The technological war between the United States and China has entered a particularly uncomfortable phase for the components ecosystem: the stage where restrictions are no longer just about “allow or ban,” but have become a mechanism for revenue collection and industrial pressure. In mid-January 2026, Washington formalized a 25% tariff on certain advanced computing chips—including NVIDIA H200 and AMD Instinct MI325X—when the final destination is the Chinese market. Beijing responded almost simultaneously with a directive for domestic companies to stop using U.S. and Israeli cybersecurity software, citing national security risks.
The U.S. move presents a double-edged sword: on one side, it maintains the narrative of national security and reducing reliance on foreign manufacturing; on the other, turns what was previously a blockade into “income”, ensuring that the state captures a share of the value from AI accelerators sold to Chinese buyers. The interpretation within the data center components channel is clear: policy is shifting from binary to transactional.
A tariff designed not to hinder domestic AI development
According to official information and leaks gathered by economic media, the tariff is based on the Section 232 of the Trade Expansion Act of 1962, a legal framework that allows imposing trade measures for national security reasons. The White House explained that the goal is to regulate imports of semiconductors, equipment, and related products, with a focus on advanced computing chips. The unique aspect is that the scheme aims to avoid harming domestic AI deployment: the tariff does not apply—or has exemptions—when chips are destined to strengthen the supply chain and manufacturing capacity within the U.S., or when used in domestic infrastructure such as data centers.
In other words: Washington seeks to sustain AI development at home without completely closing the door to major chip designers still selling to one of the world’s largest markets. For the industry, this “selective exemptions” architecture adds regulatory complexity but also sends a message: priority is domestic deployment, with the rest managed through licensing, limits, and tariffs.
Which chips are at the center of the clash
Although the public narrative talks about “GPUs,” the real debate involves AI accelerators aimed at training and inference in data centers: parts purchased in thousands, mounted on trays of 4 or 8 units, justified by cost per token or training hour.
To understand the product scope, here are the two references mentioned in announcements and analyses:
Table 1 — Quick comparison of affected accelerators
| Accelerator | Segment | Memory | Bandwidth | Why it matters in AI |
|---|---|---|---|---|
| NVIDIA H200 (Hopper) | Data center (SXM/PCIe) | 141 GB HBM3e | 4.8 TB/s | More memory and bandwidth for large models and intensive workloads |
| AMD Instinct MI325X (CDNA 3) | Data center (OAM) | 256 GB HBM3E | 6 TB/s | High capacity per GPU to consolidate models and reduce “sharding” |
Within the components sector, the key nuance is that the bottleneck isn’t just the silicon: it’s the HBM memory, advanced packaging, assembly capacity, and verification and compliance logistics. Geopolitics thus infiltrates inventory planning and delivery forecasts.
China, caught between dependence and pressure to replace technology
China’s response has not yet been articulated as a mirror tariff on chips but as a direct blow to dependence on foreign software in sensitive sectors. According to sources cited by Reuters and expanded upon by tech media, Chinese authorities have instructed domestic companies to stop using cybersecurity software from around a dozen American and Israeli firms. Among the affected are VMware (Broadcom), as well as cybersecurity firms like Palo Alto Networks, Fortinet, and Check Point.
The argument is familiar: risk of data leakage to governments or third countries. However, the background is equally strategic: Beijing has been promoting technology substitution programs for years, and this measure is linked to the Xinchuang initiative, which aims to replace foreign software with domestic alternatives in government agencies and state-owned enterprises, targeting a timeline around 2027 in some sectoral roadmaps.
For Western suppliers, the impact is not just commercial but operational: support and renewals, partner networks, and managed services. For Chinese buyers, the pressure shifts to deciding what to substitute first (perimeter, endpoint, SIEM, virtualization) and assessing the risks involved.
The H200 front: massive orders and customs as leverage
Alongside the “software war,” the importation of accelerators has become a strategic maneuver. Reuters reported that Chinese customs authorities have received instructions to block H200 entries into the country, with guidelines for tech companies to avoid purchasing unless “necessary,” with possible exceptions for R&D and academic collaboration.
The figures illustrate the tension: based on the same sources, Chinese companies reportedly placed orders for over 2 million units of H200, valued at approximately $27,000 each. If you consider this figure, it explains why the 25% tariff is far from insignificant: it involves sums capable of affecting margins, deployment schedules, and strategies for local hardware substitution.
US investments: the incentive that accompanies the punishment
The tariff is accompanied by a classic political message: domestic manufacturing within the U.S. can qualify for exemptions or favorable conditions. In this context, promises of industrial investments from major players have been cited. Reuters reports that NVIDIA has outlined plans to spend $500 billion over several years to expand manufacturing and design capacity in the U.S., while TSMC’s projects in Arizona would total over $165 billion in planned investments.
For the component industry, this fosters an already visible trend: reorganizing supply chains, diversifying packaging, and pushing for local production in “friendly” jurisdictions. The collateral effect is a market with increased friction, more controls, and, predictably, higher administrative costs.
Frequently Asked Questions
Will the 25% tariff directly increase the cost of H200 and MI325X for Chinese customers?
Yes: the design of the tariff aims to capture part of the chip’s value when its final destination is China, though the exact impact depends on the import, reexport, and exemption framework.
Why has China targeted software (VMware, Palo Alto, Fortinet, Check Point) in its response?
Because it’s an immediate lever to reduce technological dependence and increase operating costs with Western providers in sensitive sectors, supporting substitution strategies like Xinchuang.
What is Xinchuang, and how does it affect technology procurement in China?
It is a technology substitution initiative to replace foreign software and platforms with domestic alternatives, especially in government and state-owned enterprises, with phased targets up to 2027 in certain sectors.
What are the implications for AI hardware integrators and distributors?
Increased compliance burdens (licenses, verification, traceability), logistical uncertainties (customs and approvals), and demand volatility—complicating stock planning and delivery schedules.
via: techspot