Micron, Sandisk, Marvell: Wall Street Stopped Pricing AI Memory and Interconnect as a Commodity Cycle
There is one argument running underneath every chip-stock target reset this week, and it is not really about chips. It is about whether memory, storage, and the wires between accelerators are commodity components that move on the old PC-and-mobile cycle, or mission-critical AI infrastructure whose demand scales with every model upgrade, every reasoning capability, and every agentic deployment.
Bank of America just answered that question with its wallet. On June 23 — a day the group was getting hit, not bid — Vivek Arya raised Micron to $1,500 from $950 and reframed DRAM and high-bandwidth memory as structural AI infrastructure rather than a cyclical good. The same desk lifted Marvell to $365 the same session and circulated a note arguing the broader memory-plus-interconnect complex represents another trillion-dollar opportunity for chip names. That is the tell. When one analyst makes the identical structural call across DRAM, NAND, and custom silicon on a down day, it is not a price target. It is a thesis.
The reframe: infrastructure, not inventory
For two decades, memory was the textbook cyclical. Capacity gets added, ASPs collapse, fabs idle, the cycle resets. The bear case on every memory name was always the same: this is as good as it gets, and the down-leg is coming.
The claim now is that AI broke that pattern. The memory required to serve a frontier model scales with the model — bigger context windows, longer reasoning chains, and persistent agents all consume capacity that does not follow the seasonal consumer-device cycle that defined the market for decades. HBM is the sharpest expression: every leading-edge HBM wafer Micron and its competitors commit to AI accelerators is a wafer not making commodity DRAM, so the supply that used to crater pricing is being cannibalized by the highest-margin product in the building. Supply tightness becomes structural rather than transient.
Micron handed the bulls a concrete proof point on June 22 with a multi-year supply and architecture agreement with Anthropic, paired with a strategic investment into the lab’s funding round. The detail that matters is not the dollar figure — it is that a frontier AI lab publicly tied its infrastructure scaling roadmap to a specific memory vendor’s product roadmap. That is the opposite of how you treat a commodity input you can second-source on price.
Micron (MU): the anchor
Micron is where the thesis is most fully priced and most fully tested. The stock ran more than 800% in a year before the June 23 macro drop, which knocked it roughly 12% to below $1,060 — not on anything Micron did, but on a South Korean chip-led selloff that dragged the entire complex lower into the print.
The Q2 setup frames the stakes: a record $23.86 billion in revenue, DRAM up more than 200% year over year, and Q3 guidance pointing toward a ~$33.5 billion midpoint with gross margins guided near 81%. The expected beat is already in the stock. The variable is forward guidance. A Q4 outlook stepping toward $38–40 billion with HBM pricing strength carrying into 2027 turns the macro dip into a textbook buy-the-dip. Caution on HBM volumes or pricing trajectory does the opposite, and the reaction would be uglier than a macro-driven flush. The analyst cluster — roughly $1,200 to $1,750, against a stock trading near $1,100 — implies the Street still sees upside even after the run, but that is consensus, and consensus is exactly what a derating unwinds first.
Sandisk (SNDK): the NAND leg of the same trade
If memory is infrastructure, storage is the part of the thesis the market has chased hardest. Sandisk — the pure-play NAND business spun out of Western Digital — is up roughly 4,800% over the past year and has multiplied several times in 2026 alone, ranking ahead of Micron among the year’s top S&P performers.
The structural argument is identical to Micron’s, ported to NAND. Datacenter SSD demand for AI inference and data pipelines has made the biggest end-market for flash far less sensitive to the price swings that PC and mobile dictated for years. Gross margins approaching 80% on a business that historically lived in the 20s tell you the mix shift is real, not cosmetic, and longer-term contracts are starting to give the kind of visibility memory businesses never used to have. Morgan Stanley’s framing — that the change in NAND’s demand structure is the thing that actually matters — is the storage-side echo of the BofA memory call.
The catch is that Sandisk has priced more of this future than Micron has. At a forward multiple in the high-50s to mid-60s, the stock is discounting sustained shortage and margin durability. The bull case needs NAND pricing to hold and datacenter mix to keep climbing; the bear case is a cyclical pricing reset arriving before the multiple has any cushion. This is the highest-beta expression of the thesis — most upside if it holds, most air underneath it if it doesn’t.
Marvell (MRVL): the bottleneck moves to the wires
The third leg is the one the memory framing tends to miss. As clusters scale, the constraint shifts from compute to data movement — getting bits between GPUs, across racks, and between data centers fast enough that the accelerators are not sitting idle. That is Marvell’s territory: custom silicon plus the interconnect and optical stack — DSPs, silicon photonics, co-packaged optics, retimers, and the Teralynx T100 switch — that the hyperscalers build their fabrics around.
The Street has re-rated it accordingly. Marvell is up more than 300% over the past year, joined the S&P 500 on June 22, took a strategic investment from Nvidia, and got the kind of endorsement that moves a tape when Jensen Huang called it the “next trillion-dollar company” at Computex. BofA’s $365 target sits inside a wide and contested range — some models reach toward $400 and beyond on optics and ASIC ramps, while more cautious desks flag execution risk and ASIC share questions at leading customers. The bull logic is clean: the more bandwidth AI models demand, the more packets touch Marvell silicon. The risk is that custom-silicon revenue is lumpy and customer-concentrated, and the valuation already assumes the optics and switching ramp shows up on schedule.
The broader AI-infrastructure read
Zoom out and the same logic governs the entire picks-and-shovels layer — Broadcom in custom silicon and networking, the optical names, the HBM supply chain. The market’s earnings growth has become startlingly concentrated in this cohort: strip Nvidia and Micron out of the information-technology sector and a recent quarter’s blended earnings growth roughly halves. That concentration is the bull case and the risk in one sentence. A handful of infrastructure names are carrying the index’s earnings, which is exactly why they trade like the whole AI trade — and why they sell off together on macro noise that has nothing to do with their order books, as June 23 demonstrated.
For the structural bull, those macro flushes are the entry points the thesis keeps providing: the names get marked down on Korean equity volatility or AI-bubble headlines, then recover on company-specific catalysts. That pattern has worked repeatedly this cycle. The question is always whether this is the dip that gets bought or the one that doesn’t.
What breaks it
The honest bear case does not argue the demand is fake. It argues three things. First, valuation: forward multiples across this group already discount years of sustained shortage and margin durability, which leaves no cushion if the numbers merely come in good rather than great. Second, the funding loop: a meaningful share of AI infrastructure demand is debt-funded hyperscaler capex, and any wobble in that spending hits memory, storage, and interconnect simultaneously — the same concentration that makes the cohort powerful on the way up makes it correlated on the way down. Third, the oldest risk in memory: NAND and DRAM supply growth outrunning demand and resetting pricing before the “this time is structural” thesis has been tested through a real down-cycle.
The cleanest way to hold the bull thesis and stay honest about it: this is a cohort trade. It re-rates together and it derates together. The structural argument — that AI turned memory, storage, and interconnect from commodities into infrastructure — is the most credible it has ever been. It has also never been priced this aggressively, and it has not yet survived a downturn. Both can be true at once, and the next two earnings cycles are where the market finds out which one dominates.