US vs China AI Chips: Architecture, Performance, and Ecosystem Compared
The US currently leads in documented high‑end AI accelerator performance and ecosystem scale, with chips like AMD’s MI325X (≈1.3 PFLOPS FP16 and 256GB HBM3E) and Google’s TPU v6e (918 TFLOPs bf16 per chip), while Chin... China’s strategy focuses on domestic alternatives such as Huawei’s Ascend 910‑series and Biren’s...
Published byEdited with GPT-5.5Images generated with GPT Image 2
The US currently leads in documented high‑end AI accelerator performance and ecosystem scale, with chips like AMD’s MI325X (≈1.3 PFLOPS FP16 and 256GB HBM3E) and Google’s TPU v6e (918 TFLOPs bf16 per chip), while Chin...
China’s strategy focuses on domestic alternatives such as Huawei’s Ascend 910‑series and Biren’s BR100 GPUs, designed for large‑model training and inference and often paired with large cluster systems to compete with...
The US advantage currently extends beyond chips to software ecosystems and manufacturing supply chains, while China’s progress centers on scaling domestic designs and building independent AI infrastructure.
Research US vs China AI Chips and compare them as comprehensively as possible in table formatThe global AI accelerator race increasingly centers on competing chip ecosystems in the United States and China.
AI Prompt
Create a landscape editorial hero image for this Studio Global article: Research US vs China AI Chips and compare them as comprehensively as possible in table format. Article summary: The US side in this evidence set includes Nvidia H200, AMD MI325X, and Google TPU v6e, while the China side is represented mainly by Huawei’s Ascend 910B.. Topic tags: deepresearch, documentation, general web, education, user generated. Reference image context from search candidates: Reference image 1: visual subject "RAND's divisions conduct research on a uniquely broad front for clients around the globe. #### U.S. research divisions. U.S. and China flags on a computer chip on a motherboard. **" source context "China's AI Models Are Closing the Gap—but America's Real Advantage Lies Elsewhere | RAND" Reference image 2: visual subject "RAND's divisions conduct research on a uniquely broad front for clients a
openai.com
The global race for artificial‑intelligence computing is largely defined by competition between US and Chinese chip ecosystems. The United States dominates the high‑performance accelerator market with companies like Nvidia, AMD, and Google, while China is building alternatives through Huawei, Biren, and Cambricon.
Although both countries now produce specialized AI accelerators for training and inference, the competition differs not just in raw chip performance but also in memory bandwidth, manufacturing capability, software ecosystems, and large‑scale cluster design.
Below is a comprehensive comparison of major AI chips from each side using available documented specifications.
Major AI Chips in the US–China Competition
United States
AMD Instinct MI325X
Google TPU v6e (Trillium)
China
Huawei Ascend 910 series (including 910C)
Biren BR100 / BR104
Cambricon MLU370‑X8
These chips are primarily designed for large‑scale AI workloads such as training large language models (LLMs), inference services, and scientific computing.
Comprehensive Chip Comparison
Chip
Country / Company
Architecture / Node
Peak Compute
Memory
Memory Bandwidth
Power
Notable Features
AMD Instinct MI325X
US / AMD
CDNA3 accelerator architecture
~1.3 PFLOPS FP16 (2.6 PFLOPS FP8 with sparsity)
256GB HBM3E
~6 TB/s
up to ~750–1000W
Large HBM capacity designed for large‑model training and inference workloads.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
What is the short answer to "US vs China AI Chips: Architecture, Performance, and Ecosystem Compared"?
The US currently leads in documented high‑end AI accelerator performance and ecosystem scale, with chips like AMD’s MI325X (≈1.3 PFLOPS FP16 and 256GB HBM3E) and Google’s TPU v6e (918 TFLOPs bf16 per chip), while Chin...
What are the key points to validate first?
The US currently leads in documented high‑end AI accelerator performance and ecosystem scale, with chips like AMD’s MI325X (≈1.3 PFLOPS FP16 and 256GB HBM3E) and Google’s TPU v6e (918 TFLOPs bf16 per chip), while Chin... China’s strategy focuses on domestic alternatives such as Huawei’s Ascend 910‑series and Biren’s BR100 GPUs, designed for large‑model training and inference and often paired with large cluster systems to compete with...
What should I do next in practice?
The US advantage currently extends beyond chips to software ecosystems and manufacturing supply chains, while China’s progress centers on scaling domestic designs and building independent AI infrastructure.
US vs China AI Chips: Architecture, Performance, and Ecosystem Compared
The US currently leads in documented high‑end AI accelerator performance and ecosystem scale, with chips like AMD’s MI325X (≈1.3 PFLOPS FP16 and 256GB HBM3E) and Google’s TPU v6e (918 TFLOPs bf16 per chip), while Chin... China’s strategy focuses on domestic alternatives such as Huawei’s Ascend 910‑series and Biren’s...
Published byEdited with GPT-5.5Images generated with GPT Image 2
The US currently leads in documented high‑end AI accelerator performance and ecosystem scale, with chips like AMD’s MI325X (≈1.3 PFLOPS FP16 and 256GB HBM3E) and Google’s TPU v6e (918 TFLOPs bf16 per chip), while Chin...
China’s strategy focuses on domestic alternatives such as Huawei’s Ascend 910‑series and Biren’s BR100 GPUs, designed for large‑model training and inference and often paired with large cluster systems to compete with...
The US advantage currently extends beyond chips to software ecosystems and manufacturing supply chains, while China’s progress centers on scaling domestic designs and building independent AI infrastructure.
Research US vs China AI Chips and compare them as comprehensively as possible in table formatThe global AI accelerator race increasingly centers on competing chip ecosystems in the United States and China.
AI Prompt
Create a landscape editorial hero image for this Studio Global article: Research US vs China AI Chips and compare them as comprehensively as possible in table format. Article summary: The US side in this evidence set includes Nvidia H200, AMD MI325X, and Google TPU v6e, while the China side is represented mainly by Huawei’s Ascend 910B.. Topic tags: deepresearch, documentation, general web, education, user generated. Reference image context from search candidates: Reference image 1: visual subject "RAND's divisions conduct research on a uniquely broad front for clients around the globe. #### U.S. research divisions. U.S. and China flags on a computer chip on a motherboard. **" source context "China's AI Models Are Closing the Gap—but America's Real Advantage Lies Elsewhere | RAND" Reference image 2: visual subject "RAND's divisions conduct research on a uniquely broad front for clients a
openai.com
The global race for artificial‑intelligence computing is largely defined by competition between US and Chinese chip ecosystems. The United States dominates the high‑performance accelerator market with companies like Nvidia, AMD, and Google, while China is building alternatives through Huawei, Biren, and Cambricon.
Although both countries now produce specialized AI accelerators for training and inference, the competition differs not just in raw chip performance but also in memory bandwidth, manufacturing capability, software ecosystems, and large‑scale cluster design.
Below is a comprehensive comparison of major AI chips from each side using available documented specifications.
Major AI Chips in the US–China Competition
United States
AMD Instinct MI325X
Google TPU v6e (Trillium)
China
Huawei Ascend 910 series (including 910C)
Biren BR100 / BR104
Cambricon MLU370‑X8
These chips are primarily designed for large‑scale AI workloads such as training large language models (LLMs), inference services, and scientific computing.
Comprehensive Chip Comparison
Chip
Country / Company
Architecture / Node
Peak Compute
Memory
Memory Bandwidth
Power
Notable Features
AMD Instinct MI325X
US / AMD
CDNA3 accelerator architecture
~1.3 PFLOPS FP16 (2.6 PFLOPS FP8 with sparsity)
256GB HBM3E
~6 TB/s
up to ~750–1000W
Large HBM capacity designed for large‑model training and inference workloads.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
What is the short answer to "US vs China AI Chips: Architecture, Performance, and Ecosystem Compared"?
The US currently leads in documented high‑end AI accelerator performance and ecosystem scale, with chips like AMD’s MI325X (≈1.3 PFLOPS FP16 and 256GB HBM3E) and Google’s TPU v6e (918 TFLOPs bf16 per chip), while Chin...
What are the key points to validate first?
The US currently leads in documented high‑end AI accelerator performance and ecosystem scale, with chips like AMD’s MI325X (≈1.3 PFLOPS FP16 and 256GB HBM3E) and Google’s TPU v6e (918 TFLOPs bf16 per chip), while Chin... China’s strategy focuses on domestic alternatives such as Huawei’s Ascend 910‑series and Biren’s BR100 GPUs, designed for large‑model training and inference and often paired with large cluster systems to compete with...
What should I do next in practice?
The US advantage currently extends beyond chips to software ecosystems and manufacturing supply chains, while China’s progress centers on scaling domestic designs and building independent AI infrastructure.
One TensorCore with two matrix‑multiply units; designed for large TPU pods of up to 256 chips.
Huawei Ascend 910 (original)
China / Huawei
Da Vinci architecture, ~7 nm
~256 TFLOPS FP16
HBM memory
~1.2 TB/s bandwidth
~350W
Introduced in 2019 as Huawei’s flagship AI accelerator.
Huawei Ascend 910C
China / Huawei
Chiplet design (dual 910B dies), ~7 nm class
~800 TFLOPS FP16 estimate
up to ~96–128GB HBM
~3.2 TB/s
~310W
Designed to compete with Nvidia A100/H100‑class chips for AI training.
Biren BR100
China / Biren Technology
Dual‑die GPU, TSMC 7 nm CoWoS
256 TFLOPS FP32 / ~2,048 TOPS INT8
64GB HBM2E
up to ~2.3 TB/s
~550W
Chiplet‑style GPU with ~77B transistors targeting data‑center AI.
Biren BR104
China / Biren Technology
Single‑die GPU variant
~128 TFLOPS FP32
32GB HBM2E
~819 GB/s
~300W
Lower‑tier variant designed for PCIe accelerator cards.
Cambricon MLU370‑X8
China / Cambricon
MLUarch03 architecture, 7 nm
24 TFLOPS FP32 / 96 TFLOPS FP16 / 256 TOPS INT8
48GB LPDDR5
~614 GB/s
~250W
Supports multi‑card clusters via MLU‑Link interconnect.
Architecture and Compute Performance
US accelerators currently lead in documented raw compute throughput for large‑scale AI training. For example, AMD’s MI325X reaches roughly 1.3 PFLOPS FP16 performance, while Google’s TPU v6e delivers 918 TFLOPs bf16 per chip.
Chinese chips such as Huawei’s Ascend 910C aim to narrow this gap, with estimates of around 800 TFLOPS FP16 using a dual‑chiplet configuration derived from earlier Ascend chips.
Biren’s BR100 accelerator represents another Chinese attempt to compete at the high end, delivering up to 256 TFLOPS FP32 compute and around 2,048 TOPS INT8 performance in a multi‑die GPU design.
Cambricon’s MLU370‑X8 targets AI inference and training workloads with 256 TOPS INT8 and 96 TFLOPS FP16 compute capability.
Memory and Bandwidth
Modern AI models rely heavily on memory capacity and bandwidth, making HBM integration critical.
The AMD MI325X includes 256GB of HBM3E memory and about 6 TB/s bandwidth, one of the largest memory capacities among AI accelerators.
TPU v6e uses 32GB HBM with roughly 1.6 TB/s bandwidth, but is designed to scale to hundreds of interconnected chips in TPU pods.
Huawei Ascend 910C reportedly reaches about 3.2 TB/s memory bandwidth, supporting large‑scale model workloads.
Biren BR100 includes 64GB HBM2E memory and around 2.3 TB/s bandwidth.
Higher memory bandwidth helps accelerate matrix operations and large‑model training, which often move massive tensors across memory and compute units.
Interconnect and Scaling
Large AI models are rarely trained on a single chip. Instead, hundreds or thousands of accelerators are connected into clusters.
Google’s TPU v6e chips connect using a dedicated Inter‑Chip Interconnect (ICI) network designed for TPU pods containing up to 256 chips.
Cambricon’s MLU370‑X8 uses MLU‑Link, providing 200GB/s communication throughput between cards for multi‑GPU training.
Biren GPUs include high‑speed GPU‑to‑GPU interconnect links designed for multi‑accelerator clusters.
Cluster architecture is increasingly as important as the individual chip’s raw compute performance.
Manufacturing and Supply Chains
Manufacturing technology strongly influences performance and energy efficiency.
Many Chinese AI chips rely on external fabrication technologies. For example, Biren’s BR100 was built using TSMC’s 7‑nm process and advanced CoWoS packaging.
Huawei’s newer Ascend chips combine designs produced on SMIC’s 7‑nm‑class process with earlier wafers produced before US export restrictions.
In contrast, US‑designed chips typically rely on advanced foundries and supply chains capable of producing cutting‑edge nodes and packaging technologies.
Ecosystem and Software
Hardware performance alone does not determine success in AI computing.
US chips benefit from mature ecosystems such as CUDA (Nvidia), ROCm (AMD), and Google’s TPU software stack.
Huawei promotes the CANN (Compute Architecture for Neural Networks) framework for Ascend chips to build a domestic AI software ecosystem.
Developer tooling, frameworks, and cloud integration often determine which hardware is adopted for real‑world AI workloads.
What the Comparison Shows
Several patterns emerge from the current generation of AI accelerators:
Performance leadership: US accelerators currently show higher documented peak compute and memory capacities.
Domestic alternatives: China has built multiple competing chip lines—including Ascend, Biren GPUs, and Cambricon accelerators—to reduce dependence on US hardware.
Cluster strategy: Both sides increasingly compete through large AI systems and supercomputer‑scale clusters rather than individual chips.
The AI chip race is therefore not just a contest of transistor counts or TFLOPs—it is also a competition between ecosystems, manufacturing capability, and the ability to scale thousands of accelerators into efficient AI infrastructure.
As generative AI models continue to grow, these differences in architecture, memory systems, and software ecosystems will likely determine which platforms dominate future AI computing.
cset.georgetown.edu
Pushing the Limits: Huawei's AI Chip Tests U.S. Export Controls
One TensorCore with two matrix‑multiply units; designed for large TPU pods of up to 256 chips.
Huawei Ascend 910 (original)
China / Huawei
Da Vinci architecture, ~7 nm
~256 TFLOPS FP16
HBM memory
~1.2 TB/s bandwidth
~350W
Introduced in 2019 as Huawei’s flagship AI accelerator.
Huawei Ascend 910C
China / Huawei
Chiplet design (dual 910B dies), ~7 nm class
~800 TFLOPS FP16 estimate
up to ~96–128GB HBM
~3.2 TB/s
~310W
Designed to compete with Nvidia A100/H100‑class chips for AI training.
Biren BR100
China / Biren Technology
Dual‑die GPU, TSMC 7 nm CoWoS
256 TFLOPS FP32 / ~2,048 TOPS INT8
64GB HBM2E
up to ~2.3 TB/s
~550W
Chiplet‑style GPU with ~77B transistors targeting data‑center AI.
Biren BR104
China / Biren Technology
Single‑die GPU variant
~128 TFLOPS FP32
32GB HBM2E
~819 GB/s
~300W
Lower‑tier variant designed for PCIe accelerator cards.
Cambricon MLU370‑X8
China / Cambricon
MLUarch03 architecture, 7 nm
24 TFLOPS FP32 / 96 TFLOPS FP16 / 256 TOPS INT8
48GB LPDDR5
~614 GB/s
~250W
Supports multi‑card clusters via MLU‑Link interconnect.
Architecture and Compute Performance
US accelerators currently lead in documented raw compute throughput for large‑scale AI training. For example, AMD’s MI325X reaches roughly 1.3 PFLOPS FP16 performance, while Google’s TPU v6e delivers 918 TFLOPs bf16 per chip.
Chinese chips such as Huawei’s Ascend 910C aim to narrow this gap, with estimates of around 800 TFLOPS FP16 using a dual‑chiplet configuration derived from earlier Ascend chips.
Biren’s BR100 accelerator represents another Chinese attempt to compete at the high end, delivering up to 256 TFLOPS FP32 compute and around 2,048 TOPS INT8 performance in a multi‑die GPU design.
Cambricon’s MLU370‑X8 targets AI inference and training workloads with 256 TOPS INT8 and 96 TFLOPS FP16 compute capability.
Memory and Bandwidth
Modern AI models rely heavily on memory capacity and bandwidth, making HBM integration critical.
The AMD MI325X includes 256GB of HBM3E memory and about 6 TB/s bandwidth, one of the largest memory capacities among AI accelerators.
TPU v6e uses 32GB HBM with roughly 1.6 TB/s bandwidth, but is designed to scale to hundreds of interconnected chips in TPU pods.
Huawei Ascend 910C reportedly reaches about 3.2 TB/s memory bandwidth, supporting large‑scale model workloads.
Biren BR100 includes 64GB HBM2E memory and around 2.3 TB/s bandwidth.
Higher memory bandwidth helps accelerate matrix operations and large‑model training, which often move massive tensors across memory and compute units.
Interconnect and Scaling
Large AI models are rarely trained on a single chip. Instead, hundreds or thousands of accelerators are connected into clusters.
Google’s TPU v6e chips connect using a dedicated Inter‑Chip Interconnect (ICI) network designed for TPU pods containing up to 256 chips.
Cambricon’s MLU370‑X8 uses MLU‑Link, providing 200GB/s communication throughput between cards for multi‑GPU training.
Biren GPUs include high‑speed GPU‑to‑GPU interconnect links designed for multi‑accelerator clusters.
Cluster architecture is increasingly as important as the individual chip’s raw compute performance.
Manufacturing and Supply Chains
Manufacturing technology strongly influences performance and energy efficiency.
Many Chinese AI chips rely on external fabrication technologies. For example, Biren’s BR100 was built using TSMC’s 7‑nm process and advanced CoWoS packaging.
Huawei’s newer Ascend chips combine designs produced on SMIC’s 7‑nm‑class process with earlier wafers produced before US export restrictions.
In contrast, US‑designed chips typically rely on advanced foundries and supply chains capable of producing cutting‑edge nodes and packaging technologies.
Ecosystem and Software
Hardware performance alone does not determine success in AI computing.
US chips benefit from mature ecosystems such as CUDA (Nvidia), ROCm (AMD), and Google’s TPU software stack.
Huawei promotes the CANN (Compute Architecture for Neural Networks) framework for Ascend chips to build a domestic AI software ecosystem.
Developer tooling, frameworks, and cloud integration often determine which hardware is adopted for real‑world AI workloads.
What the Comparison Shows
Several patterns emerge from the current generation of AI accelerators:
Performance leadership: US accelerators currently show higher documented peak compute and memory capacities.
Domestic alternatives: China has built multiple competing chip lines—including Ascend, Biren GPUs, and Cambricon accelerators—to reduce dependence on US hardware.
Cluster strategy: Both sides increasingly compete through large AI systems and supercomputer‑scale clusters rather than individual chips.
The AI chip race is therefore not just a contest of transistor counts or TFLOPs—it is also a competition between ecosystems, manufacturing capability, and the ability to scale thousands of accelerators into efficient AI infrastructure.
As generative AI models continue to grow, these differences in architecture, memory systems, and software ecosystems will likely determine which platforms dominate future AI computing.
cset.georgetown.edu
Pushing the Limits: Huawei's AI Chip Tests U.S. Export Controls
