Navigating AI Volatility: Why We Hold US Memory

The Volatility Is Real. Our Answer Is Process.

AI-linked stocks have swung hard in recent weeks. Memory and storage names move even more than most, because they carry high beta. When headlines turn, they fall fast. When demand news is good, they jump. That volatility is uncomfortable. On its own, it is not a reason to sell.

Our answer is not a guess about the next move. It is a repeatable process. Our picks are data-driven, not gut-driven, and every name must re-qualify each month against the same systematic screens. Only the strongest opportunities stay in the portfolio. Three pillars run in sequence: Stock Universe, then Optimizer, then Risk Overlay.

We are also not fully invested. As of early July our equity exposure sat near 49% of the account, in Cruising mode. That is the honest signal of how our systematic process reads the market right now. Constructive, but far from euphoric.

The image below is an illustration of that idea, not a screen from our system.

Why We Hold These Names

Four forces drive the decision, and the sections below explain each one. Forecast sales and profit growth run far above the broad technology sector. A supply squeeze keeps memory pricing firm. A shift to multi-year contracts is making revenue steadier than in past memory cycles. And this same group already led the portfolio in a down June. The headline numbers stand out.

The US Memory and Storage Names We Hold

Inside the current portfolio, four names make up our memory and storage allocation. Each earned its place through the same monthly screen, and each is priced and rebalanced as an equal-weight position.

Micron (MU) is the one true US-based memory maker. It produces DRAM and flash storage, including the HBM that AI accelerators depend on. We have held it since December 2025. SanDisk (SNDK) is the flash-memory business that separated from Western Digital, focused on NAND flash. We added it in March 2026.

Seagate (STX) and Western Digital (WDC) build the high-capacity hard drives that store the huge data sets AI systems create and consume. Seagate has been in the portfolio since February 2026, and Western Digital since December 2025. Together, these four sit across the full memory and storage chain, from AI accelerator memory to bulk data storage.

This group did more than hold up in June. It led. In a month when the S&P 500 fell, our memory and storage names were the four largest positive contributors to the portfolio.

Combined, these four positions added about +5.5% to portfolio performance in June, while the broader market declined. SanDisk was the single largest contributor. You can see the full attribution in our June review.

Why Memory Growth Runs Ahead of the Broad Tech Sector

The reason we hold these names is not their recent price move. It is the strength and direction of their forecasts. When we screen the market, we weigh profitability and growth, the quality of those earnings, the direction of analyst revisions, and price strength. On those measures, the memory complex currently stands out against the wider technology sector.

The table below compares our four names to the broad US technology sector on forecast growth and on forward P/E. The sector figure is the median of the largest US information-technology companies, computed from our own primary data.

Consensus analyst estimates as of July 9, 2026, computed from primary data. Memory and storage are recovering from a cyclical low, so forecast growth is high relative to a normal year. These are analyst estimates, not forecasts we endorse.

Two things stand out. First, all four carry forecast sales and profit growth well above the sector median. Second, Micron trades at a strikingly low forward multiple of about 6.4 times expected earnings, far below the sector. That reflects the market pricing in the memory cycle eventually cooling. Seagate, Western Digital and SanDisk trade at higher multiples than Micron. Our screen applies no fixed valuation cap. It backs a higher multiple when growth, earnings quality and the direction of revisions support it, and it leaves the market-wide downside to the Risk Overlay rather than to a valuation rule. Paying up is a genuine trade-off, and we respect investors who choose differently.

The Supply Picture: Korea, Long-Term Contracts, and Policy

Strong forecasts only matter if the demand behind them is durable. Here, the supply side of memory is doing something unusual. It is getting tighter, and staying tight.

The two largest memory makers in the world, Samsung and SK hynix, are both based in South Korea. Both are pouring wafer capacity into high-bandwidth memory for AI, because that is where the best pricing is. Every wafer moved to that fast AI memory is a wafer not making ordinary memory. That discipline tightens supply across the whole market, and firmer pricing lifts every player in the complex, including our US names.

The way memory is sold is also changing. For years, memory was bought on short cycles at spot-like prices. Now the fastest AI memory is being committed years in advance under long-term agreements. Buyers want to lock in supply, and sellers want to lock in demand. That shift makes revenue steadier and more visible than in past memory cycles.

There is also a demand reason for supply staying tight. Applied AI means real AI features built into everyday products and daily business work. It is still in its early stages, and most companies are only beginning to roll it out. As more of these systems move from testing into daily use, they will need far more memory and storage. That points to demand rising for years, rather than fading after one buying wave.

The scale of demand is not our claim alone. Nvidia's chief executive has said publicly that both high-bandwidth memory and networking equipment are likely to stay in short supply for several years, simply because AI demand is so high. That comment reaches beyond memory. It also points to the networking layer, where we hold Ciena (CIEN) since January 2026, Lumentum (LITE) since December 2025, and Viavi (VIAV) since June 2026. These companies supply the optical connections that tie AI data centers together.

Policy adds another layer of support. The US government treats domestic memory and advanced chips as a national-security priority. Funding support for US chip manufacturing, and export limits on the most advanced memory reaching some overseas rivals, both tilt the field toward established US and allied producers. These are slow-moving forces, but they point in the same direction as the demand story.

The image below illustrates that tightening supply picture.

Why We Hold So Many Chip Names This Month

A reader scanning the July portfolio will notice something. Most of the names sit in one broad area: the hardware that powers AI. That is not a theme we chose by hand. It is where the data keeps pointing.

Each month, every position must re-qualify. Whatever names and sectors post the strongest fundamentals rise to the top of the list. Right now, the companies with the best combination of growth, earnings quality and rising forecasts are clustered in the AI supply chain. So the portfolio reflects that, without us forcing a view. Our fifteen July positions group into a few connected layers:

Memory & storage

MU and WDC (both held since December 2025), STX (since February 2026) and SNDK (since March 2026): the AI memory and data-storage chain.

Semiconductor equipment

AMAT (since April 2026) and FORM (new in July 2026): the tools and test systems that build and verify advanced chips.

Logic & custom silicon

AMD (since June 2026), MRVL and INTC (both new in July 2026): the processors and custom chips that do the computing.

Optical networking

LITE (since December 2025), CIEN (since January 2026) and VIAV (since June 2026): the optical gear that connects AI data centers.

Power & boards

BE (since December 2025), TTMI (since March 2026) and VICR (new in July 2026): on-site power, printed circuit boards, and power-conversion modules for AI systems.

This is where our process shows its value. The Stock Universe screen surfaces the strongest fundamentals. The Optimizer then decides which qualified names to hold, not how big each position should be, since every position is equal-weight. The Risk Overlay sets how much of the account is invested at all. None of it rests on a personal opinion about AI.

How We Navigate the Volatility

Owning strong companies is only half the job. The other half is not being hit too hard when the whole market falls. That is the Risk Overlay's role. Every night, the system recalculates more than 20 risk indicators across markets and sets our equity exposure for the next day.

It is not a prediction machine. It does not claim to know where the market goes next week. It is time-tested to control drawdown, not to forecast prices. As of early July it had us near 49% invested, in Cruising mode. During June's weakness it had reduced exposure to about 26%, before returning to 49% at the start of July.

This is the F1 discipline. Know when to accelerate, and know when to brake. In a market-wide selloff, high-beta memory and storage names tend to fall hardest. Because the overlay cuts total exposure in exactly those conditions, our most volatile names are the ones that gain most from the brake.

One honest limit matters here. The overlay reads market-wide signals and sets exposure for the whole book. It does not protect a single stock that falls on its own while the rest of the market holds up. If one memory name's demand story were to break, the overlay would not offset that specific drop. Two other defenses handle single-name risk instead. Every position is equal-weight, so no single name is oversized. And every name must re-qualify each month, so one whose case weakens can drop out at the next rebalance.

The live reading below shows where that leaves us today.

“We hold memory for the forecasts, the tightening supply picture, and a process that keeps re-testing the case each month. And we let the risk system, not our nerves, decide how much of the account is exposed.”

What Could Go Wrong

A balanced view has to include the other side. Memory is a cyclical business. Prices that rise on tight supply can fall just as fast when new capacity arrives or demand slows. Today's high forecast growth partly reflects a recovery from a recent low, and recoveries do not last forever.

Valuation is a real risk too. Seagate, Western Digital and SanDisk are not cheap on forward earnings. If AI hardware demand were to cool, or if buyers worked through their current orders, the same forecasts that look strong now could be revised down. A broken demand story in any one name would hurt, and the overlay would not shield that single position.

This is exactly why our exposure sits near 49% rather than fully invested. Our systematic process is constructive on this group, but it is not betting the account on it. We do not take the bull case or the bear case on faith. The rules decide, they re-test every month, and the risk overlay sets the exposure. That is how we aim to navigate the volatility rather than predict it.

If You Choose to Copy

Veloris Capital is a Popular Investor on eToro, and community members can choose to copy the AlphaWizzard strategy. We do not manage anyone's money. If you choose to copy, eToro technically replicates our trades proportionally into your own account. You stay in control of your own account at all times, and copying happens at your own responsibility. When we rebalance the portfolio each month, those changes mirror into copiers' accounts through eToro.

Related reading: our deep dives on Micron, SanDisk and Western Digital go name by name, and Sector Rotation: Are We an AI Portfolio? explains how these positions fit the wider book.

Micron (MU), SanDisk (SNDK), Seagate (STX) and Western Digital (WDC) are current Veloris Capital portfolio holdings as of the publication date. Past performance is not an indication of future results. Your capital is at risk.

Weekly investment newsletter, every Sunday — positioning, performance vs SPY/QQQ, and what we're watching next. No spam.