Memory Customers Can Get Only 65% of What They Need, VanEck Manager Says

Customers in the AI memory supply chain are securing only 65% of their actual needs, according to VanEck Emerging Markets Equity Portfolio Manager Ola El-Shawarby, who says the shortfall reinforces her long-term conviction in the AI theme. She returned from a recent research trip to Korea and Taiwan, where she met with participants across the memory supply chain. Her central finding: the industry is comfortable with its outlook through 2027.

That comfort is not casual optimism. It is rooted in a structural imbalance between what hyperscalers want and what memory makers can physically deliver. Demand significantly outpaces supply, El-Shawarby said, and that gap is doing the heavy lifting for sector fundamentals.

The numbers frame the stakes. VanEck, the firm behind the analysis, oversees $161.7 billion in assets as of September 30, 2025. When a manager at a firm of that size flags a supply bottleneck in AI memory, the positioning implications reach across emerging markets portfolios, semiconductor equities, and the broader AI investment complex.

Why does the memory supply chain matter for AI?

Memory is not a glamorous corner of the semiconductor industry, but El-Shawarby's framing puts it at the center of the AI build-out. In her words, the supply chain is what "basically enables the build-out of data centers and for the hyperscalers to continue to develop and advance their capabilities."

That is the causal chain behind the story. Hyperscalers want more data centers. Data centers need advanced memory. Memory makers — led in Korea by Hynix and Samsung — are the gatekeepers. When those gatekeepers cannot meet demand, every downstream AI ambition inherits the constraint.

Her trip covered both Korea and Taiwan, the two geographies most intertwined with global memory and chip manufacturing. Meetings with what she described as "different participants, in particular, the memory supply chain" gave her a ground-level view that public filings and earnings calls rarely capture.

The verdict from those meetings was direct. Participants are "very comfortable with 2027," she said — a three-year visibility horizon that most industries would envy.

What are Hynix and Samsung actually doing?

The most strategically significant detail from El-Shawarby's trip concerns pricing behavior. Hynix and Samsung are favoring long-term trade agreements over maximizing short-term profits, she reported.

This is a deliberate trade-off. The two Korean giants are, in her words, "sacrificing immediate earnings for prolonged earnings duration and greater visibility."

Read that against the 65% fulfillment figure. In a market where customers cannot get a third of what they need, sellers hold enormous pricing power. A pure short-term optimizer would squeeze every dollar out of that scarcity now. Hynix and Samsung are instead locking in multi-year arrangements.

Why would they do that? El-Shawarby's framing supplies the answer: duration and visibility beat a single blowout year. Long-term agreements convert today's shortage into a durable revenue base, align memory makers with hyperscaler build-out timelines that extend toward 2027 and beyond.

For investors, this behavior matters because it changes the shape of the earnings story. Instead of a spike-and-crater cycle — the historical pattern in commodity-like memory markets — the sector is signaling a plateau of sustained demand.

Is the 65% shortage a warning sign or a buying signal?

El-Shawarby treats it as the latter. With customers only able to secure 65% of their needs, she argues, the imbalance "reinforc[es] strong sector fundamentals and support[s] conviction in positioning."

There is a second reading worth naming, even if the source does not spell it out as a risk: a 35% unmet-need gap is also a production constraint on the AI build-out itself. The same shortage that flatters memory makers' order books limits how fast data centers can actually come online. El-Shawarby does not frame it this way — her emphasis is on sector fundamentals — but the figure cuts both ways by definition. What customers cannot buy, they cannot deploy.

Her conviction, though, rests on the direction of travel. The shortage exists because demand keeps growing faster than supply can respond. That is the signature of a structural theme, not a cyclical blip.

What should investors watch next?

El-Shawarby is precise about the monitoring points. Investors should closely watch two things, she said:

• Hyperscaler capex trends — the pace of data center spending by the large cloud and AI platforms.
• The sustainability of AI investment and monetization on the client side, primarily in the United States.

The second point is the load-bearing one. The memory supply chain can stay tight only as long as someone downstream is willing to keep paying. U.S. client-side monetization — whether AI products generate returns that justify continued spending — is the ultimate source of the demand El-Shawarby observed in Korea.

The chain runs in one direction. U.S. monetization funds hyperscaler capex. Hyperscaler capex pulls memory orders. Memory orders justify the long-term agreements Hynix and Samsung are signing. If the first link weakens, the comfort with 2027 gets tested.

The ground-truth premium

Part of what gives El-Shawarby's findings weight is method. "I visited Korea recently, and it was very exciting to meet with different participants, in particular, the memory supply chain," she said. On-the-ground conversations with supply chain participants in Korea and Taiwan offer a read on order books and pricing posture that arrives before it shows up in reported financials.

Her takeaway from those conversations was not ambiguity but alignment. Supply chain participants see demand outrunning supply far enough, for long enough, that 2027 sits comfortably inside their planning horizon.

VanEck's own history frames the institutional context. The firm, founded in 1955, has a record of thematic early calls — gold investing in 1968, emerging markets in 1993, and exchange traded funds in 2006. Its emerging markets team treating the AI memory chain as a conviction position extends that pattern into the current cycle.

The bottom line

One number carries the story: 65%. Customers in the AI memory supply chain can fill only about two-thirds of their requirements. Two of the world's dominant memory producers, Hynix and Samsung, are responding by trading near-term profit maximization for multi-year agreements and visibility.

El-Shawarby's condition for the thesis to keep working is equally concrete: hyperscaler capex must hold, and U.S. client-side AI monetization must prove sustainable. Watch those two indicators, she advises, and the rest of the supply chain — including the shortage itself — will tell you whether comfort with 2027 still holds.