NVIDIA has already notified some major clients that, due to the sharp rise in storage chip costs, the prices of servers equipped with NVIDIA AI chips will increase by more than 15% in most cases.

This price increase will take effect starting with systems shipped early next year, affecting servers equipped with flagship Vera Rubin and Grace Blackwell chips.

Currently, factories that manufacture servers for major companies like Microsoft, Google, and Oracle have gradually received price increase notices. The specific increase will vary depending on chip generation and storage configuration.

Server price hikes will not only directly increase the construction costs of AI data centers, but also draw more attention to NVIDIA's upcoming quarterly results announced on August 26.

| Memory shortages are the real driver behind price increases

The fundamental reason for this round of price increases is that the supply of DRAM memory chips is seriously unable to keep up with demand.

The performance of NVIDIA's AI chips heavily depends on the capacity of the DRAM they are paired with. Global DRAM production capacity is almost monopolized by three giants: Samsung, SK Hynix, and Micron.

Although these three companies have been expanding capacity, capacity growth has lagged far behind the explosive growth in AI infrastructure demand. With supply outstripping supply, the prices of these chips naturally soared.

According to reports, memory chip manufacturers have leveraged this supply-demand imbalance to gain unprecedented bargaining power.

Even industry giants like NVIDIA have failed to withstand cost pressures and have to pass price increases downstream, which precisely shows how dominant Samsung, SK Hynix, and Micron are in the current AI wave. In fact, both Apple and Qualcomm have recently stated that they have been forced to raise product prices due to chip shortages.

| No matter how high Nvidia's profits are, it can't withstand the rising memory prices

Nvidia is one of the most profitable companies in the semiconductor industry, with a gross margin as high as 75%. Thanks to TSMC's manufacturing capacity advantage, its AI chips often sell for tens of thousands of dollars. This product, originally derived from PC gaming graphics cards, saw its price soar due to sustained demand and the lack of competitors on the market. Nvidia has even recently raised the prices of PC graphics cards targeting the gaming market.

But even with such generous profits, Nvidia couldn't absorb the cost increase alone this time and had to pass the pressure on downstream.

In the face of price increases, the reactions of major clients such as Amazon, Microsoft, Google, and Meta are crucial. Although these companies are advancing their self-developed chip plans and trying to reduce dependence on Nvidia, data center construction still heavily depends on Nvidia's products. According to Bloomberg, whether these customers can truly achieve greater independence in the future also depends on whether they can secure sufficient memory supply from Samsung, SK Hynix, and Micron.

Server price increases will further intensify the difficulty of large-scale expansion of AI data centers. Project delays, labor shortages, tightening capital markets, and community resistance have already disrupted many construction plans, and now rising cost pressures are undoubtedly adding insane damage.

As the world's most valuable publicly traded company, Nvidia will release its latest financial quarterly results next week. Its performance has long been a key indicator for the tech industry and investors betting heavily on AI infrastructure.

KingTech's Perspective | Understanding the Power Restructuring of the Industry Chain Behind the '15% Price Increase'

In the past, the market generally believed that Nvidia monopolized the AI industry chain pricing power with GPUs, but this price increase reveals an underestimated fact—memory chips are becoming a "bottleneck" restricting the expansion of AI computing power.

In the Vera Rubin architecture, memory cost has surged from single digits in the previous generation to 25%-30%, with cost increases reaching 435%, exceeding a quarter of the total device cost for the first time. This means the pricing logic for AI servers is shifting from "GPU-centric" to "storage-centric," with storage configuration becoming the core variable for price increases.

Memory chips are the most direct beneficiaries of this round of price increases

Samsung, SK Hynix, and Micron together control over 90% of global DRAM capacity and nearly all HBM capacity. Against the backdrop of explosive growth in AI demand, they not only enjoy HBM premiums of up to 300%, but also prioritize allocating advanced capacity to high-margin AI storage products through the "squeeze effect," further driving up the prices of all memory chips.

For investors, the memory chip industry chain (including HBM, DRAM, memory interface chips, memory modules, etc.) is currently the most certain beneficiary.

Focus on the August 26 financial report

Nvidia's quarterly financial report, to be released on August 26, will be a key window for the market to test its supply chain control capabilities: whether gross margin can be maintained around 75%, verifying smooth cost transmission; Will next quarter's revenue guidance be revised upward due to server price increases; Management's latest assessment of the tight supply situation for memory chips.

If these indicators continue to improve, Nvidia is expected to maintain a strong pricing position during the cost upward cycle.

Price increases are forcing domestic substitution to accelerate

The soaring cost of overseas high-end computing power servers has objectively opened a replacement window for domestic computing solutions.

The cost-effectiveness advantages of domestic AI servers (such as Inspur Information, Sugon), domestic storage chips (such as Yangtze Memory, Changxin Technology), and domestic computing chips (such as Cambricon, Hygon Information) will become even more prominent. Especially in the domestic government and enterprise computing power procurement sector, the pace of domestic substitution is expected to accelerate further.