On September 15, 2025, the State Administration for Market Regulation of China (hereinafter referred to as the "Market Regulation Administration") issued an announcement, officially recognizing that the global artificial intelligence and graphics computing giant—NVIDIA Corporation—violated the Anti-Monopoly Law of the People's Republic of China and a merger review decision with additional restrictive conditions made five years ago, and has initiated further investigation in accordance with the law. Upon the news, the capital market reacted swiftly, with NVIDIA's stock price in the U.S. pre-market trading dropping by more than 2%.
This is not the first time NVIDIA has faced regulatory headwinds in China. As early as December 2024, the Market Regulation Administration had already launched an investigation into it, with the accusations focusing on the 2020 acquisition of Israeli networking equipment manufacturer Mellanox, believing it to have violated the additional restrictive conditions promised at the time of the merger. After nine months of preliminary investigation, the regulatory agency's "upgrade" action marks the formal entry of the case into a new phase that is deeper and more substantive.
For this company, which holds an absolute dominant position in the global AI chip market and is deeply involved in the construction of China's computing power infrastructure, "further investigation" is not merely a procedural advancement but a critical juncture that may trigger a series of chain reactions. It concerns not only NVIDIA's own business strategy and financial performance but may also reshape its competitive landscape in China, the world's largest semiconductor consumer market, and even set a new benchmark for multinational tech giants' compliance operations in China. This article aims to delve into the legal implications and regulatory logic of this "further investigation" and systematically assess its potential far-reaching impacts on NVIDIA, the Chinese market, and the global semiconductor industry.
Looking back five years at the "marriage": a highly publicized merger and the additional "tightening spell." To understand the core of the current regulatory storm, one must rewind to 2020. At that time, NVIDIA completed its acquisition of Mellanox for a staggering $6.9 billion, which was not only the largest transaction in its company's history but also a key step in its transformation from a single GPU supplier to a comprehensive data center platform provider. Mellanox held an almost monopolistic market position in high-performance networking interconnect devices (such as InfiniBand adapters). By deeply integrating the powerful computing capabilities of GPUs with Mellanox's high-speed, low-latency data transmission technology, NVIDIA was able to build a complete data center solution from the computing core to the network hub, with its strategic intent clear: to create an unshakeable "computing + networking" ecological moat.
However, this strong union raised significant antitrust concerns across multiple jurisdictions globally, and the Chinese market was no exception. The core concern of the regulatory agency was that the newly merged entity would hold an overwhelming market share in several key markets, including GPU accelerators, dedicated networking interconnect devices, and high-speed Ethernet adapters, potentially leading to exclusionary and restrictive competition effects, harming downstream customers (such as cloud service providers and server manufacturers) and end consumers.
Based on the above considerations, the Market Regulation Administration formally approved the transaction in April 2020 under additional restrictive conditions. These conditions constituted "behavioral commitments" that NVIDIA must strictly adhere to, including: prohibiting bundling and additional unreasonable conditions: it must not require Chinese customers to purchase Mellanox's networking equipment simultaneously when buying its GPU accelerators, and vice versa. Ensuring customer choice: it must not obstruct or restrict customers from purchasing or using a particular type of product separately and must ensure equal treatment for customers who purchase a particular type of product separately, without discrimination. Adhering to the principles of fairness, reasonableness, and non-discrimination (FRAND): it must continue to supply its GPUs and Mellanox's related products to the Chinese market under fair, reasonable, and non-discriminatory conditions.
Maintaining interoperability: it must ensure that its GPUs and Mellanox devices can maintain good interoperability with third-party networking or computing devices, without artificially setting technical barriers. Upholding open-source commitments: it must continue to maintain the open-source commitment of Mellanox's related communication software, preventing it from locking in customers through closed-source practices. These restrictive conditions are essentially a "tightening spell" set to prevent NVIDIA from abusing its market dominance. The direct trigger for this "further investigation" was the regulatory agency's determination, after the preliminary investigation, that NVIDIA is suspected of violating the aforementioned commitments.
From case filing to "further investigation": the upgrade and substantialization of legal procedures. In the antitrust enforcement system, "further investigation" is not a vague term but a key link with clear legal meaning and strict procedural requirements. According to China's Anti-Monopoly Law and related supporting regulations, for cases suspected of illegal implementation of operator concentration, enforcement agencies typically go through two main stages: preliminary investigation stage: the Market Regulation Administration, upon receiving a report or discovering clues on its own, will file a case and notify the investigated operator in writing. In this stage, the Administration will conduct preliminary factual verification and legal judgment on whether the transaction constitutes illegal implementation of operator concentration, usually within a period of 30 days.
Further investigation stage: if the preliminary investigation finds that "it constitutes illegal implementation of operator concentration," the enforcement agency will make a decision to conduct further investigation in accordance with the law and notify the operator in writing again. This is the core and deep water area of the entire investigation procedure. In this stage, the enforcement agency will conduct a comprehensive and in-depth assessment of whether the transaction "has or may have the effect of excluding or restricting competition." The investigation period is usually 120 days, and in special circumstances, it can be extended according to the law. Therefore, from "case filing" to "further investigation" is not merely a change in wording but a clear signal from the regulatory agency that, based on preliminary evidence, it formally recognizes the case as having illegal suspicion and decides to invest more enforcement resources for in-depth verification. It signifies that the investigation into NVIDIA has crossed the "shallow water area" of factual sorting and has officially entered the "core area" of evidence certainty, behavioral illegality, and competitive impact assessment.
In-depth analysis: where is the regulatory agency targeting? Although the wording of the Market Regulation Administration's announcement is rigorous and does not disclose specific details of NVIDIA's suspected violations of the restrictive conditions, combining industry dynamics and expert analysis, its investigation direction may mainly focus on the following aspects: suspected bundling or tying sales: this is the most direct and easiest to prove violation. For example, the regulatory agency may have evidence that NVIDIA, when selling its highly sought-after A100, H100, and other high-end AI chips or specially supplied H20 chips in the Chinese market, made the purchase of Mellanox's InfiniBand switches or DPUs (data processing units) a hidden or explicit necessary condition.
Discriminatory pricing or supply: whether NVIDIA offers better prices, more stable supply guarantees, or prioritized technical support to customers who purchase both its GPUs and Mellanox devices (such as large cloud service providers), while setting higher procurement thresholds or longer delivery periods for customers who only purchase its GPUs (such as some AI startups or OEM manufacturers choosing other brand networking devices), thereby indirectly restricting customers' choices. Weakening interoperability: whether NVIDIA, through software updates, changes in technical interfaces, etc., causes its GPUs to perform worse when used with third-party networking devices or fail to achieve optimal performance, thereby artificially "deep binding" its chips with Mellanox devices.
Liu Xu, a special researcher at Tsinghua University's National Strategy Institute, pointed out that since two of the additional restrictive conditions are confidential, the outside world cannot yet determine which specific commitment NVIDIA has violated. However, regardless of which one it is, its essence is the destruction of market competition order, placing competitors at an unfair disadvantage, and ultimately harming the interests of China's downstream industries.