NVIDIA RTX 5090D v2 Banned in China Amid Export Curbs

๐กUnderstand how GPU export bans are reshaping the availability of high-end AI hardware in global markets.
โก 30-Second TL;DR
What Changed
RTX 5090D v2, a China-exclusive GPU, is now officially banned.
Why It Matters
These restrictions force AI developers in China to rely on older hardware or cloud-based alternatives, potentially slowing down local AI model training capabilities.
What To Do Next
If your AI infrastructure relies on NVIDIA hardware, diversify your cloud provider regions to mitigate risks associated with regional export bans.
Key Points
- โขRTX 5090D v2, a China-exclusive GPU, is now officially banned.
- โขThe ban follows previous restrictions on high-performance computing chips.
- โขMarket prices for standard RTX 5090 units in regions like South Korea are surging due to supply chain shifts.
๐ง Deep Insight
Web-grounded analysis with 24 cited sources.
๐ Enhanced Key Takeaways
- โขThe ban on the RTX 5090D v2 was imposed by China itself, not the US, and occurred during a high-level diplomatic visit involving Nvidia CEO Jensen Huang and US President Donald Trump.
- โขDespite being designed to comply with US export controls, Chinese AI developers were reportedly using the RTX 5090D v2 as a workaround to access Blackwell architecture capabilities for AI workloads.
- โขThe RTX 5090D v2 features further reduced VRAM (24GB GDDR7) and memory bandwidth (384-bit interface) compared to the original RTX 5090D, specifically to limit its AI performance and comply with US regulations.
- โขChina's ban signals a strategic shift towards reducing dependency on downgraded US chips and actively promoting domestic alternatives from companies like Huawei and Cambricon.
- โขEven with US approval for some Chinese companies to purchase Nvidia's H200 AI chips, China has not permitted these imports, indicating a deliberate policy to foster its own semiconductor industry.
๐ ๏ธ Technical Deep Dive
- GPU Name: GB202 (Variant: GB202-240-K1-A1)
- Architecture: Blackwell 2.0
- Process Node: 5 nm (TSMC 4N FinFET)
- Transistors: 92,200 million
- Die Size: 750 mmยฒ
- Shading Units: 21760
- Texture Mapping Units (TMUs): 680
- Render Output Units (ROPs): 176
- Tensor Cores: 680 (5th generation)
- Raytracing Acceleration Cores: 170 (4th generation)
- Memory: 24 GB GDDR7
- Memory Interface: 384-bit
- Memory Clock: 1750 MHz (28 Gbps effective)
- Memory Bandwidth: 1,344 GB/s
- GPU Clock: 2017 MHz (Base), 2407 MHz (Boost)
- Bus Interface: PCI-Express 5.0 x16
- Thermal Design Power (TDP): 575 W (up to 600W for factory OC models)
- AI TOPS Performance: 2375 AI TOPS (compared to 3352 AI TOPS on the full RTX 5090)
- Launch Price: 2299 US Dollars / ยฅ16499 CNY
๐ฎ Future ImplicationsAI analysis grounded in cited sources
โณ Timeline
๐ Sources (24)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
- overclock3d.net
- thenextweb.com
- wccftech.com
- investing.com
- seekingalpha.com
- tomshardware.com
- videocardz.net
- wccftech.com
- youtube.com
- chosun.com
- wccftech.com
- moderndiplomacy.eu
- techpowerup.com
- computerweekly.com
- wikipedia.org
- idtechex.com
- laweconcenter.org
- videocardz.com
- tomshardware.com
- wccftech.com
- videocardz.com
- videocardz.net
- wikipedia.org
- bisi.org.uk
Weekly AI Recap
Read this week's curated digest of top AI events โ
๐Related Updates

Taobao expands instant retail 'Convenience Store' network

Samsung Secures $200B Broadcom AI Infrastructure Deal

PACA: Open-source tool for ancient fossil coordinate mapping

First atmosphere detected on habitable-zone exoplanet
AI-curated news aggregator. All content rights belong to original publishers.
Original source: cnBeta (Full RSS) โ