The memory margin, and what else is inside the $2 trillion chip rally

Three chip stocks tripled or nearly tripled in ninety days and added roughly $2 trillion of market capitalisation between them. Micron is up about 240%, Intel about 216%, AMD about 186%, and the interesting number is buried inside the first of those — Micron's gross margin (the share of revenue left after the cost of making the product) went from around 39% to 84.9% in a single quarter. That is the number to read the rally through, and it does not explain everything the rally is doing.

What actually happened. Over Q2 2026, the AI chip trade broadened out of Nvidia. Marvell doubled. Arm gained 134%. Micron, Intel and AMD are now the tenth, eleventh and twelfth most valuable US technology companies by market cap.¹ Bloomberg Technology called it the best quarter ever for chip stocks.² Alongside the rally, Taiwanese authorities raided Super Micro Computer's offices in connection with alleged Nvidia chip smuggling to China — export-control enforcement tightening while prices ran.¹

The convenient reading is that the AI trade has matured: capital is spreading down the stack from GPUs into memory, logic, networking and IP licensing, because every layer is bottlenecked at once. That is one reading. It is not the only one the evidence supports.

The memory margin is the load-bearing datum

Start with Micron, because Micron is the anomaly. Memory is a commodity business. It has historically traded on capacity cycles, not pricing power. In the 2022–2023 downturn Micron posted negative gross margins. Software companies run at 80%-plus gross margins because their marginal cost of production is roughly zero. DRAM (dynamic random-access memory, the working memory that sits next to a processor) does not work that way; every bit shipped has a wafer cost.

An 84.9% gross margin on memory is, structurally, a software margin on a commodity product. The only way to sustain that is if demand outstrips supply for long enough that customers pay whatever is asked, and if the specific product being sold, in Micron's case, high-bandwidth memory stacked next to AI accelerators, is genuinely scarce and genuinely non-substitutable.

Both conditions look real for now. Micron reported HBM (high-bandwidth memory, the fast memory bolted directly onto AI training and inference chips) capacity sold out through 2026, with roughly $22bn in long-term customer contracts already signed.³ Revenue grew 346% year-over-year.³ The inference-economics frame — the working thesis that the binding constraint on AI at scale is running the models, not training them — makes a sharp prediction here: if inference throughput is memory-bound rather than compute-bound, memory suppliers capture rent that used to sit with GPU vendors. Micron's Q3 print is that prediction landing.

That is the piece of the rally I would treat as structural. The margin is anchored to a product-market fit that has a physical explanation.

Intel is the piece that doesn't fit

Now Intel. A 216% quarterly gain requires a story. The story on offer is that AI infrastructure demand is broad enough that even the laggard silicon vendor gets pulled up. This is not a story about product; it is a story about tape.

Intel's Gaudi AI accelerator line has not materially displaced Nvidia's H100 or H200 parts in datacenter deployments through the first half of 2026. Its foundry ambitions remain expensive and unproven at leading nodes. None of the primary datapoints that would justify a 216% repricing — a major hyperscaler design win, a foundry customer at 18A, a credible AI accelerator roadmap update — landed in Q2 at the scale needed to support the move.

The inference-economics frame does not extend cleanly here. Memory is bottlenecked; Intel's products are not the constrained part of the stack. What is left, once you strip out the memory story, is closer to what I would call AI performativity — the frame in which the scale of the AI spend narrative moves prices whether or not the underlying products deliver. Intel is inside the AI trade because Intel is a chip company and the AI trade is on. That is a momentum reading, not a fundamentals reading, and it should be labelled as such.

Broadening, or rotation?

The optimistic gloss on the quarter is broadening: capital moving down the stack because every layer is bottlenecked. The less flattering gloss is rotation: capital moving down the stack because Nvidia got expensive and money had to go somewhere adjacent.

The evidence is mixed. Micron and, on a smaller scale, Marvell fit the broadening story — there is a product-specific demand signal you can point to. Arm's 134% gain has a plausible thesis (custom AI ASICs licensing Arm IP) though the company's non-AI revenue muddies the read.¹ Intel and, to a lesser extent, AMD look more like rotation, with AMD at least having a defensible AI GPU roadmap that Intel currently lacks.

The two readings have different implications. Broadening survives a Nvidia earnings wobble because the demand is distributed. Rotation does not — it unwinds back into whatever the next narrative is, and the beta names give back the most.

Export controls as an invisible bid

One more thing to name. The Super Micro raid in Taiwan is part of intensifying enforcement against grey-market flows of Nvidia parts into China.¹ If enforcement is now biting, two things happen at once: Chinese demand that had been supporting prices exits the market (bearish for pricing), and domestic hyperscaler allocation gets less contested (bullish for pricing to buyers who can still buy).

Which effect dominates depends on how much of prior demand was Chinese pass-through. Nobody outside the enforcement agencies knows. It is worth naming that some non-trivial portion of the current pricing regime is being set by an export-control variable that is neither disclosed nor stable. This is an intelligence-explosion-signals data point in the map — export controls are now a price input, not just a policy input.

What to watch

Micron's next two quarters. If gross margin holds above 70%, the memory-margin regime is real and the frame extends. If it compresses back toward 50% as HBM supply catches up, the Q3 print was a peak-cycle artefact and the stock is priced for a margin that will not repeat.

Intel's foundry and accelerator disclosures at its next investor update. Without a product datapoint, the 216% gain is a tape story that will unwind at the first sign of AI-trade fatigue.

The next round of export-control enforcement actions, and any Chinese-demand disclosures in hyperscaler capex commentary. If the Super Micro raid is followed by further named enforcement, the supply constraint is durable. If it is a one-off, it is noise.

I would treat the quarter as one clear structural signal (memory), one clear momentum signal (Intel), and a lot of things in between that need another print to sort.

Glossary

Gross margin Revenue minus the direct cost of producing the goods, expressed as a percentage of revenue. Software runs high; commodity hardware runs low.

HBM (high-bandwidth memory) Fast memory stacked next to AI accelerators, used to feed data into training and inference chips.

DRAM Dynamic random-access memory; the standard working memory used across computing, historically a commodity market.

Inference Running a trained model to produce outputs, as distinct from training the model in the first place.

Hyperscaler capex The capital spending of the largest cloud operators (Microsoft, Google, Amazon, Meta) on datacentres and chips.

ASIC Application-specific integrated circuit; a chip designed for one workload, often licensed from IP vendors like Arm.

---

Footnotes

Footnotes

1. CNBC, "Record chip rally adds $2 trillion in combined value to Micron, Intel and AMD," 30 June 2026. https://www.cnbc.com/2026/06/30/ai-chip-rally-in-q2-adds-2-trillion-in-value-to-micron-intel-amd-.html ↩ ↩² ↩³ ↩⁴

2. Bloomberg Technology, "Chip Stocks on Track for Best Quarter Ever," 30 June 2026. https://www.youtube.com/watch?v=Hk4nN3lzHDY ↩

3. CNBC, "Micron stock jumps on Q3 earnings; memory demand and HBM contracts drive results," 25 June 2026. https://www.cnbc.com/2026/06/25/micron-stock-3q-earnings-memory.html ↩ ↩²