the factories that flood the market every time it finally cools