The SAFE Chips Act is a proposed Senate bill, not a current export ban. S.3374 would require the Commerce Department to deny licenses for certain advanced integrated circuits shipped, reexported, or transferred in China and five other designated adversary destinations for at least 30 months after enactment. That could leave Nvidia and AMD selling China-specific accelerators such as the reported H20 and MI308 class while newer products move beyond the permitted limits—but the bill has not passed Congress or taken effect.
The bill was introduced on December 4, 2025, and referred to the Senate Committee on Banking, Housing, and Urban Affairs. The official legislative record does not show enactment. Read the bill record at GovInfo.
What the SAFE Chips Act is
The Secure and Feasible Exports of Chips Act of 2025 is Senate bill S.3374. Senators Pete Ricketts and Chris Coons introduced it with Tom Cotton, Jeanne Shaheen, Dave McCormick and Andy Kim as cosponsors. It was read twice and sent to committee; it has not passed either chamber, been signed by the president or created a new export restriction.
The proposal would amend the Export Control Reform Act of 2018. Its mechanism is license denial for covered advanced integrated circuits, rather than an immediate blanket prohibition on every Nvidia or AMD GPU.
#1 Best Overall
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
What it would do if enacted
A minimum 30-month restriction
The Commerce Secretary would have to deny licenses for covered advanced chips for not less than 30 months. The clock would begin after enactment, not on the December 4, 2025 introduction date. A “mid-2028” endpoint is therefore only a conditional estimate based on enactment around the introduction date, not a fixed deadline in current law.
Destinations covered
The bill defines foreign-adversary countries to include the People’s Republic of China, Hong Kong, Macau, Russia, Iran and North Korea. China is the central commercial focus, but the text is not China-only.
Exports, reexports and in-country transfers
The covered transactions include an export from the United States, a reexport through another country and an in-country transfer. Product legality would still depend on destination, end user, end use, classification and licensing requirements; a model name alone would not decide the result. See the proposed statutory language.
Rank #2
- High-Performance AI Processing: The MX3 is designed to handle the most demanding AI computer vision workloads, delivering exceptional performance and efficiency.
- Flexible Integration: The MX3 can be easily integrated into your existing systems via its M.2 M-key form factor and support for Linux operating systems.
- Energy Efficient: The MX3 is designed to provide high performance while minimizing power consumption.
- Comprehensive Software Development Kit (SDK): The MX3 is supported by a comprehensive SDK that simplifies development and deployment.
- Hardware compatability: The MX3 is compatible with the PCI-SIG M.2 M-key 2280 Specification. It can be used with the Raspberry Pi 5 with a M-key 2280 HAT.
How “advanced” would be determined
The bill ties its definition to chips more capable than those currently permitted for export to adversary destinations. Reporting on the proposal describes thresholds involving total processing performance (TPP), performance density, DRAM bandwidth, interconnect bandwidth and combined bandwidth. Frequently cited figures include:
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problems| Reported parameter | Threshold |
|---|---|
| TPP | 4,800 or more |
| TPP plus performance density | 2,400 or more with density of 1.6 or more |
| TPP plus performance density | 1,600 or more with density of 3.2 or more |
| DRAM bandwidth | 4,100 GB/s or more |
| Interconnect bandwidth | 1,100 GB/s or more |
| Combined DRAM and interconnect bandwidth | 5,000 GB/s or more |
These figures reflect reporting about the proposed framework and underlying Commerce classifications, not a simple statutory list of named products. A future accelerator would need a technical and export-control analysis before anyone could say whether it crossed a limit. Tom’s Hardware’s analysis explains the reported thresholds.
Why H20, L20, L2 and MI308 matter
Secondary reporting identifies Nvidia’s H20, L20 PCIe and L2 PCIe, and AMD’s Instinct MI308 as examples of China-oriented or export-compliant accelerators. They are not named as the only legal products in S.3374. They illustrate the practical ceiling created by the existing technical rules.
Rank #3
- ✅Powered by 26 Tera-Operations Per Second (TOPS) Hailo-8 AI Processor. 2.5W typical power consumption
- ✅Scalable, enabling simultaneous processing of multi-streams & multi-models
- ✅Enabling real-time, low latency and high-efficiency AI inferencing on the edge devices
- ✅Supports TensorFlow, TensorFlow Lite, ONNX, Keras, Pytorch frameworks
- ✅Supports Linux and Windows. Supports the temperature range of -40°C to 85°C
If Nvidia or AMD develops a newer architecture that exceeds the frozen parameters, the companies could be unable to export that part under the bill even if it is deliberately configured below the global flagship. Conversely, a product described as exportable under the reported thresholds could still require a license or face separate end-user and end-use controls.
It would not ban every GPU
Consumer and gaming products not designed or marketed for data-center deployment may be treated differently. Classification, marketing, customer identity and intended use remain important, and the bill does not erase other entity-list, end-use or enforcement rules. It is therefore inaccurate to describe S.3374 as a ban on all Nvidia and AMD graphics processors.
Why supporters want a statutory freeze
Ricketts and Coons argue that advanced AI computing has military and strategic importance and that China’s military-civil fusion system could apply powerful chips to military modernization. Their proposal would prevent an administration from loosening technical limits unilaterally, provide a more durable policy signal and require congressional visibility into later changes. Those are the sponsors’ national-security arguments, not findings that the bill has already produced those outcomes. The sponsors’ explanation is published here.
Rank #4
- 48GB AI graphics accelerator
The commercial case against a fixed ceiling
Nvidia’s reported position is that controlled sales of useful but restricted accelerators keep Chinese customers connected to U.S. hardware and software. That can preserve revenue, developer relationships and influence over standards while slowing a switch to domestic alternatives. It may also help U.S. companies fund research and development.
Industry critics of a freeze warn that total exclusion could instead make Huawei, Biren, Moore Threads and other Chinese suppliers the default. Customers may invest in local software stacks and never return, while lost sales reduce U.S. companies’ scale. A fixed ceiling also risks becoming commercially obsolete as memory, packaging, interconnects and performance per watt improve.
“Outdated” is relative. An accelerator that trails frontier training hardware may remain useful for inference, less demanding workloads or systems that value software compatibility. Real competitiveness depends on more than peak arithmetic: memory capacity, bandwidth, multi-GPU links, power, yields, packaging, complete servers, cloud access, CUDA or ROCm support and developer adoption all matter.
Best Value
- Professional AI & Creator Workstation: AMD Radeon AI PRO R9700 GPU with 32GB GDDR6 is engineered for AI development, professional content creation, and compute-intensive workloads.
- Massive 32GB Memory Capacity: 32GB of GDDR6 memory on a 256-bit bus provides ample bandwidth for large AI models, 8K video editing, and complex 3D rendering.
- Advanced RDNA 4 with AI Accelerators: 64 Compute Units with 3rd Gen Ray Tracing and dedicated 2nd Gen AI Accelerators for groundbreaking AI performance and visual computing.
- Professional Blower Cooling: Efficient single blower design exhausts heat directly out of the chassis, ideal for multi-GPU workstation and server configurations.
- Enterprise-Grade Thermal Solution: Vapor chamber heatsink with industrial Honeywell PTM7950 thermal interface material ensures reliable cooling under sustained professional loads.
China’s alternatives and the measurement problem
Huawei and other Chinese vendors are developing competing accelerators, but headline comparisons require caution. Reported performance figures can use different precision formats, sparsity assumptions, software stacks, system configurations and workloads. A claimed advantage in one metric does not establish equivalent training capacity, production volume or deployment scale.
A prohibition could therefore have two opposing effects: deny Chinese buyers access to the newest U.S. compute, while giving domestic suppliers a larger captive market and more time to improve hardware and software.
What happens after the initial period
After the minimum restriction period, the bill would permit the executive branch to update technical parameters, but with additional checks:
- A majority of the End-User Review Committee would have to approve proposed updates.
- Proposed changes would be published in the Federal Register.
- Congress would receive a briefing at least 30 days before the changes were proposed.
- The administration would assess effects on Chinese military capabilities, cyber operations and the global AI balance.
The bill would not freeze every export-control policy. Other classifications, end-use restrictions, entity designations and enforcement actions could continue to change.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What to watch next
- Committee action: S.3374 must move beyond its referral before the Senate can consider it.
- Legislative changes: Congress could amend, replace or delay the proposal, or take no further action.
- Enactment date: Only enactment would start the 30-month calculation.
- Product classifications: Nvidia and AMD offerings would need to be evaluated against the final text and Commerce rules, not press shorthand about H20, MI308 or a future architecture.
Implications for investors and enterprise buyers
Nvidia and AMD
- Chinese data-center demand could shrink or shift permanently to domestic suppliers.
- Companies may need separate products and designs for different regulatory markets.
- Lost revenue and developer relationships could affect ecosystem influence and research budgets.
- Uncertainty would persist until Congress, the administration and Commerce clarify the final rules.
Enterprise AI infrastructure
- Organizations operating in China could face greater reliance on local accelerators and software stacks.
- Cross-border deployments would require closer checks of destination, end user, end use and licensing.
- Different hardware availability and support models could increase infrastructure fragmentation.
The Bottom Line
The SAFE Chips Act is a proposed 30-month statutory freeze on exports of certain advanced AI chips to China and other listed adversaries—not an enacted ban. If Congress passes it, the measure could keep Nvidia and AMD from refreshing China offerings as quickly as their global products, but it could also accelerate China’s move to domestic hardware and software.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
