Intel CEO Lip-Bu Tan warned on September 15, 2026, that the memory shortage would become more serious in 2027. Two days later, the consequences are no longer confined to AI data centers: smaller phone and laptop makers are redesigning hardware, placing orders earlier and testing incoming chips because getting memory has become a supply problem—not merely a pricing problem.
The squeeze starts with high-bandwidth memory, or HBM. This specialized form of DRAM sits alongside AI accelerators and high-performance computing systems. Its production competes for manufacturing resources with conventional DRAM used in PCs, smartphones, networking equipment and other electronics.
Intel warns the memory shortage will worsen in 2027
Tan said production capacity was very limited and that businesses were being delayed because they could not secure enough memory. He also said memory prices had risen five, six or seven times, but did not identify a specific memory category or baseline for that comparison.
That warning follows a broader rise in memory costs. One industry estimate puts DRAM prices more than 400% higher between the beginning of 2024 and the end of 2026. The estimate covers DRAM over a defined period; it is not the same measurement as Tan’s category-unspecified price statement.
The pressure is already visible in downstream forecasts. A 2026 forecast projects 1.08 billion global smartphone shipments, a 13.9% year-over-year decline. That projection attributes the expected contraction to the memory crisis and geopolitical shocks together, so it does not isolate AI demand’s individual contribution.
Why AI servers are competing for conventional memory supply
AI accelerators need HBM because it can move data at very high speed between memory and the processor. The catch is that HBM is more resource-intensive to produce than standard DDR5. A Micron earnings-call statement cited in industry analysis described roughly a 3-to-1 capacity trade-off between HBM and DDR5: supplying more HBM can consume manufacturing resources that might otherwise support conventional memory.
The market is also split between reserved and spot supply. Large technology customers can use long-term agreements to secure allocation, while smaller manufacturers are more exposed to volatile spot-market prices and shorter quote windows. Paying more can improve access to some supply, but it cannot instantly create cleanroom space, specialized equipment, trained workers or completed factories. Building and outfitting new semiconductor capacity takes years.
That is why the shortage can affect an ordinary laptop even when the original demand comes from an AI server. Manufacturers are competing inside the same constrained memory ecosystem, but they do not all have the same purchasing power or ability to reserve production in advance.
Smaller device makers are redesigning around availability
The response from smaller manufacturers is practical and, frankly, a little unusual for a component that most buyers never think about until a product is delayed.
| Maker | Product or scope | Response to constrained memory | Reader consequence |
| Jolla | Phones | Uses two motherboard designs for different memory packages and tests samples from every incoming batch for refurbished or counterfeit chips | More flexibility when a particular memory package is difficult to source |
| Framework | Modular laptops | Places non-cancellable memory orders in advance and supports salvaged or second-hand memory through its modular design | More procurement risk up front, but more options for replacing or reusing memory |
| Fairphone | Phones around $400 | Reports that memory can account for almost 60% of the bill of materials | Less room for other components or pricing decisions in lower-cost phones |
Jolla’s dual-board strategy is a direct hedge against component availability. Instead of locking every product to one memory package, the company is preparing for more than one supply path. Its batch testing also addresses the risk of receiving refurbished or counterfeit chips when legitimate supply is tight.
Framework is taking the opposite side of the same problem: secure memory earlier, then preserve flexibility after the sale. Its modular laptops allow customers to install salvaged or second-hand memory, which can matter when new components are expensive or difficult to obtain.
Fairphone’s cost figure shows why budget and midrange phones may feel the squeeze particularly sharply. If memory approaches 60% of the materials cost for phones priced around $400, even a component increase can put pressure on the rest of the bill of materials.
What the shortage could mean for phones and laptops
For consumers, the likely effects are straightforward: higher system prices, fewer configuration choices and longer lead times. A manufacturer may keep a product on sale by offering fewer memory configurations, changing the board design or securing parts well before production. Those choices can preserve availability, but they reduce the freedom to tailor a device to a particular budget or workload.
The smartphone forecast adds a broader market consequence. Shipments are projected to fall from 1.255 billion units in 2025 to 1.08 billion in 2026, before rising to 1.108 billion in 2027 and 1.16 billion in 2028. Those figures are forecasts, not a measure of how much of the decline comes from memory alone.
The duration is similarly unsettled in industry projections. One forecast discussed shortages through 2027 and into 2028, while a separate projection linked to SK Hynix’s outlook extends the crunch to 2030. Together, they point to a multi-year supply problem rather than a short-lived interruption, but they do not provide a single end date.
What Intel says it is doing
Intel is pursuing advanced packaging that combines CPUs, AI accelerators and HBM in larger packages. Tan also described cooperation with Nvidia involving Intel CPUs, Nvidia GPUs and NVLink, while Intel continues preparing its own manufacturing roadmap.
The strategy reflects the same constraint driving the shortage: AI systems need more computing power and more memory bandwidth in tightly integrated packages. As those systems scale, the competition for advanced memory capacity reaches further into the devices people use every day.
