Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesReuters reported on June 24, 2024, that ByteDance was working with Broadcom on a custom 5-nanometer AI chip intended to meet U.S. export restrictions, with TSMC reportedly expected to manufacture it. That was a report of a development effort—not confirmation of a launched product. The available evidence does not establish that the chip taped out, entered production, shipped to China, or remains eligible under rules that have changed since 2024.
What Reuters reported
Reuters, citing two people familiar with the matter, reported that ByteDance and Broadcom were working on a customized 5 nm application-specific integrated circuit (ASIC) for artificial intelligence. The chip was reportedly intended to comply with U.S. export restrictions, and TSMC was expected to manufacture it. Reuters also reported that ByteDance and Broadcom had been business partners since at least 2022. Reuters report mirrored by Investing.com; Reuters report mirrored by StreetInsider.
The reports did not disclose a public contract, product name, performance figures, price, production volume, or shipment schedule. They also did not provide public confirmation from ByteDance or Broadcom. “Working with” should not be read as proof of a definitive partnership agreement, an order for finished chips, or a completed design.
Why ByteDance might want a custom AI chip
ByteDance runs services that rely on computing for recommendations, advertising, search, generative AI, and content moderation. A custom accelerator could be designed around some of those workloads rather than serve as a general-purpose graphics processor. That can offer tighter workload fit and, in principle, better cost or energy efficiency for selected tasks.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →#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
Supply planning is another possible motive. U.S. restrictions have limited Chinese access to some advanced accelerators, while export rules and availability for China-oriented products have shifted. A direct design program could give ByteDance more influence over its supply plans and reduce reliance on merchant GPU availability. That is a rationale for diversification, not evidence that the proposed chip could replace Nvidia products across ByteDance’s computing needs.
Custom silicon also brings trade-offs. It can be less flexible than a GPU when workloads change, and its value depends on software support, memory, networking, packaging, production yield, and the cost of designing and deploying it. Nvidia’s mature software ecosystem and broad compatibility remain relevant advantages; domestic Chinese suppliers may offer local availability and strategic alignment but are not interchangeable on every technical or operational measure.
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.
What Broadcom and TSMC’s roles would mean
Broadcom is a chip-design and infrastructure supplier, so the reported arrangement fits a customer-specific ASIC program. In a typical program, a design partner may help translate a customer’s requirements into architecture and logic, integrate elements such as memory controllers and high-speed interfaces, prepare the design for manufacturing, and support validation. Those are common ASIC-program functions, not confirmed details of this particular effort. The public reporting does not establish which tasks Broadcom would perform or which intellectual property it would supply.
Reuters said TSMC was expected to manufacture the chip. A foundry relationship is not automatic production: the design must reach a manufacturable state, a foundry must accept and schedule it, capacity must be available, and the relevant transactions must comply with applicable rules and any licensing requirements. Advanced packaging and access to suitable memory can also constrain a product even if wafer manufacturing is possible.
Free tools Windows power users keep installed
One-click scans. No signup required.
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
TSMC’s reported involvement matters because it would place the project in a cross-border supply chain subject to controls on advanced computing chips and certain semiconductor-manufacturing activities. U.S. rules can reach some foreign-produced items and foundry-related transactions, depending on the applicable rule and facts. BIS guidance on advanced-computing controls and foundry due diligence; BIS controls concerning China’s advanced-semiconductor production capabilities.
What “5 nm” and “U.S.-compliant” do—and don’t—tell you
5 nm is a process label, not a performance rating
The 5 nm description came from Reuters’ unnamed sources; no public product specification is available. A process-generation label does not by itself establish AI performance, energy efficiency, transistor count, memory bandwidth, interconnect speed, or software compatibility. Two chips made on a process described with the same node label can differ substantially in architecture and system performance.
Rank #4
- 48GB AI graphics accelerator
Compliance is not the same as government approval
In the 2024 report, “compliant” described the reported design intention: to meet applicable U.S. export-control thresholds and licensing requirements. It did not mean the U.S. government had certified the chip or authorized every transaction involving it.
Whether a chip or system may be exported, transferred, or used can depend on technical characteristics as well as destination, customer, end use, server configuration, and transaction structure. Controls can also involve memory and interconnect characteristics, packaging, manufacturing, software, and rules concerning foreign-produced items. BIS guidance issued on May 13, 2025, says certain advanced-computing integrated circuits and related commodities used to train AI models may require authorization for activities involving China or other D:5 destinations. BIS policy statement on AI-model training and advanced computing.
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.
Broadcom’s inclusion in an EAR list of approved integrated-circuit designers does not establish that this ByteDance project was approved. EAR Part 740. A design that aimed to fit rules in June 2024 would need fresh legal and technical review under rules in effect later; a company’s internal compliance assessment is not a BIS determination.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How U.S. policy changed after the report
| Date | Development | Why it matters to this project |
|---|---|---|
| October 7, 2022 | The United States introduced major controls on advanced-computing chips and semiconductor-manufacturing equipment destined for China. BIS announcement. | These controls formed part of the regulatory backdrop for a chip reportedly designed with export restrictions in mind. |
| October 2023 | BIS updated and expanded controls on advanced computing and semiconductor manufacturing. | The applicable technical and transaction rules evolved after the initial controls. |
| April 2024 | BIS issued clarifications concerning advanced computing, supercomputing, and semiconductor-manufacturing controls. BIS announcement. | The clarified framework was part of the environment in which Reuters reported the proposed project in June. |
| June 24, 2024 | Reuters reported the ByteDance–Broadcom 5 nm ASIC effort. | The reported compliance objective was tied to the rules and circumstances at that time. |
| January 15, 2025 | BIS strengthened advanced-computing restrictions and added foundry due-diligence measures. BIS announcement. | Foundry participation and customer screening are part of the practical compliance picture, not merely the chip’s node label. |
| May 13, 2025 | BIS rescinded the Biden-era AI Diffusion Rule and issued additional semiconductor-control guidance. BIS announcement. | The regulatory framework continued to change; the rescission did not mean that all controls on advanced computing disappeared. |
| May 13, 2025 | BIS stated that certain advanced-computing ICs and related commodities used to train AI models may require authorization for activities involving China or other D:5 destinations. BIS policy statement. | Use and activity can matter alongside the chip’s technical design. |
| January 13, 2026 | BIS announced case-by-case license review for exports of Nvidia H200, AMD MI325X, and similar chips to approved Chinese customers, subject to specified security and supply conditions. BIS announcement. | A change in review policy for certain chips is not blanket authorization for other products, customers, or transactions. |
| April 7, 2026 | BIS extended the timeline for certain authorized integrated-circuit designers through December 31, 2026. BIS. | That extension does not itself establish authorization for this project. |
The rules applicable to any proposed chip would need to be assessed as of the relevant transaction and use—not inferred from the phrase “U.S.-compliant” in a 2024 news report. The Export Administration Regulations, Part 744 include end-use and end-user controls relevant to advanced-node and supercomputer activity.
How the project fits Nvidia and China’s chip alternatives
Reuters later reported that Chinese firms, including ByteDance, increased orders for Nvidia’s China-oriented H20 chips ahead of subsequent restrictions affecting them. It also reported that Nvidia discussed China-tailored chips with customers including ByteDance, Alibaba, and Tencent. Reuters report on Nvidia’s H20; Reuters report on China-tailored Nvidia chips.
Those reports show why a custom ASIC should be viewed as one possible part of a broader supply strategy. Nvidia GPUs offer flexibility and an established software stack; a ByteDance-specific ASIC could be more optimized for selected workloads but would likely have a narrower ecosystem. Huawei and other Chinese suppliers are another part of the landscape. The reporting does not establish that ByteDance’s proposed Broadcom chip was meant to replace Nvidia entirely, or that it achieved performance parity with any named accelerator.
What remains unverified
- Whether the reported effort reached a completed design or tape-out.
- Whether TSMC accepted or manufactured the design, and whether it reached production.
- The product name, specifications, memory configuration, price, order volume, and shipment schedule.
- Whether ByteDance deployed the chip, where it would be used, or whether it received any specific export authorization.
- Whether the reported collaboration remains active.
Advanced-node feasibility would not settle the commercial questions by itself: the project would also need adequate yield, packaging, memory supply, software readiness, and production capacity. Nor does a reported design objective establish legal permission for the chip, server, or downstream use.
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.
