When all three major US indices fell in the second week of March, a small group of stocks moved in the opposite direction. SandDisk (SNDK) gained nearly 7 percent for the week, while Micron Technology (MU) rose more than 5 percent, both landing near the top of the S&P 500 movers list. The common thread behind both rallies is a single premise that the market had previously underestimated: AI inference computing is driving a new wave of memory demand, one that is accelerating faster than consensus models had assumed just a quarter ago.
SandDisk: A NAND Flash Leader Returns to the Public Market
The SandDisk name is familiar to anyone who followed technology markets in the 2000s and early 2010s. The company built its identity around consumer NAND flash storage products, becoming one of the dominant brands in portable memory cards and USB drives before Western Digital acquired it in 2016 for roughly $19 billion. In the years that followed, SandDisk was folded into Western Digital's corporate architecture and faded from view as a standalone entity.
In 2025, Western Digital announced it would spin off its NAND storage business and relist it as an independent company under the SandDisk name, with the ticker symbol reverting to SNDK. The business that emerged from that separation looks quite different from the consumer gadget brand of the previous era. The focus has shifted to enterprise-grade NAND solutions, particularly solid-state drives aimed at data centers, storage modules for edge computing infrastructure, and high-density storage arrays used in AI server deployments.
That repositioning matters because it changes the valuation logic for the stock. A company selling USB drives and SD cards is priced as a consumer electronics cyclical, sensitive to device upgrade cycles and retail demand. A company supplying storage infrastructure to AI data centers is priced as an AI infrastructure play, with demand tied to the expansion of model deployment and the growing scale of inference workloads. The week's rally reflects the market beginning to apply the second frame rather than the first.
The timing coincided directly with the approach of NVIDIA's GTC 2026 conference, with the keynote address scheduled for March 18. The central theme of GTC this year is the evolution of AI inference architecture, and data storage bandwidth is a critical variable in inference performance. Systems performing inference need to read large volumes of model parameters from storage at very low latency. SandDisk's enterprise SSD product line sits exactly at that intersection, and investors positioned ahead of the keynote in anticipation of renewed attention to the storage layer of the AI stack.
Micron Technology: HBM and the Memory Bandwidth Bottleneck
Micron Technology operates as one of three global leaders in DRAM manufacturing, alongside Samsung and SK Hynix. Its position in the AI memory cycle is more direct than almost any other company outside of the GPU makers themselves, because the type of memory that matters most at the frontier of AI computing, High Bandwidth Memory, is one of Micron's core product lines.
High Bandwidth Memory, or HBM, differs from conventional DRAM in how it is built and deployed. Rather than sitting on a separate memory module connected to a processor through a standard bus, HBM stacks multiple DRAM dies vertically and integrates them directly alongside the GPU on a shared substrate. The result is a dramatic increase in memory bandwidth. A single HBM3e module can sustain more than a terabyte of data transfer per second, roughly ten times the throughput of standard DDR5. For large language models running in inference mode, that bandwidth determines how many requests the system can process per second, which in turn determines the cost per query for the operator. Higher memory bandwidth means lower inference cost, which directly affects the economics of deploying AI at scale.
Micron's HBM3e production ramp has been one of the most closely tracked metrics among institutional investors covering the semiconductor sector. In its most recent earnings call, the company raised its revenue outlook for fiscal year 2026 and pointed specifically to AI-related memory demand, covering server DRAM and HBM, as the factor extending order visibility well into the second half of the year. Customer pull-through has been strong enough that supply remains the binding constraint rather than demand. NVIDIA's upcoming inference architecture announcement is widely expected to require greater HBM capacity per chip than the previous generation, which would mean Micron's production expansion is timed almost precisely to the demand uptick.
During model inference, the system must repeatedly read billions of parameters from memory to generate each output. Conventional DDR memory creates a bottleneck because its bandwidth cannot keep pace with the GPU's computational throughput, leaving processing cores idle while they wait for data. HBM eliminates most of that bottleneck by making memory bandwidth roughly proportional to compute power. As a result, every major new GPU architecture targeting inference workloads has demanded more HBM capacity, and that trend is expected to continue through at least 2028. For memory manufacturers like Micron, inference demand growth is more predictable than consumer electronics cycles and less sensitive to macroeconomic swings.
Ares Management and Blackstone: Alternative Assets Gain Ground
The week's strong performers extended beyond the semiconductor sector. Ares Management (ARES) gained 5.5 percent and Blackstone (BX) rose 4.6 percent, two of the largest alternative asset managers in the world moving in the same direction for a related reason. Both companies benefit when institutional investors conclude that traditional stock-and-bond portfolios can no longer deliver the diversification they were designed to provide, and the current environment is producing exactly that conclusion.
When government bonds decline alongside equities, institutions managing pension funds, endowments, and sovereign wealth vehicles need to find non-correlated sources of return. Private credit, infrastructure debt, and real assets have lower day-to-day price correlation with public markets, and both Ares and Blackstone are among the largest managers of those strategies globally. Their management fees are largely contractual, set at the time capital is committed and not immediately subject to market price volatility, which makes their revenue streams more visible and predictable in periods of public market uncertainty.
One additional factor sharpened the contrast this week. Reports emerged that JPMorgan is pulling back on its lending to private credit firms and writing down the value of certain software-backed loans. A major bank reducing its participation in private credit is, counterintuitively, positive for firms like Ares and Blackstone, whose lending platforms do not depend on bank facilities in the same way and who potentially benefit from reduced competition for deals as bank capital retreats. The structural shift toward private credit as a meaningful component of institutional portfolios did not begin this week, but this week's market dynamics put a sharper point on why that shift is continuing.
Adobe's Problem: Building AI Features Is Not the Same as Monetizing Them
Adobe (ADBE) fell 7.6 percent following its earnings release, and the reaction illuminated a tension that runs through a wide range of enterprise software companies right now. The headline figures were not dramatically bad. The broader problem was that Adobe's AI-powered creative tools, led by the Firefly image generation platform, are generating trial activity and engagement but converting to paid upgrades or higher-tier subscriptions more slowly than analysts had modeled. The gap between feature deployment and revenue realization proved wider than expected.
This is a pattern that markets are increasingly focused on as AI feature announcements have become standard across the software industry. In 2024 and 2025, many companies received valuation credit for embedding AI capabilities into their products, on the assumption that monetization would follow. In early 2026, patience with that assumption has shortened noticeably. Investors are now distinguishing more sharply between companies where AI features are producing measurable revenue acceleration and companies where AI is an internal productivity tool or a marketing narrative without a clear path to customer-level pricing power. Adobe's earnings placed it in an uncomfortable intermediate position: clearly investing in AI, clearly attracting users to AI features, but not yet converting that engagement into the kind of revenue growth that justifies an AI-premium valuation multiple.
The longer-term debate about Adobe's competitive moat is genuinely unsettled. Adobe's creative software suite, encompassing Photoshop, Illustrator, Premiere Pro, and the broader Creative Cloud ecosystem, carries switching costs that are unusually high. Professionals who have built years of workflows, keyboard shortcuts, file formats, and plugin dependencies around Adobe's tools do not switch lightly. That friction provides a durable floor under subscription retention. The risk is that AI-native design platforms, which start from scratch with generative tools at the core rather than bolting them onto legacy software, could attract the next generation of creative professionals without ever encountering the switching cost barrier at all.
Ulta Beauty and the Consumer Spending Signal
Ulta Beauty (ULTA) fell more than 14 percent after the company revised its guidance for same-store sales growth in fiscal 2026 lower than the market had expected. The sell-off was sharp, but the underlying logic is straightforward: when oil prices spike, the increase functions as a de facto tax on consumer spending, particularly for households in lower and middle income ranges that spend a larger share of their income on fuel. The extra cost of filling a gas tank does not disappear from the household budget. It comes out of somewhere, and discretionary spending categories, including beauty, personal care, and specialty retail, tend to absorb those cuts earliest.
Ulta's guidance revision may be the first publicly reported corporate acknowledgment that the oil price surge is already filtering through to consumer behavior in ways that are measurable at the business level. If that is the case, it is a leading indicator worth watching carefully across other consumer discretionary names. Restaurant chains, apparel retailers, and entertainment venues are the categories most exposed to the same spending compression dynamic. Whether those companies confirm or contradict Ulta's signal in their own upcoming reports will say something important about the depth and breadth of the consumer-level oil shock transmission.
| Stock | Weekly Move | Primary Driver |
|---|---|---|
| SNDK (SandDisk) | +6.9% | AI inference storage demand; pre-GTC positioning |
| ARES (Ares Management) | +5.5% | 60/40 breakdown; private credit allocation grows |
| MU (Micron Technology) | +5.1% | HBM demand strength; NVIDIA inference cycle |
| BX (Blackstone) | +4.6% | Alternative assets inflows; bank credit pullback |
| IP (International Paper) | +4.5% | Packaging demand recovery; inventory restocking |
| ADBE (Adobe) | -7.6% | AI monetization lag; Firefly conversion below consensus |
| ULTA (Ulta Beauty) | -14.2% | Guidance cut; consumer spending compressed by oil prices |
| PODD (Insulet) | -6.9% | Medical device reimbursement pressure |
| MOS (Mosaic) | -6.5% | Near-term fertilizer demand revision lower |
What This Week's Movers List Is Actually Saying About Market Structure
The distribution of gains and losses this week traces two clear lines through the market. The first is that the AI infrastructure investment cycle remains intact. Capital continues to flow toward semiconductor companies, memory manufacturers, and AI computing infrastructure, and the demand signals from both the chip makers and their customers are pointing in the same direction. The second line is that the consumer economy is beginning to absorb the pressure from elevated oil prices, and Ulta's revised guidance may be an early warning that the pass-through from pump prices to discretionary spending is already underway rather than merely theoretical.
The structural divergence, specifically the pattern of strength close to AI infrastructure and weakness close to the consumer, is not a signal to aggressively reposition a portfolio in a single direction. It is a signal to be more deliberate about understanding what each position is actually exposed to. A broad index ETF blends both sides of this divergence, smoothing out what is in reality a significant internal rotation. Investors who want to understand what the market is doing on a fundamental level need to look past the index numbers and track where the actual money is moving within them.
If NVIDIA's GTC keynote on March 18 confirms specific memory bandwidth requirements for the next-generation inference architecture, Micron and SandDisk could extend their gains into the following week. A vague or underwhelming announcement could trigger near-term profit-taking in both names. For Ares and Blackstone, the medium-term outlook depends on the Federal Reserve's rate path, since the cost of capital for private credit strategies is sensitive to how long rates stay elevated. Adobe's next meaningful data point will be the following quarter's earnings, where management will need to present concrete evidence that AI feature engagement is translating into subscription revenue acceleration. Without that evidence, the stock's AI premium will remain difficult to justify.
This week's market did not deliver an AI-driven rally for the whole market. What it delivered was a clearer picture of where genuine demand conviction exists inside a market that is otherwise navigating real uncertainty. For investors trying to separate signal from noise, that picture is worth more than most of what the headline index moves communicate.