AMD–Chinese joint venture
Based on Wikipedia: AMD–Chinese joint venture
In 2016, a quiet but monumental shift occurred in the global semiconductor landscape that would eventually ripple through geopolitics, technology markets, and the very architecture of computing power. The American chip giant Advanced Micro Devices (AMD) struck a deal with Chinese partners to license its x86 CPU designs, aiming to build processors specifically for the Chinese market. This was not merely a business transaction; it was an acknowledgment of a stark reality: despite China's massive manufacturing prowess in textiles, electronics assembly, and later electric vehicles, the nation had failed to produce a competitive central processing unit based on its own indigenous technology. The microchip industry demanded a depth of research and development that China simply did not possess at the time, creating a vacuum that only American innovation could fill. Unlike the established and lucrative chip ecosystem in Taiwan, or the natural manufacturing dominance China held in other sectors, building a CPU from scratch remained an insurmountable hurdle for Beijing. To bridge this gap, AMD navigated a complex legal and geopolitical labyrinth to create a joint venture that would eventually become known as Hygon Information Technology.
The timing of this agreement is critical to understanding its significance, particularly when viewed through the lens of subsequent history. The deal was negotiated and finalized well before the eruption of the 2018 trade war between the United States and China, and years before the global chip shortage that would cripple industries in the early 2020s. In that pre-conflict era, the logic seemed straightforward: AMD gained access to a massive, hungry market without having to build factories there, while China hoped to leapfrog decades of R&D by licensing proven architecture. The United States government, initially seeing an economic opportunity, granted permission from both the Department of Defense and the Department of Commerce for AMD to export its Zen 1 core design to China. This was a rare concession, given that x86 technology is considered the bedrock of global computing security and performance. To comply with American legal restrictions while still facilitating the deal, AMD constructed a corporate edifice of staggering complexity, setting up multiple distinct entities to manage the licensing flow.
At the heart of this structure was Tianjin Haiguang Advanced Technology Investment Co. Ltd., known by the acronym THATIC. This overarching joint venture was a peculiar hybrid of American and Chinese interests. It was owned jointly by AMD and a consortium of both public and private Chinese companies, including the prestigious Chinese Academy of Sciences. However, reports suggested that despite the diverse ownership structure, AMD itself held the major share, retaining significant control over the intellectual property being transferred. From THATIC, two operational arms were spun off to execute the vision: Haiguang Microelectronics Co. Ltd. (HMC) and Chengdu Haiguang Integrated Circuit Design Co. Ltd., which traded under the name Hygon. The division of labor was precise. HMC held the local intellectual property rights and subcontracted the actual manufacturing of the chips, effectively acting as the bridge between design and silicon fabrication. Meanwhile, Hygon took on the role of designer, marketer, and seller. It was Hygon that would present the processors to the world, selling them under a Chinese banner while relying entirely on American blueprints.
The announcement in 2016 set off a chain reaction that culminated in the release of the first processor in 2018. This chip, named the Dhyana system-on-a-chip, was a marvel of engineering replication. It was a variant of the AMD Epyc series, designed for server environments. The technical fidelity was so high that industry observers noted there was "little to no differentiation between the chips." When analysts examined the code supporting these processors in the Linux kernel, they found something startling: less than 200 lines of new code were required to make the Dhyana run. This figure stood as a testament to the fact that the Hygon processor was not an independent invention but rather a licensed adaptation of existing Western technology. In essence, it was a re-branded Zen CPU tailored for the Chinese server market, allowing China to claim domestic production capabilities while relying on American core architectures.
By 2019, the trajectory of this partnership began to curve sharply downward as geopolitical tectonic plates shifted beneath the industry. At the Computex trade show in June 2019, AMD CEO Lisa Su addressed the future of the collaboration with unambiguous clarity. She confirmed that the license granted to Hygon was strictly limited to the original Zen architecture. The promise of continued access to newer, more powerful generations of chips ended there; the license would not be extended to Zen 2 or any subsequent iterations. This limitation signaled that the era of open technological transfer between the US and China in high-performance computing was drawing to a close. Just weeks later, on June 24, 2019, the United States government took decisive action. The Department of Commerce placed one of the parent companies involved in the venture onto its Export Control Entity List.
The implications of being added to the Entity List were severe and immediate. It effectively banned any further technology transfers from AMD to Hygon or its associated entities. While the ban did not necessarily shut down existing operations overnight, it cast a long shadow over the future of the joint venture. The restriction meant that Hygon could no longer expect updates, patches, or new architectural designs from their American partner. Analysts at Anandtech were quick to assess the situation, predicting that further AMD involvement would become minimal to non-existent due to the ban. The strategy of "leapfrogging" through licensing had hit a hard wall erected by national security concerns and trade hostilities. The chip that China hoped would be the foundation of its digital independence was suddenly frozen in time at the Zen 1 level, a relic of a pre-trade war world.
The technical reality of the Hygon Dhyana processor, when scrutinized under the microscope of independent testing in 2020, revealed the nuances of this "indigenous" success story. While the chip was largely identical to its AMD counterpart, subtle but significant differences emerged in performance and functionality. Testing suggested that integer performance—the kind of calculation used for general computing tasks—was essentially identical to the equivalent Zen 1 processor. However, when it came to floating-point operations, which are critical for scientific modeling and complex simulations, as well as specific security instructions known as RDRAND and RDSEED, performance was noticeably reduced. These were not minor tweaks; they represented a degradation in capability that could impact high-performance computing workloads.
Furthermore, the cryptographic landscape of the processor had been altered to suit Chinese regulatory standards. The standard cryptography extensions found in Western processors, such as AES for data encryption, were replaced by Chinese national algorithms known as SM2, SM3, and SM4. According to technical analysis, these new algorithms bore a strong resemblance to existing international standards: SM2 was similar to ECC-based algorithms, SM3 mirrored SHA-256 for hashing, and SM4 corresponded to AES-128. This substitution ensured that the chips complied with Chinese security protocols but further distanced them from the global ecosystem of software compatibility. Additionally, support for AVX and AVX2 instruction sets, which accelerate multimedia and vector processing, appeared to be disabled. While some speculated this was a deliberate strategic choice to limit performance, researchers suspected it might have been an unintended bug in the porting process, highlighting the fragility of taking a complex chip design and attempting to modify it for a different market without full access to the source code or future support from the original architect.
The story of the AMD–Chinese joint venture is not just about chips; it is a case study in the limits of technological transfer in an era of rising nationalism. It highlights the fundamental challenge China faces: while the country has excelled in scaling manufacturing, logistics, and assembly—evident in its recent electric vehicle boom—the creation of high-end microprocessors requires a depth of fundamental research that cannot be bought or licensed indefinitely. The joint venture was a bridge built on sand, relying on the goodwill of American regulators who eventually withdrew their support as strategic competition intensified. Unlike other industries where China has naturally excelled through sheer scale and state investment, the microchip industry demands a continuous cycle of innovation that is difficult to sustain once external knowledge flows are cut off.
The parallel can be drawn to another joint venture in the same sector: Zhaoxin, which was supported by VIA Technologies. Like Hygon, Zhaoxin represents an attempt to localize x86 technology through partnership with Western firms. Yet, both ventures serve as proof of concept for a single truth: China has been unable to produce a CPU based on its own technology without significant foreign assistance. The AMD deal provided that assistance, but only temporarily. When the geopolitical climate soured, the lifeline was severed, leaving Hygon with a product that, while functional, was technologically stagnant compared to the rapid pace of global innovation.
The human and economic stakes of this technological dependency are immense. For the Chinese government, the inability to produce sovereign silicon represents a critical vulnerability in its digital infrastructure. If a conflict were to arise, or if trade restrictions tightened further, the supply of high-performance computing power could be threatened. The Dhyana processor was an attempt to mitigate this risk, to build a domestic fortress of computing capability. Yet, by relying on a design that was effectively frozen in 2016 and cut off from future updates in 2019, China found itself with a solution that solved yesterday's problems but left it ill-equipped for tomorrow's challenges. The "less than 200 lines" of new kernel code stands as a metaphor for the entire endeavor: impressive in its replication, but limited by its lack of original foundation.
As we look at the broader context of China's technological ascent, the contrast is stark. In the realm of electric vehicles, China has built a dominant industry from local battery chemistry, supply chains, and manufacturing scale. It was a success story of "local" innovation driving global markets. But in semiconductors, the narrative remains one of dependency. The AMD joint venture was the most ambitious attempt to replicate that EV success story in the chip world, and its ultimate stalling illustrates why microchips are different. They cannot be scaled simply by building more factories; they require the relentless, often invisible, work of physicists and engineers pushing the boundaries of what silicon can do. That work is hard to license.
The legacy of this joint venture extends beyond the Dhyana chip itself. It serves as a historical marker, capturing a specific moment in time when the US and China were still willing to engage in deep technological integration despite growing tensions. The agreement was a product of an era that believed trade would eventually bind nations together, that economic interdependence would act as a shield against conflict. The subsequent banning of Hygon from receiving further technology transfers shattered that illusion. It marked the transition from a world of globalized supply chains to one of fractured technological blocs. For AMD, it was a difficult decision to walk away from a lucrative market, but one dictated by national security directives. For China, it was a wake-up call that the path to technological sovereignty would be longer and more arduous than anticipated.
The technical modifications made to the Hygon processor, such as the swapping of cryptographic algorithms and the disabling of certain instruction sets, were perhaps unavoidable compromises. They reflect the reality of adapting foreign technology for domestic use in a hostile environment. The reduced floating-point performance and the suspicions around AVX/AVX2 functionality suggest that without full support from the original designer, maintaining peak efficiency is nearly impossible. These are not just technical glitches; they are symptoms of a deeper disconnect between the design intent and the final product. When you take a chip designed for a global market and try to force it into a siloed ecosystem, friction is inevitable. The "bug" that may have disabled AVX/AVX2 is a fitting symbol: an unintended consequence of trying to do something for which the original tools and support were no longer available.
In the end, the AMD–Chinese joint venture stands as a complex chapter in the history of technology. It was a bold experiment that demonstrated both the power of collaboration and the fragility of trust in a geopolitical crisis. The Dhyana processor remains on servers across China, doing its job, but it is a fossilized version of American innovation. It represents a ceiling that Hygon cannot break without new breakthroughs from within. As the global chip industry continues to race toward nanometer-scale precision and quantum computing frontiers, a license for a 2016 Zen architecture design becomes increasingly irrelevant. The venture proves that while you can build a factory in China, you cannot simply manufacture ingenuity through a joint agreement. The gap between assembling chips and designing them remains the widest chasm in the industry, one that no amount of state funding or licensing deals has yet been able to fully bridge.
The story is also a reminder of the human element behind these corporate structures. Behind the acronyms THATIC, HMC, and Hygon are thousands of engineers, researchers, and managers who believed they were building a new future for Chinese technology. They worked with the tools given to them, trying to make the Dhyana perform as well as possible within its constraints. The limitations on their work—the frozen architecture, the missing instructions—were not failures of their skill, but rather the result of forces far beyond their control. Their experience mirrors that of many in the global tech sector who found themselves caught in the crossfire of a trade war they did not start and could not end.
As we reflect on this history, it becomes clear that the path to self-sufficiency in semiconductors is not paved with joint ventures alone. It requires a fundamental rethinking of how innovation happens. China's success in electric vehicles was built on a foundation of local R&D and supply chain control, elements that were largely absent in the semiconductor deal with AMD. The lesson for the future is that true technological independence cannot be imported; it must be cultivated from the ground up, through decades of investment in education, basic research, and the patient accumulation of knowledge. The Hygon experiment was a necessary step in that journey, but it also serves as a clear boundary marker: the point where collaboration ends and isolation begins.
The fate of the AMD–Chinese joint venture is now sealed by history. It will be remembered not for the longevity of its products, but for the brief window of optimism it represented before the walls came down. In the annals of semiconductor history, the Dhyana processor will likely appear as a footnote—a curious artifact of an era when the world was still connected enough to share its most valuable secrets. Yet, for those who understand the stakes of computing power in the 21st century, it is a story with profound implications. It illustrates that in the race for technological supremacy, no amount of licensing can substitute for original innovation. And as the US and China continue to drift further apart, the silence following the end of this partnership speaks louder than the processors themselves ever did.
The future of Chinese computing will depend on whether it can finally generate its own "Zen" moments—breakthroughs that are not licensed, but born from within. Until then, the joint venture remains a testament to what was possible when walls were low, and a warning of what happens when they rise. The chips may still run, but the engine driving them has stopped receiving fuel. In a world where speed is everything, standing still is the same as falling behind. And for a nation with ambitions to lead the digital age, that is a position it cannot afford to occupy for long.
The narrative of Hygon and AMD is more than a business case; it is a geopolitical allegory. It shows how quickly the ground can shift beneath the most carefully laid plans. The deal was signed in 2016, a time when the future seemed open. By 2019, the gates were closed. In those three years, the world changed forever, and the semiconductor industry became a battleground rather than a marketplace. The Dhyana chip is a relic of the past, a reminder that in the high-stakes game of global technology, the rules are written not just by engineers, but by politicians. And when the political winds change, even the most advanced silicon can become obsolete overnight.