Omdia: AI PC shipment share to reach 60% by 2027, but only 9% of buyers purchase for AI features
Omdia predicts that the share of AI PC shipments will reach 60% by 2027, but only 9% of buyers actually place orders for AI features; most users' motivation for upgrading remains that their old machines have slowed down or broken down. What is driving the penetration rate up is mainly the natural replacement cycle and the cost pressure brought by cloud AI token-based billing. 78% of office users use AI multiple times a week, yet only 12% access it through company accounts. Memory costs as a share of the entire device bill of materials have risen from 15% to nearly 40%, becoming a hardware bottleneck for running large models locally.
Body · AI translation
On October 30, a forecast report recently released by market research firm Omdia revealed the false prosperity of the AI PC market: on one hand, industry penetration is advancing rapidly; on the other, consumer response has been lukewarm.
Data shows that although AI PCs are expected to account for 60% of shipments by 2027, only 9% of buyers actually place orders for AI features; most users' upgrade motivation is simply that their old machine has slowed down, malfunctioned, or they want a better model.
Since users are not coming for AI, why will penetration still climb rapidly? The answer lies in the replacement cycle itself. Whether or not users actively choose AI, each natural replacement cycle pushes the installed base of local AI capabilities one step further, and this passive adoption is precisely the main driver of AI PC expansion.
A deeper driving force comes from the reverse pressure of cloud AI. After mainstream cloud vendors shifted to token-based billing, task costs often become clear only after execution ends, and AI agents trigger multiple calls in long chains, making cloud AI spending increasingly difficult to predict and manage.
At the same time, pure cloud AI also brings problems such as compliance, data sovereignty, and IT blind spots. Data shows that 78% of office users use AI multiple times a week, but only 12% access it through company accounts, meaning a large amount of AI usage lies outside enterprise IT control.
Moreover, progress on the technology side has also created conditions for localization: the gap between open-source model weights and frontier closed-source models is rapidly narrowing, quantization technology allows models to be compressed to run in PC memory, and NPUs are gradually becoming standard components of the new generation of PCs. Together, these factors make it possible for more AI workloads to be completed locally.
However, the truly fatal hardware bottleneck for AI PC adoption is still memory prices. Since 2025, because running large models locally places extremely high demands on memory capacity and speed, memory costs as a share of the whole-machine bill of materials have soared from the original 15% to nearly 40%.
This means vendors must strike a difficult balance between configuration and selling price; otherwise, high premiums will directly deter ordinary buyers. Vendors need to prove that the cloud subscription fees saved through local AI are enough to offset the premium brought by higher-end hardware configurations.
If this logic cannot be closed in the business model, AI PCs may, for a considerable period, remain in the awkward position of having a big reputation but thin profits.