Why GPU Prices Jumped Thousands Overnight: AI Demand and Memory Shortage

Price Surge Data (Early August 2024)

Shanghai Securities Daily and Wccftech reported the following retail and wholesale price changes:

RTX 5060: ¥2,300‑2,500 → ¥2,700‑3,100 (↑¥800 in four days, >30%).

RTX 5080: >¥9,000 → >¥11,000 (↑¥1,700).

RTX 5090: ↑¥4,000 within one week.

RTX 5070 Ti wholesale price: ¥8,200 on July 31 → ¥8,400 on August 1 (↑¥200).

RTX 5070 Ti listed on some e‑commerce platforms at ¥10,700, a ~50% increase year‑over‑year.

Manufacturer Price Adjustments

According to the Taiwan Economic Daily, NVIDIA sent a price‑increase notice to AIC (add‑in‑card) manufacturers for GPU‑plus‑memory kits, raising bundled prices by 20‑30% and covering both the latest GDDR7 and previous‑generation GDDR6 memory.

Wccftech cited Samsung’s quarterly memory‑price rise of roughly 20%, which NVIDIA mirrored. AMD later announced a minimum 10% increase for GPU‑plus‑memory bundles starting in August, after postponing a planned July hike.

Root Cause: AI‑Driven Memory Allocation

TrendForce reported that memory makers such as Samsung and Hynix prioritize HBM production for AI accelerators because HBM yields the highest profit margin among DRAM segments, followed by DDR5, with consumer‑grade GDDR at the bottom.

Consequently, GDDR6 and GDDR7 supply is constrained, driving up chip prices: 2 GB GDDR7 chips sell for $20, while 3 GB chips cost $60‑70 (a three‑fold increase for only 50% more capacity).

Impact on AIC Manufacturers and Channels

AIC manufacturers receive GPU‑plus‑memory kits, which constitute the majority of a graphics card’s cost. With only NVIDIA and AMD supplying GPUs and Samsung/Hynix controlling memory, AICs have virtually no bargaining power, so kit price hikes directly raise card costs.

Channel reports from Huaqiangbei indicate that many AIC factories have locked inventory and halted bulk shipments. Dealers note that stock exists but upstream suppliers are unwilling to sell at previous prices, leading to daily price adjustments and a reluctance to hoard inventory.

Effect on DIY Builds and Consumer Choices

DIY system costs have risen sharply: RTX 5060 now costs ~¥3,000 and RTX 5080 exceeds ¥10,000, placing mid‑range cards at the price level of last year’s high‑end models. This reduces the performance‑per‑budget ratio and disrupts upgrade cycles.

As a result, price‑stable gaming consoles (PS 5, Xbox Series X, Switch 2) become comparatively attractive, though console prices have also seen modest increases.

Implications for Local AI Deployment

Running a 7‑billion‑parameter quantized model requires at least 6‑8 GB VRAM; fine‑tuning or larger models need ≥24 GB. Within the RTX 50 series, only the RTX 5070 Ti (16 GB) and RTX 5090 (32 GB) meet these memory requirements, and they experienced the steepest price hikes.

Given the high hardware cost, many users find cloud AI APIs more cost‑effective, paying a few hundred yuan per month for top‑tier model access instead of purchasing a single high‑end GPU.

Broader Market Context

The GPU price surge is not solely due to NVIDIA’s pricing decisions; the broader AI boom has commandeered memory capacity, forcing higher prices across the consumer‑electronics supply chain. As long as AI compute demand remains strong, elevated GPU pricing is likely to persist.

Code example

1、
Markdown：被误解了二十二年的老实人
2、
我把网站塞进图标后，慌了
3
、
爷青回！无需安装，浏览器一键“复活”Windows
4、
因AI无法被拖欠工资，很多公司开始重新招聘人类程序员。
5、
老黄刀法被破解：10GB老矿卡“开核”80GB，身价暴涨20倍