Intel CEO Lip-Bu Tan is warning that the AI boom’s memory problem is getting worse—and that the strain is reaching the hardware ordinary people buy.
Speaking at the AI Infrastructure Forum in Santa Clara on September 15, Tan said limited memory supply was delaying projects and making components difficult to secure for mid- and low-priced phones and laptops, according to Seoul Economic Daily’s report.
That warning connects two markets people usually think about separately: enormous AI data centers and the next reasonably priced computer on a store shelf.
The connection runs through memory manufacturing. AI systems need large amounts of specialized memory, and suppliers are directing resources toward that demand. The consequences depend on which products they expand, which customers they serve, and how quickly additional production arrives.
For readers trying to understand Tan’s warning, the central question is straightforward: How does building more AI infrastructure make memory harder to obtain for an ordinary laptop?
Why memory matters to Intel’s warning
A processor needs a steady supply of data to do useful work. Memory holds that data close enough for the processor to use it quickly. Improving the processor alone has limited value when the rest of the system cannot keep up.
AI makes that relationship especially important. High-bandwidth memory, or HBM, stacks memory chips and connects them through a very wide interface, allowing large amounts of data to move quickly. It sits alongside specialized processors used for demanding AI workloads. Micron’s technical explanation of HBM describes how this arrangement helps keep processors busy.
Micron is a memory and storage chip manufacturer, separate from Intel. Along with Samsung Electronics and SK hynix, it supplies memory used in computing hardware. These companies matter to Tan’s warning because processor companies and equipment makers depend on the availability of complementary components.
How AI demand reaches phones and laptops
An AI server and a budget laptop do not simply compete for an identical memory stick. The pressure spreads through manufacturing resources and product priorities.
Several categories sit underneath the broad word “memory”:
| Type | Main job | Common uses |
| --- | --- | --- |
| High-bandwidth memory, or HBM | Moves large amounts of data quickly through stacked chips | Specialized AI processors |
| Conventional DRAM | Holds data currently in use | PC and server working memory |
| Low-power DRAM, including LPDDR | Provides working memory with lower power consumption | Phones, laptops, and some AI servers |
| NAND flash | Keeps stored data when power is off | Solid-state drives and device storage |
HBM is part of the DRAM family, but producing it places different demands on manufacturing resources.
In its December 2025 earnings presentation, Micron warned that growing HBM demand puts additional pressure on manufacturing capacity. Producing more of this specialized memory uses resources that otherwise support conventional DRAM production.
AI demand also extends into other memory categories. SK hynix’s August product presentation included low-power DRAM modules optimized for AI servers. Technologies associated with portable devices also have applications inside data centers.
These production tradeoffs help explain the pressure Tan describes. Their contribution varies by memory product and supplier, but they connect a laptop’s component supply to decisions being made for much larger computing systems.
What the dramatic price claim actually tells us
Seoul Economic Daily’s report attributes a five-to-sevenfold memory-price increase to Tan. It does not specify the products, comparison period, or purchasing terms behind that figure. Its English edition also cautions that translated quotations may differ from the original wording.
The figure therefore offers limited guidance about the price of a particular device.
More specific market data gives a clearer, if less dramatic, picture. In its third-quarter mobile-memory outlook, research firm TrendForce projected an 8%–13% quarter-over-quarter increase in contract prices. It described slower price growth as inventory restocking finished and demand softened, while expecting continued shifts toward server memory and HBM to support prices.
That forecast measures a defined segment over a defined period. It does not verify the basis for Tan’s broader multiple.
It also supplies an important counterweight to a simple scarcity narrative: AI-related production shifts and weakening consumer demand operate at the same time. Suppliers’ priorities influence availability, while buyers’ willingness to pay influences prices.
Tan’s warning deserves attention. Turning it into a prediction that every memory product—or every laptop—is about to become several times more expensive goes beyond the evidence.
Suppliers’ business decisions show the pressure
One supporting example comes from Micron.
In December 2025, the company announced its exit from the Crucial consumer business, a brand associated with consumer memory and storage products. Micron linked the decision to AI-driven data-center demand and its intention to improve supply for larger strategic customers in faster-growing markets.
Its plan called for Crucial consumer shipments to continue through February 2026, with warranty support continuing afterward.
The announcement concerned a particular consumer business. It did not establish that Micron stopped supplying every component ultimately used in a PC or phone.
Its significance for Intel’s warning is narrower and concrete: a major memory manufacturer explicitly changed its business priorities in response to data-center demand.
For smaller hardware makers, that raises questions about their own suppliers’ commitments. Public announcements do not reveal exactly which customers receive what allocation, but they show why total industry production is only part of the story. The product mix and customer commitments matter too.
Why building more capacity takes time
New investment offers a route to more supply, with a delay between announcing a facility and shipping usable products.
SK hynix’s Indiana project illustrates that gap. In August, the company described an investment exceeding $4 billion, with next-generation HBM mass production planned for the second half of 2029.
The facility will package and test memory using wafers produced in South Korea. Its role is therefore specific: it expands a later stage of HBM manufacturing.
That detail matters when assessing relief for ordinary device makers. More HBM packaging capacity does not directly tell us how much conventional laptop memory will become available.
The 2029 date also belongs to this project, rather than the entire industry. Other expansions and manufacturing improvements operate on their own schedules.
The useful measure is what reaches customers: additional output that meets their technical requirements, shorter delivery times, and more dependable availability for particular products.
Tan’s forecast depends on demand continuing to outpace usable supply. Stronger AI demand and slow expansion support that concern. Weaker device demand, slower AI investment, or faster production improvements would ease the pressure. The evidence explains why shortages remain a risk; it does not establish that every memory category will become harder to supply.
Intel’s broader point: AI needs a complete working system
According to Seoul Economic Daily’s account, Tan also highlighted electricity, cooling, and connectivity, and described an industry moving toward integrated systems that combine processors, networking, and software in equipment racks.
The practical implication is that AI expansion depends on several resources arriving together. Additional processors have limited value to a deployment waiting for memory or sufficient power.
This connects Intel’s warning to the broader issue of who controls the infrastructure underneath AI. Making AI software more capable does not automatically make the equipment required to run it readily available.
For a business planning an AI deployment, that means evaluating the complete system and its delivery commitments. For a business replacing employee laptops, it means checking the actual configuration, price validity, and promised arrival date.
A higher component cost has several possible destinations: the manufacturer’s margin, the device’s selling price, its specifications, or its production schedule. Contracts, existing inventory, and competition influence which outcome appears.
Intel’s warning connects the AI buildout to those everyday purchasing decisions. The same suppliers serving enormous data centers also make components used in ordinary devices, and their manufacturing choices reach both markets.
That is why someone buying a basic laptop has a stake in AI infrastructure spending. Their computer does not need to run an advanced AI model for the race to build one to affect its supply chain.
