AI Makes Intelligence Cheaper, but the Hardware Carrying It More Expensive
The demand for HBM and DRAM from AI servers has squeezed consumer memory production capacity, driving the share of memory and SSDs in PC bill-of-materials costs from approximately 15% to nearly 40%, while global PC shipments fell more than 20% year-over-year. Meanwhile, the LLM price index is dropping at a pace far exceeding historical PC price declines. As intelligence becomes cheaper, the hardware required to access it grows more expensive—a structural contradiction that is reshaping survival strategies in the PC industry.
There is a contradiction in the AI industry that few address directly: the price index for Large Language Models (LLMs) is plummeting rapidly, while the hardware required to run these models is rising in cost. The well-known venture capital firm Andreessen Horowitz (a16z) once published a chart citing data from Goldman Sachs, showing that the LLM price index completed in about three years what took the personal computer price index roughly fifteen years. Intelligent services are increasingly resembling cheap digital fast-moving consumer goods, yet the memory, chips, and computing devices that carry this intelligence remain constrained by production capacity and supply-demand dynamics.
The transmission mechanism of this contradiction is not complex. The growing demand for High Bandwidth Memory (HBM) and server DRAM from AI servers has led storage manufacturers to allocate more production capacity to higher-margin data center products, squeezing the room for growth in consumer-grade memory capacity. Estimates from market research firm Omdia show that the share of memory and SSDs in PC bill-of-materials costs has risen from a typical 15% to nearly 40%. TrendForce calculated using a mainstream laptop with a suggested retail price of $900 as an example: in Q3 2026, CPUs, DRAM, and SSDs already accounted for approximately 68% of material costs.
The result of this price transmission to the end user is a significant contraction in PC shipments. Preliminary data from IDC shows that global PC shipments in Q3 2026 fell 20.1% year-over-year, shrinking from 78.5 million units in the same period last year to 62.7 million units. Omdia’s statistics point to the same trend: global desktop, notebook, and workstation shipments fell 21.2% year-over-year. Even Macs, touted as the "best AI PCs," were not immune to this chill, with shipments also declining year-over-year during the same period.
However, attributing the shipment slump entirely to memory price hikes is unfair. The replacement demand triggered by the end of support for Windows 10 was concentrated in 2025, raising the baseline for comparison; additionally, some brands placed advance orders in the first half of the year to stock up before further memory price increases, disrupting the rhythm for the second half. The combination of multiple factors meant that Q3, traditionally driven by the back-to-school season, did not see its usual peak. Jitesh Ubrani, an analyst at IDC, believes that inventory pressure may force channel promotions, but prices returning to levels from a year ago seem unlikely—once the bow is drawn, there is no turning back.
Facing cost pressures, PC manufacturers, after multiple price hikes failed to fill the cost gap, began shifting toward extending the service life of mature platforms. On October 9, 2026, Gigabyte announced that all its B760 and H610 motherboards would receive BIOS updates to support new Intel LGA1700 processors expected to launch in early 2027, including notably DDR4 motherboards. The LGA1700 platform debuted in 2021 with the 12th Gen Core processors, originally supporting both DDR4 and DDR5 motherboard types; existing B760 and H610 users can now upgrade their processors without replacing memory after updating their BIOS. On the AMD side, Gigabyte announced a new batch of AM4 motherboards in August this year, paired with the reintroduction of the Ryzen 7 5800X3D, adding options for assembly solutions sticking with DDR4.
In the past, new platforms often bundled processor, motherboard, and memory upgrades together, forcing users to replace everything. Now, how much existing hardware a product can help users keep has become part of its competitiveness. A reporter from The Verge recently found that a 32GB Corsair Vengeance DDR5 kit sold for $620, while the corresponding DDR4 kit was only $260. A price difference of $360 is enough to sway an assembly decision.
Software is also attempting to bypass hardware bottlenecks. An open-source project named Strata has gained popularity on GitHub, accumulating approximately 19,500 stars, with the goal of enabling a standard gaming PC to drive the Qwen3.8-Flash-Next model with 125 billion parameters. According to self-test data released by the Strata project, a computer equipped with an RTX 5070 graphics card, 12GB VRAM, a Ryzen 5 7600 processor, and 64GB system memory can achieve a response generation speed of 94 tokens per second under the Q2_0 quantization configuration.
The principle lies in exploiting the sparsity of the MoE (Mixture of Experts) architecture: not all experts need to be active for every token generated. Frequently used experts are cached in VRAM, while weights for all experts are stored in system RAM. The CPU processes uncached experts in parallel, and large lookup tables are stored on the SSD. Quantization further compresses weight volume, while speculative decoding allows auxiliary modules to predict subsequent tokens in advance, which are then batch-verified by the main model. According to developer notes, this can bring a 1.6x to 1.8x acceleration.
But these solutions come with clear trade-offs. The 94 tokens per second corresponds to the Q2_0 low-bit quantization version; on the same hardware, the milder IQ3_S version generates at 53 tokens per second—trade-offs between speed, memory usage, and model quality remain necessary. Speculative decoding accelerates output but cannot eliminate precision loss caused by low-bit quantization. Upon first launch, the project requires moving tens of gigabytes of data into memory, potentially freezing the computer for 1 to 3 minutes. A 32GB memory configuration typically forces the selection of a streamlined Coder version, which has weaker capabilities in non-programming tasks such as Chinese.
Strata demonstrates a possibility: the same hardware, through more rational allocation of computing and storage tasks, can run large models that were previously difficult to accommodate. How much a computer can do depends not only on the hardware purchased but also on how effectively software utilizes that hardware. However, it still has thresholds for daily usability; the trade-offs in startup wait times, memory requirements, and answer quality mean that deploying local large models on consumer hardware remains an engineering exercise requiring compromise.
Returning to that structural contradiction: the tech industry is simultaneously putting stronger AI into ordinary people's hands and making ordinary people split every penny for a single stick of memory. We may be entering an era where intelligence becomes increasingly affordable, yet we are simultaneously experiencing a hardware cost crisis pushed up by the very demand for intelligence. Software optimization and extending platform lifecycles are paths for consumers paying for this AI boom to feel less heartburn—but only if these paths are truly established as standards.