Intel chief executive Lip-Bu Tan told an audience at the AI Infrastructure Summit 2026 in Santa Clara, California, on September 15, 2026, that the memory chip shortage squeezing the computer industry is not easing. It is getting worse. Tan said memory prices have already risen “five, six, and seven times” over prior levels, and warned that low-cost smartphones and laptops now face a reality where 70-80% of the device’s cost has to be spent on memory alone.
The remarks, delivered at the Santa Clara Convention Center, land at a moment when AI data center buildouts are absorbing memory supply that used to go to phones, laptops, and game consoles. Tan’s message was blunt: production capacity is limited, and that scarcity is delaying business projects across the industry. He has also said, in separate comments reported earlier this year, that there is “no relief until 2028.” For engineers, procurement teams, and consumers trying to plan around device prices, that is a specific and uncomfortable timeline.
This piece breaks down what Intel’s chief executive actually said, what is driving the shortage, how the numbers compare across the wider hardware supply chain, and what it means for anyone buying a phone, a laptop, or a GPU between now and 2028.
What Intel’s Lip-Bu Tan Said About the Memory Shortage
Tan’s keynote at the AI Infrastructure Summit 2026 focused on the bottlenecks holding back AI infrastructure expansion, and memory topped the list. He said he had expected memory to become a serious constraint as early as last year, and that the problem has now arrived in force. According to the Seoul Economic Daily, Tan told the Santa Clara audience: “The production capacity is very limited, so many businesses are being delayed because they do not have enough memory.” That is a direct statement from Intel’s CEO, not an analyst projection, and it puts a specific voice behind numbers that chipmakers have been reporting for months.
Tan also connected the shortage to the sharpest end of the pricing problem: entry-level devices. He said the situation where 70-80% of a low-cost smartphone or laptop’s bill of materials goes to memory alone is difficult for device makers to absorb. That statement, reported by the Seoul Economic Daily, reframes the shortage from an abstract supply-chain story into something that shows up directly on a store shelf.
Why AI Demand Is Draining the Memory Supply
The root cause is straightforward: high-bandwidth memory, or HBM, is the memory type AI accelerators need, and it competes for the same fabrication capacity as ordinary DRAM. According to the Seoul Economic Daily’s September 16 report, the shortage stems from HBM demand surpassing production capacity at Samsung Electronics, SK hynix, and Micron. As those three companies reorganize fabs around HBM output, general-purpose memory supply for phones, laptops, and PCs shrinks by default. Nobody built a new factory to make consumer DRAM scarcer. It happened as a side effect of chasing AI demand.
This is why Tan’s warning is about more than Intel’s own supply chain. When a fab line gets converted to produce HBM3E or HBM4 stacks for Nvidia or AMD accelerators, that capacity stops making the DDR5 modules that go into a budget Chromebook. Industry estimates suggest HBM output growth in 2026 has leaned heavily on converting existing DDR4 and DDR5 lines rather than building new ones, which speeds up AI memory supply but stretches out the wait for mainstream DRAM and NAND prices to normalize. Tan’s own framing lines up with that: memory is not scarce because nobody is investing, it is scarce because the investment is going somewhere specific.
Memory Prices by the Numbers: 2026 Snapshot
Tan’s “five to seven times” figure describes the scale of the increase rather than a single spot price, and it tracks with what memory market trackers have been recording through 2026. TrendForce’s spot-price data placed mainstream DDR5 16Gb modules in the low-to-mid $50 range per chip as of September 15, 2026, a level that reflects a market that has firmed up considerably compared to the oversupply years earlier in the decade. On the AI side, HBM stack pricing has also moved, with premium HBM3E capacity commanding higher per-unit costs than standard DRAM by a wide margin, according to memory pricing trackers.
| Memory Segment | 2026 Status | Primary Driver | Who Feels It |
|---|---|---|---|
| HBM (AI accelerator memory) | Tight supply, premium pricing | Nvidia/AMD accelerator demand | AI cloud providers, GPU buyers |
| DDR5 commodity DRAM | Firming spot prices, TrendForce-tracked | Fab capacity shifted to HBM | PC and server OEMs |
| Mobile LPDDR / NAND | Rising cost share of device BOM | Reduced general-purpose output | Budget smartphone buyers |
| Overall price trend (Tan’s figure) | Up “5x-7x” versus prior levels | AI-driven capacity reallocation | Entire device supply chain |
| Expected relief window | “No relief until 2028” (Tan) | Multi-year fab conversion cycle | Anyone buying memory-heavy devices through 2028 |
Samsung, SK Hynix and Micron: Where the Capital Is Going
Tan did not name specific rivals in his keynote, but the shortage he described maps directly onto capital spending decisions at the three companies that make almost all the world’s advanced memory. SK hynix has pushed a large share of its 2026 investment into HBM4 capacity expansion, with industry estimates putting its annual capital expenditure in the neighborhood of $20 billion, a double-digit increase year over year. Samsung is running a comparable playbook, directing new capacity toward next-generation HBM at its advanced process lines, while Micron has been described by analysts as the most aggressive of the three in percentage terms, with 2026 spending increases reported in the low double digits as it races to catch up on HBM share.
None of that capital is aimed at making a $200 laptop cheaper. It is aimed at winning AI accelerator contracts, which pay far better per wafer than commodity DRAM ever did. That is the uncomfortable math behind Tan’s remarks: the memory industry is investing more money than it has in years, and the shortage is still getting worse for ordinary consumer devices, because the money is chasing a different customer.
| Memory Maker | 2026 Capex Direction | Primary Focus | Consumer DRAM Impact |
|---|---|---|---|
| Samsung Electronics | Increased year over year | Advanced-process HBM expansion | Secondary priority |
| SK hynix | Estimated near $20B, up double digits | HBM4 capacity build-out | Secondary priority |
| Micron | Aggressive YoY increase, analysts say | Catching up on HBM share | Secondary priority |
| Net effect on commodity DRAM/NAND | Growth constrained | Fabs reallocated to HBM | Prices stay elevated into 2028, per Tan |
The 70-80% Problem: Budget Phones and Laptops Take the Hit First
The part of Tan’s speech most likely to affect everyday buyers is the bill-of-materials point. In a flagship phone or a high-end laptop, memory is one line item among many, alongside the processor, display, camera array, and battery. In a $200-$400 device, there is far less room to absorb a cost spike, because there are fewer expensive components to offset it against. When Tan says memory can eat 70-80% of that cost, he is describing a device category where the phone or laptop is, in effect, becoming a memory module with a screen attached.
Manufacturers facing that math have three levers, and none of them are painless: raise the retail price, cut RAM or storage specs to hold the price line, or delay the launch and hope supply loosens. Console and PC hardware makers have already been through versions of this trade-off earlier in 2026, and the pattern Tan described suggests budget smartphone and Chromebook makers are next in line. This is also why the shortage keeps showing up in headlines about gaming hardware, not just enterprise AI servers. Rising component costs do not stay confined to one product category once fab capacity gets reallocated.
How This Compares to Other Chipmakers’ Warnings
Tan is not the only chip executive flagging memory as the constraint that matters most in 2026. Nvidia has already raised prices on parts of its AI server lineup this year, citing the same memory crunch that Tan described, a move covered in Nvidia’s own price hikes on AI server hardware. On the consumer GPU side, rising memory costs have been cited as a factor behind the run-up in high-end graphics card pricing, a trend tracked in coverage of RTX 5090 pricing as AI demand bites. What makes Tan’s comments stand out is the specificity: a named CEO, a named multiplier (“five, six, and seven times”), and a named timeline (“no relief until 2028”), delivered in a public keynote rather than buried in a supplier earnings call.
That specificity matters for how the market reads the story. Vague warnings about “supply constraints” get discounted by analysts who assume some cyclical exaggeration. A CEO putting a hard multiplier and a hard year on the shortage is a different kind of signal, and it is one reason the comments spread quickly through Korean and international business media within a day of the keynote.
Lip-Bu Tan, In His Own Words: Five Statements on the Shortage
Tan has been consistent on this subject across multiple public appearances in 2026, not just the September keynote. Below are five statements attributed to him, each drawn from named reporting.
“I think in terms of the AI, the biggest challenge, I think, for a lot of my customer is memory,” Tan said, according to Livemint’s report on his remarks at Intel’s annual summit.
He put it more bluntly in the same appearance: AI is “sucking” up a lot of the memory that the computer industry has, per the same Livemint account of his comments on AI-driven memory pressure.
On timing, Tan said: “Memory, actually, there’s no relief as far as I know, when I talk to only 3 key players, 2 of them I talked to very frequently. And then they told me that Lip-Bu, there’s no relief until 2028,” a statement Livemint reported from his remarks on the expected duration of the shortage.
At a separate conference, Tan said: “As far as I know, there is no sign of memory shortages easing,” according to the Seoul Economic Daily’s report from February 2026.
He also described the source of his 2028 estimate directly: “After speaking with major players in the memory sector recently, I was told the shortages would continue until 2028,” per the same Seoul Economic Daily coverage. Taken together, the five statements show a CEO who has been tracking this problem for the better part of a year and whose outlook has not improved.
Market and Business Impact
For Intel specifically, Tan’s framing carries a strategic angle beyond the warning itself. Intel has been positioning memory-CPU stacking and closer collaboration with Nvidia and AMD as part of its response to the shortage, according to the Seoul Economic Daily’s September 16 report, which also noted Tan flagged power consumption and cooling, including air, liquid, and microfluidic systems, as parallel infrastructure bottlenecks tied to the same AI buildout. That framing lets Intel position itself as a company solving the shortage rather than merely a victim of it, even though its own PC chip business depends on the same memory supply chain it is warning about.
For memory makers, the picture is more directly favorable in the near term. Higher prices and locked-in demand from AI customers mean Samsung, SK hynix, and Micron are seeing stronger margins on the capacity they can allocate to HBM, even as they take reputational heat for constrained consumer supply. For device OEMs building phones, laptops, and consoles, the picture is worse: they are absorbing higher input costs with fewer places to pass them on before hitting consumer price resistance. Recent reporting on memory stockpiles falling to critically low levels underscores how thin the buffer has become across the supply chain, and Samsung’s own HBM4E production ramp for Nvidia shows exactly where the company’s engineering priority sits.
Historical Context: This Is Not the First Memory Squeeze
From Oversupply to Scarcity in Two Years
Memory has always been a boom-and-bust business, cycling between oversupply that crushes prices and scarcity that spikes them. What is different this time, according to the pattern Tan and other executives have described through 2026, is the cause. Past cycles were driven by PC and phone demand swings that manufacturers could eventually plan around. This one is driven by AI infrastructure spending that shows no sign of plateauing, which is why RAM prices climbing sharply earlier this year did not settle back down the way past shortages typically did within a few quarters.
Why 2028 Keeps Coming Up
The 2028 figure Tan has repeated across multiple appearances is not arbitrary. It reflects the multi-year lead time required to bring new fab capacity online and qualify it for HBM production, on top of the existing backlog of AI accelerator orders that memory makers have already committed capacity against. Executives across the sector have converged on a similar window, which is part of why Tan’s estimate has been treated as credible rather than dismissed as CEO caution-talk.
What Comes Next: Five Predictions Through 2028
- Expect more device makers, not just chipmakers, to start naming memory costs directly in earnings calls and price-hike announcements, following the pattern Tan set with a public, quotable multiplier.
- Budget smartphone and laptop segments will see either higher prices or trimmed RAM/storage configurations before the high end of the market feels comparable pressure, since low-end devices have the least room to absorb the BOM shift Tan described.
- Samsung, SK hynix, and Micron will keep prioritizing HBM4 and next-generation capacity over commodity DRAM expansion through at least 2027, meaning consumer memory relief is unlikely to arrive faster than AI accelerator demand cools.
- Intel’s push toward memory-CPU stacking and closer Nvidia and AMD cooperation, flagged in Tan’s keynote, will be positioned as a partial technical hedge against the shortage rather than a fix for the underlying capacity gap.
- Watch for Tan or other chip executives to repeat the “2028” timeline at further conferences this fall. A shift in that number, in either direction, would be a meaningful signal of whether fab conversions are ahead of or behind schedule.
What This Means for Buyers and Businesses Right Now
If you are shopping for a laptop, phone, or prebuilt PC in the near term, Tan’s comments are a reasonable signal to expect current pricing to be closer to a floor than a ceiling, particularly at the budget end of the market. Businesses planning hardware refreshes or data center expansion should treat memory lead times as a planning constraint on par with GPU availability, not an afterthought. Procurement teams that assumed memory pricing would normalize by 2027 now have a specific, on-the-record reason to revisit that assumption, since the executive closest to the problem is putting the relief date a full year later than that.
Frequently Asked Questions
What did Intel CEO Lip-Bu Tan say about the memory shortage?
At the AI Infrastructure Summit 2026 in Santa Clara on September 15, 2026, Tan said memory prices have risen “five, six, and seven times” and that limited production capacity is delaying businesses that cannot secure enough memory. He also said low-cost smartphones and laptops face a situation where 70-80% of the device cost goes to memory.
Why is AI causing a memory chip shortage?
AI accelerators require high-bandwidth memory (HBM), which competes for the same fab capacity used to make standard DRAM. As Samsung, SK hynix, and Micron shift production toward HBM for AI customers, less capacity remains for the memory used in phones, laptops, and PCs.
When will the memory shortage end?
Tan has said repeatedly, including in comments reported by the Seoul Economic Daily, that there is “no relief until 2028.” That estimate reflects the multi-year timeline needed to expand fab capacity and work through existing AI accelerator order backlogs.
Which companies make the memory affected by this shortage?
Samsung Electronics, SK hynix, and Micron produce the vast majority of the world’s advanced DRAM and HBM. All three have been directing 2026 capital spending toward HBM capacity for AI accelerators rather than commodity DRAM expansion.
How does the memory shortage affect phone and laptop prices?
Because memory is now a much larger share of a device’s total cost, particularly for budget models, manufacturers are more likely to raise prices, reduce RAM or storage specifications, or delay product launches rather than absorb the cost increase.
Is HBM the same thing as regular computer memory?
No. HBM (high-bandwidth memory) is a specialized, stacked memory format built for AI accelerators and high-performance computing. Standard DDR5 or LPDDR memory used in laptops and phones is a different product, but both draw on overlapping fabrication capacity at the same manufacturers.
Has Intel raised its own product prices because of the memory shortage?
Tan’s keynote focused on the industry-wide shortage and Intel’s technical response, including memory-CPU stacking and closer cooperation with Nvidia and AMD, rather than announcing specific Intel product price changes. Other chipmakers, including Nvidia, have already raised prices on parts of their AI server lineups citing memory costs.
What should businesses do about the memory shortage before 2028?
Treat memory availability and pricing as a planning constraint for hardware refreshes and data center buildouts, not a temporary blip. Locking in supply agreements early and budgeting for sustained higher memory costs through at least 2027 is consistent with the timeline Tan has laid out.
