China’s Review of Nvidia RTX Pro 5500 Purchase Approvals and the Realignment of US-China Semiconductor Regulatory Consistency
Executive Summary
The decision by China’s Ministry of Industry and Information Technology (MIIT) to signal its intent to approve purchases of the Nvidia RTX Pro 5500 for ByteDance and Alibaba comes despite semiconductors not being on the official agenda of the Trump-Xi summit. This demonstrates that the practical rules of US-China technological competition are being shaped not by top-level agreements, but by the individual decisions of regulatory authorities and corporations in both countries. While Beijing officially maintains its localization policy centered on the Huawei Ascend series, it adopts a dual approach by selectively yielding to practical pressure from big tech companies facing computing power shortages. The performance threshold-based export controls of the United States have exposed structural limitations, failing to block the diversion of workstation chips for AI purposes—an issue closely linked to regulatory oversight gaps, such as the prosecution of Nvidia and Supermicro representatives by the Keelung District Prosecutors Office in Taiwan. South Korea must adopt a dual strategy: managing the imperfections of the US control regime as a constant variable while maintaining its leverage through super-gap technologies in HBM and foundries.
I. Analysis of the Issue
China’s Review of Nvidia RTX Pro 5500 Purchase Approvals: Issue Analysis
1. Background and Progress
China’s Ministry of Industry and Information Technology (MIIT) has requested domestic big tech companies, including ByteDance and Alibaba, to report their purchase plans for the Nvidia RTX Pro 5500 [1]. It is reported that signals of intent to approve have already been delivered to some of these companies [1]. The backdrop to this development is the computing power shortage facing Chinese AI firms. ByteDance and Alibaba have encountered computing power bottlenecks while expanding their proprietary AI services [1].
This move appears to run counter to Beijing’s established policy stance. The Chinese government has officially maintained a path of semiconductor self-reliance, led by the Huawei Ascend series. Indeed, driven by the AI boom and localization campaigns, the first-half net profits of Chinese semiconductor companies such as CXMT and SMIC surged by 620% [13]. The *Global Times* maintains a tone that emphasizes the expansion of domestic computing capabilities, reporting projections that China’s intelligent computing power will surge by 87.9% year-on-year to reach 2,576.5 EFLOPS by 2026 [15]. Reports by Taiwan’s *DigiTimes* that MIIT is expanding the replacement of foreign technologies, including high-speed networking hardware, through an audit of Broadcom also align with this self-reliance stance [4].
However, this is not an unprecedented case. The Brookings Institution pointed out that Beijing has quietly allowed several large tech companies to purchase limited quantities of H200 chips in recent weeks [2]. This contrasts with the fact that China did not approve any purchases during and immediately after President Xi Jinping’s visit to the United States, even though President Trump had approved the export of H200 chips despite domestic political risks [2]. The current review of RTX Pro 5500 approvals can be seen as an extension of this "quiet allowance" pattern.
2. Current Situation
The market reacted immediately to the news. The CSI 300 Index fell by as much as 2.4% during intraday trading on Monday [9]. The downward pressure on the stock prices of Chinese semiconductor companies is interpreted as stemming from investor concerns that if Beijing—which has prioritized nurturing domestic chips—permits the purchase of foreign chips again, the market position of domestic alternatives like the Huawei Ascend could be undermined [9].
The time lag relative to the summit is also notable. The *South China Morning Post* reported that semiconductor export controls did not emerge as a key issue during the Trump-Xi summit [7]. Despite Jensen Huang and Lisa Su were seated alongside the two leaders at the White House state dinner, observers assess that it remains unclear whether any substantive progress was made toward the core goal of restoring market access to China for Nvidia and AMD [7]. *Nikkei* also pointed out that the direction of US-China AI cooperation remains uncertain following the summit [10]. With Silicon Valley reacting skeptically to the hotline initiative proposed by Washington, the two countries went no further than including AI on the agenda of their bilateral economic dialogue and securing leader-level endorsement [8][14].
In this context, the review of RTX Pro 5500 approvals should be viewed not as a product of a summit agreement, but as a practical decision made at the regulatory level, proceeding independently of the summit. This can be interpreted as a pragmatic measure to bridge the gap between Beijing’s semiconductor self-reliance policy and its actual demand for computing power.
3. Key Actors and Positions
China’s Ministry of Industry and Information Technology (MIIT)is the practical decision-maker in this matter. By requesting purchase plans from ByteDance and Alibaba and signaling its intent to approve purchases for certain firms, MIIT has opted to manage exceptions on an individual company basis while maintaining the principle of localization [1]. This is read not as a wholesale policy shift at the national level, but as a selective response to the realistic pressure of computing power shortages.
ByteDance and Alibabaare consumers in urgent need of computing power. To avoid falling behind in the AI service race, they require performance-proven Nvidia chips, yet they find themselves in a position where they must obtain purchase approvals under a policy regime that prioritizes the use of domestic chips [1].
The Chinese Domestic Semiconductor Industryis perceived by the market as a potential victim of this review. The decline in the CSI 300 Index and the pressure on semiconductor stock prices reflect concerns over the weakening market position of this industry [9]. However, the surging financial performance of companies like CXMT and SMIC suggests that the localization trend itself is not faltering [13].
Nvidiaprioritizes restoring its access to the Chinese market. While the seating of Jensen Huang at a prominent table during the state dinner underscores the intensity of the company's lobbying efforts, substantive outcomes are materializing through the individual decisions of Chinese regulators rather than summit agreements [7][10].
The US Department of Commerce and Administrationoccupy a relatively passive position in this matter. Washington’s export controls are designed around performance thresholds, but they reveal structural limitations by failing to fundamentally block the diversion of workstation chips like the RTX Pro 5500 for AI computing. This is similar to the loopholes in the whitelist system identified in the illegal AI server export case originating from Taiwan [5].
Taiwanese Authoritiesare attempting to strengthen oversight at the forefront of export control circumvention. The indictment of nine individuals associated with the Taiwanese subsidiaries of Nvidia and Supermicro by the Keelung District Prosecutors Office demonstrates that channels for controlled GPUs to flow into China through multi-tiered distribution networks are actively functioning [5].
4. Key Issues
The first issue is the consistency of China’s policy. The simultaneous pursuit of localization campaigns and reviews of foreign chip approvals shows that Beijing’s semiconductor self-reliance path is being flexibly adjusted between principle and practice. Crucially, this does not represent an abandonment of the policy to nurture domestic chips like Huawei's; rather, it is being managed as an exceptional measure to address short-term bottlenecks in computing power.
The second issue is the structural limitations of the US export control regime. Performance threshold-based regulations fail to prevent the diversion of borderline products like the RTX Pro 5500. When combined with the gaps in whitelist management exposed by the illegal export case from Taiwan, this could raise growing doubts about the overall effectiveness of the control regime [5].
The third issue is the disconnect between summits and actual regulatory practices. The fact that such moves are occurring despite semiconductors not being on the official agenda of the Trump-Xi summit [7] supports the view that the practical rules of US-China technological competition are being shaped by the decisions of regulatory authorities and individual corporations in both countries, rather than top-level agreements. This has direct implications for the export conditions and supply chain strategies of South Korean semiconductor firms. For allied companies that have built compliance systems aligned with US export controls, China’s unpredictable, case-by-case approval practices could introduce a new source of uncertainty.
II. In-Depth Analysis of the Issue
China’s Review of Nvidia RTX Pro 5500 Purchase Approvals: In-Depth Analysis
1. Root Cause Analysis
The primary cause of this issue is the computing power bottleneck experienced by Chinese AI firms. ByteDance and Alibaba have faced computing power shortages while expanding their proprietary large language model (LLM) services [1]. Although the Huawei Ascend series is being nurtured as a domestic alternative, industry assessments indicate that it has not yet secured the performance and volume required to fully replace Nvidia’s product lineup in both training and inference. Projections that China’s intelligent computing power will surge by 87.9% by 2026 also evidence that demand is outstripping supply capacity [15].
A more fundamental cause lies in the design of US export controls. Washington’s control regime has adopted a threshold-based approach using computing performance (FLOPS) and interconnect bandwidth as criteria. The RTX Pro 5500 is categorized for workstation and graphics use, placing it below these thresholds. However, the industry commonly points out that connecting multiple cards in parallel allows them to be diverted for AI inference and lightweight training. This reveals the structural limitations of an approach that attempts to control end-use solely through a performance baseline. The loopholes in the whitelist system and the lack of oversight in multi-tiered distribution networks were already confirmed when the Keelung District Prosecutors Office in Taiwan indicted representatives of Nvidia and Supermicro's Taiwanese subsidiaries on charges of breach of trust and forgery of documents [5]. While the RTX Pro 5500 case differs in that it involves official approval rather than illegal circumvention, it shares the same structural issue: performance-based controls cannot fundamentally block pathways for end-use diversion.
The cause on Beijing’s side can be found in the gap between political justification and pragmatic necessity. Simultaneously, MIIT is pursuing a localization campaign to expand the replacement of foreign technologies, including high-speed networking hardware, through an audit of Broadcom [4]. This demonstrates a dual stance: maintaining the official line of nurturing domestic chips while yielding to practical pressure from big tech companies facing computing power shortages.
2. Structural Context
From a political structure perspective, the key is that this decision came with a time lag relative to the Xi-Trump summit. Semiconductor export controls did not emerge as a major issue at the summit [7]. Although Jensen Huang and Lisa Su were seated alongside the two leaders at the White House state dinner, it remained unclear whether there was any tangible progress toward the goal of restoring Chinese market access for Nvidia and AMD [7]. The two countries went no further than including AI on the agenda of their bilateral economic dialogue and securing leader-level endorsement [8][14]. This shows that the semiconductor issue is not treated as a subject for a package deal between leaders, but rather as a matter to be adjusted individually by the regulatory authorities of both countries in accordance with their respective domestic political and industrial situations. As the Brookings Institution points out, even though Trump approved H200 exports despite domestic political risks, Beijing did not approve any purchases during or immediately after Xi’s visit to the United States [2]. Only weeks later did it quietly allow several large tech companies to purchase limited quantities of H200 chips [2]. The current review of the RTX Pro 5500 repeats this same pattern, proceeding at the regulatory level with a time lag after the summit.
In the economic structure, the interests of the Chinese semiconductor industry intersect. Driven by the AI boom and localization campaigns, the first-half net profits of Chinese semiconductor companies like CXMT and SMIC surged by 620% [13]. Under these circumstances, investor concerns that resuming purchases of Nvidia products could undermine the market position of domestic alternatives, including the Huawei Ascend, were immediately reflected in a 2.4% intraday drop in the CSI 300 Index [9]. The stock price reaction of domestic chip companies serves as an indicator of how the market reads the tension between Beijing’s official narrative of self-reliance and its actual purchasing decisions.
In the security structure, this intersects with moves in the United States to tighten restrictions on technology access for China. In the US Congress, legislation is being pursued to expand the scope of products subject to trade blacklists, while Lenovo has become embroiled in patent disputes [11]. Simultaneously, Chinese internet regulators are investigating allegations that DeepSeek and Moonshot AI bypassed restrictions to transmit user interaction data to Anthropic’s Claude model [12]. In other words, both the US and China are showing a trend toward expanding the scope of controls beyond semiconductor hardware to include access routes for data and models. The review of RTX Pro 5500 approvals can be read as an exceptional relaxation limited to the hardware domain amid this phase of comprehensive control expansion.
3. Historical Precedents and Comparison of Similar Cases
The closest precedent is Beijing’s "quiet allowance" of the H200 chip. Even though the Trump administration approved H200 exports despite domestic political backlash, Beijing did not immediately accept this during Xi’s visit to the United States [2]. Only weeks later did it quietly permit several large tech companies to purchase limited quantities [2]. This overlaps with the pattern of the current RTX Pro 5500 case, in that approvals were granted within a limited scope, with a time lag after the summit, and without any public agreement or announcement. The difference is that while the H200 is a high-performance chip for AI servers and was a primary target of export controls from the outset, the RTX Pro 5500 is categorized for workstation use, placing it at the periphery of the control network. From Beijing’s perspective, the latter likely represented a choice with lower political risk.
Another point of comparison is the illegal AI server export case originating from Taiwan [5]. In this case, representatives from Nvidia’s Taiwanese subsidiary and Supermicro’s Taiwanese affiliate abused their whitelist status to illegally sell 74 B300 servers to Chinese buyers [5]. The scheme involved a detour sale to Chinese buyers through a Supermicro distributor after Feihu Technology faced difficulties fulfilling the contract due to financial issues [5]. This case and the review of RTX Pro 5500 approvals differ in nature; the former is an illegal act exploiting the whitelist system, while the latter is a legal process officially reviewed by the Chinese government. However, both cases share the commonality of originating from the same structural vulnerability: performance threshold-based controls fail to filter actual end-uses. The AP characterized the Taiwan case as "another ripple in the US-China AI rivalry," assessing that advanced semiconductors made in Taiwan have emerged as a key battleground in the competition [5].
4. Key Variables Shaping Future Developments
The first variable is the subsequent response of the US Department of Commerce. If the pathway of diverting workstation chips into AI servers through parallel connections is confirmed, the possibility cannot be ruled out that the Department of Commerce will expand its control list to include product lines like the RTX Pro 5500. This could lead to discussions on redesigning the performance threshold criteria themselves.
The second variable is the extent of the gap between Beijing’s localization campaign and actual procurement behavior. Depending on how quickly domestic chip production capacity, including the Huawei Ascend, catches up with big tech demand, it will be determined whether this approval remains a one-off relaxation or solidifies into a recurring pattern. Although the domestic supply chain is growing rapidly, as evidenced by the surging profits of CXMT and SMIC [13], it remains uncertain how much they have narrowed the gap with Nvidia in the cutting-edge AI accelerator domain.
The third variable is whether the AI and semiconductor agendas within the US-China bilateral economic dialogue develop into a substantive negotiation track. Currently, efforts are limited to leader-level endorsements [8][14], and Silicon Valley has reacted skeptically to the hotline initiative proposed by Washington [10]. Unless this track evolves into concrete norms or common standards, semiconductor regulation is highly likely to continue in its current form, where the accumulation of individual decisions by both governments and corporations shapes de facto rules.
These variables interact in complex ways for the South Korean semiconductor industry. If the baselines of US export controls are readjusted, it will affect the classification of export items to China for domestic firms. At the same time, the pace of China’s localization is a variable that determines the sustainability of the windfall benefits South Korean firms enjoy in the memory and foundry sectors. In the current phase, where individual decisions by regulatory authorities substitute for established rules, a system is required to constantly track individual approval and enforcement trends, regardless of any summit agreements.
III. Final Recommended Response Measures
China’s Review of Nvidia RTX Pro 5500 Purchase Approvals: Comprehensive Response Measures
1. Comprehensive Assessment and Recommended Response Measures
This development reaffirms that the rules of US-China semiconductor competition are being determined by the individual decisions of regulatory authorities rather than top-level agreements. Semiconductor export controls were not addressed as a key issue during the Trump-Xi summit [7]. Nevertheless, in the weeks following the summit, Beijing has been reviewing approvals for the RTX Pro 5500, following its review of the H200 [2][1]. This demonstrates that China’s response is not a product of the summit, but rather stems from practical pressure caused by computing power shortages among its domestic big tech firms.
Beijing’s stance is dualistic. The localization path centered on the Huawei Ascend is officially maintained [4]. The profits of domestic semiconductor companies such as CXMT and SMIC are surging [13]. However, the demand for computing power from application-layer companies like ByteDance and Alibaba outpaces the supply rate of domestic alternatives [15]. To bridge this gap, the Chinese government is opting for a managed opening, selectively yielding between strategic justification and pragmatic necessity. The RTX Pro 5500 is categorized for workstation use, placing it below the performance thresholds of US export controls. The structural loophole that allows multiple cards to be connected in parallel for AI computing serves as the backdrop for this approval review.
In this trend, the response principles South Korea should adopt can be summarized into two points. First, assuming that the US performance-based control is not a flawless system, South Korea must manage the risk of domestic firms' transactions with China becoming problematic ex-post as a constant variable. The indictment of Nvidia and Supermicro's Taiwanese subsidiaries by the Keelung District Prosecutors Office in Taiwan demonstrates that gaps in whitelist systems and multi-tiered distribution networks can culminate in actual criminal cases [5]. Second, acknowledging that China’s localization campaign is continuing rather than retreating, South Korea must maintain its leverage through super-gap technologies in HBM and foundries, areas where domestic alternatives cannot replace them.
2. Short-, Medium-, and Long-Term Action Plans
Short-term (1–3 months): The Ministry of Trade, Industry and Energy (MOTIE) and the Ministry of Foreign Affairs (MOFA) must monitor whether the review of RTX Pro 5500 approvals translates into actual export volumes, and verify its impact on the market share of the Huawei Ascend series in China's domestic market. The decline in the CSI 300 Index can be interpreted as a signal that the market has not yet fully gained confidence in the pace of domestic chip substitution [9]. SK Hynix and Samsung Electronics must voluntarily strengthen their end-user verification procedures in memory and packaging supply contracts with China to preemptively prevent involvement in detour distribution networks, as seen in the Taiwan case [5].
Medium-term (6 months to 1 year): It is necessary to continuously track whether the U.S. Department of Commerce adjusts performance thresholds. If the diversion of workstation chips for AI servers remains unaddressed, the United States may use this issue as an opportunity to expand the scope of controls on mid-performance chips. In this case, HBM3E and lower-grade memory or general-purpose GPU-related components handled by South Korean companies could be included in the new regulations. The Ministry of Trade, Industry and Energy (MOTIE), as the trade authority, should establish prior consultation channels starting from the stage of the U.S. pre-announcement of regulatory changes to minimize compliance costs for South Korean companies.
Long-term (1 year or more): Under the premise that China's semiconductor self-reliance is progressing structurally alongside surging profits for CXMT and SMIC, South Korea must prepare for the possibility that the revenue base of South Korean companies in the legacy and mid-range AI accelerator markets will be eroded [13]. At the same time, since China will continue to court South Korea in "super-gap" technology areas where domestic substitution is impossible, a dual-track strategy is required to separately manage technology leakage controls in these areas and leverage for negotiations with China.
3. Monitoring Indicators and Trigger Points
The first indicator is the scope of approval for the RTX Pro 5500 by China's Ministry of Industry and Information Technology (MIIT) and its actual import volume. The key is whether the approval is limited to a few large companies like ByteDance and Alibaba, or if it expands to local government levels [1][9]. The second is the change in the market share of Huawei's Ascend series. Whether the approval of foreign chips actually slows down the pace of nurturing domestic alternatives will serve as a benchmark for assessing the sustainability of Beijing's policy. The third is the U.S. Department of Commerce's moves to readjust performance thresholds. If the diversion of workstation chips becomes an official issue in Congress or the Department of Commerce, it will serve as a precursor to regulatory changes directly affecting South Korean companies. The fourth is whether circumvention distribution cases between Taiwan and China recur. If additional cases similar to the Keelung District Prosecutors Office prosecution occur, the U.S. strengthening of whitelist oversight could proceed faster than expected [5].
Three trigger points can be established: first, if the United States announces new performance standards for workstation GPUs; second, if China expands RTX Pro 5500 approvals to local governments and state-owned enterprises; and third, if further declines in CSI 300 semiconductor-related stocks confirm a decline in confidence in the localization policy. If any of these occur, MOTIE and relevant companies must immediately review their semiconductor export strategies toward China.
4. Summary and Conclusion
The review of the RTX Pro 5500 approval is not a byproduct of the U.S.-China summit, but a practical adjustment resulting from the combination of Chinese Big Tech's computing power shortage and structural loopholes in U.S. export controls [1][2][7]. While not abandoning its localization path, Beijing is selectively opening its doors in necessary areas [4][13]. South Korea must not misinterpret this trend as a sign of easing U.S.-China tensions, but rather recognize it as a fluid situation where regulatory thresholds and circumvention routes are continuously readjusted. A dual-track response—managing compliance risks while maintaining negotiating leverage in super-gap technologies—must serve as the core pillar of future responses.
References
[1] [Exame] China pode liberar compra de novos chips da Nvidia para ByteDance e Alibaba
[2] [Brookings Institution] Why Beijing isn’t taking the bait with Trump
[4] [DigiTimes Asia] Beijing escalates domestic chip drive with Broadcom audits
[7] [South China Morning Post] Why US chip controls took a back seat at the Xi-Trump summit
[8] [DigiTimes Asia] Trump, Xi back AI talks as hotline details remain unclear
[9] [Jornal de Negócios] Ao minuto:Bolsas europeias em terreno misto com aumento dos juros da dívida
[10] [Nikkei Asia] Xi's human control or Trump's 'super intelligence': What's next for AI?
[12] [DigiTimes Asia] DeepSeek, Moonshot, the hidden cost of routing China's AI through Claude
[13] [Nikkei Asia] CXMT, SMIC power 620% surge in profits in China's chip industry
[14] [DigiTimes Asia] Trump, Xi back AI talks as hotline details remain unclear
*This text is an AI translation of an original written in Korean. Some translations or nuances may be inaccurate.
This report is an in-depth analysis planned by an EAI researcher, grounded in sophisticated AI-assisted research, and finalized by the EAI researcher.
How can I help you?