The global tech race is messy. If you’ve been following the drama surrounding Huawei AI chips US China relations, you know it’s less of a corporate rivalry and more of a high-stakes geopolitical chess match where the board keeps changing every single week. Huawei was once just a smartphone giant. Now, it’s the centerpiece of a massive struggle for artificial intelligence supremacy.
The US wants to keep the most advanced semiconductors out of Chinese hands. China wants to build its own, free from Western influence. In the middle of this, Huawei is trying to prove that it can thrive while being cut off from the world’s most advanced chip-making equipment. Honestly, it’s a wild story of survival, secret supply chains, and some seriously impressive engineering that caught Washington off guard.
The Ascendancy of the Ascend 910B
For a long time, Nvidia was the only name that mattered in AI. Their H100 GPUs are the gold standard for training large language models like GPT-4. But when the US Department of Commerce tightened export controls, it created a massive vacuum in the Chinese market. Huawei stepped into that gap with the Ascend 910B.
It’s not just a "knock-off." Reports from Chinese tech firms like Baidu and iFlytek suggest that the 910B is roughly comparable to Nvidia’s older A100 in terms of raw computing power. That’s a big deal. While it might not beat the newest Blackwell architecture Nvidia just released, it’s "good enough" for many Chinese companies that can no longer buy the top-tier American gear.
📖 Related: Mark Zuckerberg Trump Meeting: What Really Happened at Mar-a-Lago
Baidu reportedly ordered 1,600 of these chips back in 2023. That might sound like a small number compared to the tens of thousands Nvidia sells to Meta or Microsoft, but it signals a shift. Chinese tech giants are no longer just looking at Huawei as a backup plan. They’re looking at it as their primary path forward.
SMIC and the 7nm Mystery
How is Huawei even making these things? That’s the question that keeps US regulators up at night.
To make a cutting-edge AI chip, you need Extreme Ultraviolet (EUV) lithography machines. These are made exclusively by a Dutch company called ASML. Because of US pressure, ASML isn’t allowed to sell these machines to China. So, logically, Huawei shouldn’t be able to produce anything below 14nm or 10nm.
Yet, here we are.
Working with SMIC (Semiconductor Manufacturing International Corporation), Huawei managed to produce 7nm chips for their Mate 60 Pro phones and their AI accelerators. They basically used older DUV (Deep Ultraviolet) machines and pushed them to their absolute physical limits. It’s expensive. The "yield rate"—the percentage of chips that actually work—is rumored to be much lower than what TSMC achieves in Taiwan. But for the Chinese government, the cost doesn't matter as much as the sovereignty.
They are brute-forcing their way into the high-end chip market. It's fascinating and terrifying for competitors all at once.
Why the US-China Deadlock is Getting Weird
The tension over Huawei AI chips US China policy has led to some strange cat-and-mouse games. The US sets a performance ceiling for chips that can be exported. Nvidia then designs "lite" versions of their chips (like the H20) that sit just below that ceiling so they can keep selling to China. Then, the US moves the ceiling again.
Commerce Secretary Gina Raimondo has been very clear: the goal is to deny China the compute power needed for advanced military AI. But there's a flip side. Every time a new restriction is added, it gives Huawei more customers. If Chinese companies can't buy Nvidia, they have to buy Huawei.
It’s an accidental stimulus package for Huawei’s R&D department.
📖 Related: 2d shapes and 3d shapes: Why Your Brain Struggles to See the Difference
The Software Hurdle: CANN vs CUDA
Hardware is only half the battle. If you ask any AI developer why they love Nvidia, they won't just talk about the GPU. They'll talk about CUDA.
CUDA is the software platform that makes it easy to program GPUs for AI tasks. It has a massive ecosystem. Millions of developers have spent a decade building tools for it. Huawei has its own version called CANN (Compute Architecture for Neural Networks).
Switching from CUDA to CANN is a huge pain. It’s like trying to rewrite a whole library from English into a language that only a few people speak fluently. However, Huawei is throwing thousands of engineers at this problem. They are helping Chinese companies migrate their code, bit by bit. They're making it easier. It's a slow grind, but it's happening.
What Most People Get Wrong About the Sanctions
A common misconception is that the sanctions have "stopped" Huawei. They haven't. They've just changed Huawei's DNA.
Before 2019, Huawei was a global integrator. They bought components from everywhere—US, Japan, Germany—and put them together. Now, they are becoming vertically integrated. They are designing the chips, the software, the cloud infrastructure, and even the manufacturing tools.
Is it working?
- Their revenue is rebounding.
- Their "HarmonyOS" is replacing Android in China.
- Their AI cloud services are growing.
We also have to talk about the "teardowns." When firms like TechInsights rip open Huawei’s latest hardware, they often find components that shouldn't be there. Sometimes it's old stock. Sometimes it's sourced through third-party distributors in countries that don't have the same export rules. It's a leaky system. No matter how many "entity lists" the US creates, silicon finds a way.
The 2026 Outlook: 5nm and Beyond
The next big milestone is 5nm. If SMIC and Huawei can successfully mass-produce 5nm AI chips without ASML’s top-tier machines, the US export controls will be viewed by many as a failure.
There are rumors that Huawei is experimenting with "chiplets"—basically stitching smaller chips together to act like one big, powerful chip. It’s a clever workaround for manufacturing limitations. Instead of trying to make one giant, perfect 5nm chip, you make four smaller 7nm chips and link them with high-speed interconnects.
It’s not as efficient, and it runs hotter, but in the world of AI, raw power often trumps efficiency.
How This Affects the Global Market
You might think this is just a "China problem," but it affects everyone.
- Supply Chain Splitting: We are seeing the "splinternet" of hardware. One ecosystem for the West (Nvidia/AMD/Intel) and one for China (Huawei/Baidu/Biren).
- Pricing: If Huawei can produce AI chips at scale, it provides a price floor. Nvidia can't charge whatever they want if there's a viable, cheaper alternative, even if it's only available in certain regions.
- Innovation Pace: The fear of being left behind is making the US invest billions into the CHIPS Act. Competition, even when born out of trade wars, tends to speed up technical breakthroughs.
Honestly, the whole situation is a bit of a tragedy for global collaboration. AI is a field that usually thrives on open research and shared papers. Now, the most important part of the AI stack—the silicon it runs on—is becoming a state secret.
Actionable Insights for Tech Observers
If you’re trying to navigate this landscape, whether as an investor, a developer, or just someone interested in the future of tech, here’s what you actually need to watch. Forget the political grandstanding for a second and look at the technical markers.
Watch the "Interconnect" technology. AI doesn't happen on one chip; it happens on thousands of chips talking to each other. Huawei’s ability to build high-speed networking (since they are a telecommunications company at heart) is their secret weapon. If their "Ascend" clusters can communicate faster than Nvidia's clusters, they can win even with slower individual chips.
📖 Related: Biomedical Engineering Jobs Pay: What Most People Get Wrong
Monitor the Talent Flow. Keep an eye on where top semiconductor engineers are going. Huawei has been aggressively hiring from top Western universities and even poaching talent from competitors by offering massive salaries.
Keep an eye on "Legacy" nodes. The US is now looking at restricting older chip-making tech too. If that happens, it won't just hit AI; it will hit cars, toasters, and medical devices. That’s where the trade war could really start to hurt the average consumer.
Diversify your software knowledge. For developers, if you're working in a global context, start looking at "hardware-agnostic" AI frameworks. Tools like Triton (from OpenAI) allow code to run on different types of chips more easily. Betting everything on one proprietary platform like CUDA is becoming riskier as the world fragments.
The story of Huawei AI chips US China rivalry isn't over. Not by a long shot. We're just in the middle chapters where both sides are doubling down. Huawei is no longer just "the Chinese company that makes phones." They are becoming a full-stack AI powerhouse, driven by the very restrictions meant to slow them down.
Key Takeaways for the Future
- Sovereignty over Efficiency: China has signaled it will pay any price to have its own chip supply.
- Huawei is the National Champion: All roads in Chinese AI eventually lead back to Huawei’s Ascend architecture.
- Software is the True Moat: The hardware gap is closing, but the software ecosystem gap remains wide and will take years to bridge.
- Geopolitics is the New Moore’s Law: The speed of AI development in the next decade will be dictated as much by policy papers in D.C. and Beijing as by physics in a lab.
The situation is fluid. One week it's a new export ban, the next it's a breakthrough in SMIC's production line. The only certainty is that the "Silicon Curtain" is real, and it’s changing how the world thinks about computing.