Kwame Zaire is a towering figure in the world of manufacturing and electronics, known for his deep technical understanding of the hardware that powers our modern digital infrastructure. With an extensive background in production management and a focus on the intersection of quality and safety, he has spent years analyzing the intricate dance of global supply chains. As we sit down today, the conversation centers on the seismic shifts occurring in the artificial intelligence chip market, where the landscape is being radically redrawn by geopolitical maneuvers and a fierce drive for regional self-sufficiency. We explore the transition of dominance from Western giants to local innovators, the technical hurdles of high-end lithography, and how software developers are adapting to a world where the hardware they once relied upon is increasingly out of reach.
Nvidia’s market share in China has plummeted from a near-monopoly of 95% to a projected 8% this year. How do you interpret such a massive structural shift in a market that was once so firmly under their control?
The numbers we are seeing are nothing short of a total market transformation that would have seemed impossible just five years ago. When you consider that Nvidia sat comfortably with a 95% market share, they weren’t just a leader; they were the entire ecosystem for Chinese AI development. To see that projected to crater to a mere 8% this year suggests a decoupling that is as much about policy as it is about performance. You can almost feel the tension in the manufacturing hubs when you talk about this; it’s the sound of thousands of server racks being filled with domestic silicon rather than the Santa Clara exports they once craved. This isn’t just a dip in sales; it’s a fundamental pivot where Bernstein estimates that Nvidia’s 40% share in 2025 is already being aggressively cannibalized by Huawei, which is expected to capture 50% of that same market.
Huawei has emerged as a formidable domestic rival, with its Ascend 950 series being compared to Nvidia’s flagship H200. From a technical and manufacturing standpoint, how impressive is it that they have reached this level of parity?
It is a staggering achievement when you consider the hurdles placed in their path, specifically the lack of access to the most advanced chipmaking machinery. The Ascend 950 series is now seen by industry analysts as roughly comparable to Nvidia’s H200, which is a powerhouse chip that represents the cutting edge of what the West can produce. In the cleanrooms and fabrication plants, the “scent” of innovation is unmistakable as Huawei’s semiconductor business, led by He Tingbo, finds “pretty good solutions” to bypass the restrictions that were supposed to hold them back. Even though Huawei’s annual revenue of $126 billion is still dwarfed by Nvidia’s $216 billion, the fact that they are rolling out AI computing clusters that combine the power of thousands of Chinese-made chips shows they are no longer just playing catch-up. They are building a parallel reality where they can walk just as fast, if not faster, than their global rivals.
The U.S. export controls were intended to safeguard national security, yet they seem to have accelerated China’s drive for self-sufficiency. How are these restrictions changing the global supply chain for high-end AI hardware?
We are witnessing a classic case of necessity being the mother of invention, though the birthing process for this new supply chain is incredibly complex and fraught with friction. The global chain is so interconnected that no single nation can truly go it alone; Nvidia designs the chips, but they rely on Dutch company ASML for the extreme ultraviolet lithography, or EUV, machines, which themselves are packed with American components. Because China is barred from buying these EUV machines, they have been forced to innovate within a closed loop, relying on Taiwan’s TSMC to make a large share of the top-tier chips before the doors were shut. The frustration is palpable among Chinese researchers who still hunger for that H200 performance, leading to cases of smuggling to circumvent the controls. However, the long-term result is a hardening of the Chinese domestic industry, as experts like He Hui from Omdia point out that the country now firmly believes in its own supply capabilities and is no longer willing to rely on a reprieve that could be revoked at any moment.
With companies like DeepSeek adapting their V4 AI models specifically for Huawei’s Ascend chips, what does this tell us about the future of software-hardware integration in the region?
This is perhaps the most significant “boots on the ground” indicator that the shift is permanent because once the software layer is optimized for a specific hardware architecture, the “moat” around that hardware becomes much deeper. DeepSeek’s latest V4 model, which rolled out in April, isn’t just a generic update; it was specifically adapted for Huawei’s advanced Ascend chips, representing a massive collaborative effort between software engineers and hardware designers. This kind of synergy is what made Nvidia so dominant with their CUDA platform, and seeing it replicated with Huawei and DeepSeek suggests that the domestic ecosystem is maturing rapidly. While analysts like Phelix Lee don’t expect an “abrupt switch” overnight, the groundwork is being laid for a future where the “brain” of the AI and the “muscle” of the silicon are entirely homegrown. It’s a specialized, localized evolution that makes the H20 chips—Nvidia’s stripped-down versions designed to bypass restrictions—look less and less attractive to Chinese firms who want the full, unbridled power of their own localized clusters.
Despite losing significant ground in China, Nvidia’s global sales are still surging, with revenue expected to hit $91 billion this quarter. How long can they sustain this growth while being essentially locked out of one of the world’s largest tech markets?
Nvidia is currently riding a global wave of demand that is so massive it effectively masks the pain of the Chinese market loss. Their jump from $82 billion in the previous quarter to an expected $91 billion—excluding any data center revenue from China—is a testament to how the rest of the world is scrambling for AI hardware. However, you can’t ignore the historical context; Jensen Huang has been in China for 30 years, and losing that 95% market share is a bitter pill to swallow, regardless of how well the rest of the world is buying. The “zhajiangmian” noodles Huang ate in Beijing might have been a hit with the locals, but the business reality is that the U.S. has lost its edge in that specific geography as Chinese competitors become “giants.” Nvidia is still the designer of the world’s most powerful AI chips, but they are now operating in a world that is splitting in two, and the $216 billion in annual revenue they’ve achieved will eventually face a ceiling if they cannot find a way to compete in the 170 countries and regions where Huawei is also expanding its footprint.
As China scales its production capacity and their pricing becomes more competitive, what is your forecast for the global AI hardware market over the next five years?
My forecast is that we are moving toward a bipolar global market where the price-to-performance ratio of Chinese hardware will begin to dominate emerging markets, particularly in Southeast Asia. As China’s strategy of technological self-sufficiency matures, they will transition from being a captive consumer to a major exporter, utilizing their $126 billion revenue base to subsidize aggressive expansion into regions that are less sensitive to U.S. export controls. We will see a “pricing war” in AI infrastructure that mirrors what we’ve seen in telecommunications equipment, where Huawei’s mission of bringing digital connectivity to every home will extend to bringing AI clusters to every developing data center. Nvidia will likely maintain its lead in the most advanced research and training chips for the foreseeable future, but the “bread and butter” of global AI computing will increasingly be fought over by these two distinct, and now very capable, industrial ecosystems. The smell of ozone in the data centers of tomorrow will be coming from a much more diverse array of silicon than we ever anticipated.
