Nvidia CEO Jensen Huang has been unusually blunt lately, warning that export controls will backfire and Huawei will replace Nvidia in China if Washington continues restricting chip exports. But if you view this as mere drama-bait, then boy, you are in for a rude awakening.
Because, while America spent the last two years congratulating itself on “containing” China’s AI development, something remarkable happened. The sanctions did nothing to slow China down. Worse, Huawei actually came out of the ashes of its sanctions to shake up the global AI landscape.
America Unintentionally Fans the Storm
Let’s rewind to understand how we got here. Starting in 2019, the U.S. systematically dismantled Huawei’s global business through a series of escalating sanctions. First came the telecom equipment bans, then smartphone restrictions that killed their Android access, and finally the semiconductor export controls that cut them off from advanced chips entirely. All for the ultimate goal of crippling Huawei and sending a message to China about technological dependence.
By 2022, Huawei’s smartphone revenue had collapsed, its international telecom business was in freefall, and it seemed to be slowly fading into irrelevance. But things never ended there. In reality, ol’ Uncle Sam just took a wounded tech giant with nothing left to lose and accidentally created the perfect startup conditions.
At this point, Huawei is no longer in competition mode with the likes of Samsung and Apple anymore. The impetus for stability is no longer there. So to focus on survival, they instead made a 180 to innovate radically. When those chips got banned, they didn’t just disappear, of course, and they were immediately forced to look for alternatives.
The next turn of events for the company then transformed it into a national AI champion overnight. Beijing has been providing state support, urging local tech companies to buy more Huawei AI chips and shift away from the $3.3 trillion chipmaker just on the other side of the ocean. China may have historically represented 20-25% of Nvidia’s annual sales. But now, Huawei had unlimited resources, national prestige on the line, and a government actively pushing customers in their direction of the AI revolution.
Greatness, At Any (Power) Cost
While already having initial offers in lower-end AI hardware, Huawei’s industrial comeback came to its AI fever pitch in September 2024. This was when Huawei unveiled its CloudMatrix 384, an AI system that was outperforming the best AI hardware available prior to the release of the GB300 later this year.
| Metric | Huawei CloudMatrix 384 | Nvidia GB200 NVL72 | Advantage |
| BF16 Compute Performance | ~300 PetaFLOPS | ~150 PetaFLOPS | 2x faster |
| HBM Memory Capacity | 49.2 TB | 13.8 TB | 3.6x more memory |
| Memory Bandwidth | 1,229 TB/s | 576 TB/s | 2.1x more bandwidth |
. Companion apps are a real inference sink in China. They sit next to short video and live commerce as always-on categories, but their stack is heavier per user. A premium service like porn AI chat is not one model, it is a bundle per session: speech recognition to hear, a large language model to converse, text-to-speech to reply, an avatar or lip-sync engine to render, plus safety and billing running continuously. Long chats keep context hot, and users expect zero lag between turns. High HBM capacity and internal bandwidth let more of that pipeline live on a single node, keep longer histories resident, and raise the number of concurrent sessions per rack.
There is also a go-local incentive as operators in China favor domestic vendors for latency, data residency, and procurement stability, which puts throughput per rack and cost per token ahead of a single chip’s efficiency score. A system like CloudMatrix 384 can batch long conversations, stream audio and video, and run classification and watermark checks on the same box. That is the kind of workload where Huawei wins footprint even if performance per watt trails in Western data centers.
The obvious caveat, as many sources instantly point out, is the whopping x2.3 lower performance per watt compared to the GB200. But here’s the thing about China: electricity prices declined from $90.70 per MWh in 2022 to $56 per MWh in some regions by 2025. As such, the company’s (and by extension, the government’s) leeway for ambitious projects becomes much wider. This is especially true when they are at the cutting edge of current AI technology. Besides, even if costs weren’t cheaper, we still doubt that China would have cared for the power inefficiency either way.
Oh yes, and this isn’t just some fancy prototype, as Huawei is well into the manufacturing stage already. In fact, the company is slated to ship one hundred thousand 384 Asccend 910C chips in 2025. The trick was using TSMC workarounds via third-party entities like Sophgo.
Is Nvidia Losing the Initiative?
Well, for one thing, China is “gone” forever. Even if sanctions lifted tomorrow, Chinese AI companies aren’t coming back if they see an absolute advantage for Huawei’s offerings down the line. And technically, that would be 25% of Nvidia’s revenue that’s never returning. Huawei’s leadership has publicly expressed confidence that China has overcome its technology dependence issues and no longer needs Western suppliers. Which is a pretty direct way of declaring that they no longer need Nvidia as much.
Besides, it wasn’t even a monopoly in the first place. Huawei’s offerings serve as a one-to-one credible alternative to what Nvidia can provide. And with competition, alternatives come changes in pricing power, competitive dynamics, and customer leverage. So does this foster supply chain anxiety? It sure does feel like it. Tech companies worldwide just watched China successfully build domestic alternatives under the harshest sanctions in history. That’s going to influence supplier choices for years.
That being said, even with another direct competitor coming its way, this is in no way an existential threat to Nvidia. The AI market is exploding, CUDA still has massive switching costs, and Huawei’s efficiency problems would certainly be a dealbreaker for a few international clients. Nvidia will likely continue dominating Western markets while Huawei starts scooping up local shares. Still painful for Nvidia’s growth within the AI craze, but definitely not at the same level as Intel is experiencing with AMD right now.
Huawei’s Splash into the AI Hardware Craze
One can’t help but feel rather unsympathetic to Nvidia, as Jensen Huang saw this coming all along. During his Beijing visit, he called Huawei “a formidable competitor” and emphasized that no one should underestimate China’s manufacturing capabilities. He especially noted Huawei’s advantage in having built its own chips and entire cloud system that allows it to “go to market directly.”
But market dominance always has conditions attached. Nvidia’s supremacy was built on global access and the absence of credible alternatives. Introduce a relatively equal competitor, and suddenly, a lot of the technological logistics can be effectively bypassed.
And yes, this all boils down to the AI hardware market simply becoming more complex than it already is. For Nvidia, this is a permanent recalibration rather than a temporary setback. For the industry as a whole, it’s yet another cycle that proves that technological superiority never guarantees market control.