---
title: "A 6,000-yuan laptop, and what memory now eats"
date: 2026-11-18
category: en
site: NeuroAI
canonical: https://neuroai.site/a/07-na-chips-memory-cost-1118
language: en
---

# A 6,000-yuan laptop, and what memory now eats

> Daily AI/BCI industry briefing (EN edition, 2026-11-18).

In October, TrendForce ran an arithmetic exercise on a laptop priced at about 6,000 yuan: what share of the bill of materials goes to the processor, memory and SSD.

In the first quarter of 2025 it was 45%. In the third quarter of 2026 it is 68%.

## From 2,700 yuan to 4,080 yuan

On a 6,000-yuan machine, 45% is 2,700 yuan and 68% is 4,080 yuan. The same laptop now carries 1,380 yuan more in three parts.

That is why Singles' Day laptops cost more than they did in June. In a November 17 report by Caijing, Zhang Yingju, founder and CEO of Shenzhen Jihecheng, an industrial computer maker, said memory plus SSD now accounts for 70% of total machine cost in some of its products, and the company consumes about 200,000 memory modules a year.

His options are narrow: absorb the cost, or raise prices and lose orders. Customers simply stop buying.

Another company chose a different way to absorb it. AI infrastructure provider Xinzhi Huixiang locked in a batch of memory chips at the end of 2025 and delivered several large projects on schedule — at a cost of more than 30% of profit.

## The size of the gap

Institutions estimate the server DRAM supply gap at more than 10%, and the HBM gap at 20% to 30%.

On price, the average contract price for DDR 8Gb reached $27.5 in October 2026, up 5.77% from $26 in September.

On its October 28 earnings call, SK Hynix said that counting signed long-term agreements and orders under negotiation, its DRAM and HBM supply through 2029 is effectively sold out.

This is not an ordinary cycle. The real change is allocation: advanced capacity goes first to HBM and server DDR5, and the consumer slice is squeezed.

## The increase reaches compute rental

Cloud providers are responding in two ways.

UCloud chairman and CEO Ji Xinhua says costs are rising but demand is rising too, and customers are beginning to accept higher prices; the company is using software to raise hardware utilisation while seeking domestic substitutes, and already works with CXMT.

Parallel Tech took the other route: no price increase, absorbing costs through software optimisation. Its short-term compute rental orders — six months to a year — have grown noticeably, from customers who had planned to buy hardware and found their budgets short.

There was an earlier signal. In January 2026, AWS raised prices for EC2 Capacity Blocks for ML by about 15%, the first time in more than two decades it had lifted list pricing on a core compute service.

## The end of the chain pays

Luotu Technology's data shows the cost. In the third quarter of 2026, China's online notebook channel sold 3.12 million units, down 14.0% year on year; revenue was 24.1 billion yuan, down 2.6%; the average selling price was 7,759 yuan, up 13.4%.

Volume fell, price rose. AI PCs reached 48.7% of those sales, as vendors moved resources to more expensive configurations.

Phones are tighter. Hardware gross margins at major Chinese brands sit mostly between 10% and 20%, with net margins often below 10%. Counterpoint senior analyst Lin Keyu told Caijing that entry-level models at most vendors carry almost no hardware profit.

Institutions expect double-digit contract price increases again in the fourth quarter, led by enterprise SSDs. The rise has not peaked.

The closer a company sits to the end customer, the less pricing power it has.

## Where the cost lands

Industrial PC makers sit in an awkward place. Their customers are factories and integrators with fixed project budgets, and their products are not consumer goods that can be refreshed every year. When memory and storage take 70% of machine cost, a price increase is not a marketing decision but a survival one.

Cloud providers have more room. They can pass cost into rental rates, and they can shift the mix toward longer contracts. The growth in short-term rental orders at Parallel Tech is the market's version of a hedge: customers who cannot buy hardware rent it instead.

The macro read is simple. AI demand is pulling advanced memory capacity away from consumer devices, and the consumer market has no countervailing power. Until new capacity comes online — and SK Hynix suggests that is years away — the direction of prices is set elsewhere.

One consequence is rarely stated: when components dominate cost, design cycles shorten. Vendors re-spec machines around whatever memory is available, and the model number on the box matters less than the bill of materials behind it.

*Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.*

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/07-na-chips-memory-cost-1118
Free to quote with attribution and a link to the original.
