China has tried several approaches to building its own processors. Loongson eventually created an entirely new instruction set called LoongArch. Other companies have adopted Arm or RISC-V. Hygon took a considerably stranger route.
It started with AMD Zen.
In 2016, AMD licensed processor technology to a Chinese joint venture connected to what would become Hygon’s CPU business. The resulting processors were x86-compatible and closely related to AMD’s original Zen architecture. Three years later, U.S. export restrictions effectively stopped additional AMD technology transfers. That should have left Hygon stuck with a technological snapshot from 2016. Instead, the company kept developing it.
Today, Hygon is several generations removed from its original Dhyana processors. Its latest C86 chips support modern technologies such as DDR5 and PCIe 5.0, while upcoming designs reportedly introduce unusual features including four-way simultaneous multithreading.
So how did China end up with its own x86 processor, and how much AMD is still hiding underneath?
Definitely related: YMTC Explained: Can China Really Overtake Samsung in NAND?
It Started With an AMD That Needed Money
To understand Hygon, we have to return to AMD in the mid-2010s. This was before Ryzen transformed the company. AMD was struggling against Intel, Bulldozer had failed to deliver the competitive performance the company needed, debt was a serious problem, and Zen was still an unproven architecture under development. Then came an opportunity from China.
In 2016, AMD entered a joint-venture arrangement with Tianjin Haiguang Advanced Technology Investment Co., commonly known as THATIC. According to AMD’s financial filings, the agreement involved licensing certain AMD processor technology for $293 million, plus royalties from future products.
For AMD at the time, hundreds of millions of dollars were rather more attractive than they would be to the considerably wealthier AMD of today. China, meanwhile, got access to something extremely difficult to develop independently: modern high-performance x86 CPU technology. However, one common description of the deal is misleading. AMD did not simply sell China an x86 license.
AMD Didn’t Sell China x86
The x86 ecosystem is wrapped in decades of patents, licensing agreements, extensions and cross-licensing arrangements, particularly between Intel and AMD. AMD could not simply photocopy its legal rights to x86 and hand the copy to another company.
Instead, the THATIC arrangement used a complicated structure involving two joint ventures with different ownership arrangements. AMD licensed specific processor intellectual property to those ventures while retaining different levels of ownership and control. The important distinction is that the Chinese side received access to specific AMD technology rather than an unrestricted license to future x86 processors.
And the technology at the heart of that agreement was particularly valuable. It was Zen. Not Zen 5. Not Zen 4. Not even Zen 2. The original Zen. But in 2016, that was exactly the technology AMD was betting its future on.
Meet Dhyana, Zen’s Chinese Relative
The first major product to emerge from the partnership was Hygon Dhyana. Calling Dhyana merely “inspired by Zen” would undersell the relationship considerably. Linux kernel patches submitted for Hygon processors explicitly described the first-generation Dhyana architecture as originating from AMD technology and sharing most of its architecture with AMD Family 17h. Family 17h is Zen.
The relationship was close enough that Linux could reuse significant portions of its existing AMD Zen support for Hygon processors. There were changes, of course. Hygon processors identified themselves differently. AMD CPUs traditionally report the vendor string AuthenticAMD, while Hygon adopted the wonderfully confident HygonGenuine. Dhyana also received its own Family 18h identifier.
Underneath those identifiers, however, the ancestry wasn’t particularly mysterious. When AnandTech and Level1Techs eventually tested real Dhyana hardware in 2020, the results largely confirmed what the software evidence had suggested. Dhyana behaved much like an early Zen processor, inheriting many of its characteristics.
This wasn’t a case of a Chinese company secretly reverse-engineering Ryzen. AMD had legally provided the technological foundation. For a brief period, Hygon essentially had a shortcut into the modern x86 CPU business. Then the shortcut disappeared.
The Door Closed in 2019
In June 2019, the U.S. Department of Commerce added THATIC and several related entities to the Entity List. The restrictions effectively prevented AMD from continuing to provide new processor technology to the partnership without government authorization. That meant Hygon wasn’t going to receive Zen 2, 3, or 4.
The technology it had already received, however, didn’t magically vanish from its servers because Washington changed the rules. Hygon still had its existing technological foundation. This created a much more interesting engineering problem. AMD had effectively given Hygon the first chapter of the Zen story, but Hygon would have to write the sequels itself.
That is where the modern C86 family begins to matter.
What Happens After Zen 1?
Hygon continued developing its x86 processors through successive generations known broadly as the C86 family.
The early generations retained an obvious relationship with AMD’s original architecture, but Hygon gradually modified more of the design. The third generation, for example, improved areas such as instruction fetching, I/O bandwidth, functional units and security while moving to features including PCIe 4.0 and DDR4-3200.
Higher-end Hygon 3 processors reached 32 cores and 64 threads, making it clear where the company’s priorities were heading. Hygon isn’t primarily trying to build China’s answer to the Ryzen 7 9800X3D.
Its important markets are servers, cloud computing, databases, virtualization and enterprise infrastructure. Those workloads reward high core counts, memory bandwidth, I/O connectivity and throughput more than squeezing another 17 FPS out of Cyberpunk 2077. This also explains why comparing Hygon purely against consumer Intel and AMD processors can become misleading. Its fourth generation makes that even clearer.
C86-4G Is Where Things Get Complicated
By the C86-4G generation, Hygon processors had acquired considerably more modern platform capabilities. Known configurations include processors with 16 cores and 32 threads, alongside DDR5 and PCIe 5.0 support. Hygon describes the generation as using a self-developed microarchitecture.
That sounds like a clean break from AMD. The reality appears more nuanced.
In 2026, GCC added dedicated optimization targets for several C86-4G variants, including the M4, M6 and M7, followed by another model known as M8 or Suzhou. LLVM and Clang have also gained Hygon-specific targeting. This is significant because compilers care about what processors actually do internally. Instruction latencies, execution resources and scheduling behavior affect how a compiler should arrange code. Hygon is therefore becoming distinct enough to deserve its own compiler optimization rather than simply being treated as generic Zen.
At the same time, the compiler work still reveals plenty of architectural similarities to AMD Zen. The best description is consequently neither “Chinese Ryzen” nor “completely original Chinese x86 architecture.” C86-4G appears to be a heavily evolved descendant of the original AMD technology, containing increasingly significant Hygon-specific modifications while retaining recognizable Zen ancestry.
CPU family trees, it turns out, can become as awkward as human ones.
How Much AMD Is Still Inside?
This is probably the most interesting question surrounding Hygon today, and unfortunately there isn’t a percentage meter we can check.
A processor microarchitecture consists of numerous interacting components. There is the branch predictor, instruction decoder, scheduler, execution units, caches, memory subsystem, interconnects, power management and countless smaller structures. Changing one component doesn’t suddenly make the entire processor new. But neither does retaining some design concepts mean the processor remains identical to its ancestor.
Imagine AMD gave Hygon the blueprint for a house. Hygon then spent years replacing the plumbing, moving walls, redesigning rooms, adding another floor and rebuilding the electrical system. At some point it becomes reasonable to call the resulting house Hygon’s design, even though anyone examining the foundation can still tell where it came from.
C86-4G appears to occupy somewhere along that transition. The upcoming fifth generation could move considerably further.
C86-5G Could Be Hygon’s Biggest Departure Yet
Details about C86-5G remain much less independently verified, so specifications should be treated cautiously for now. Current reports describe configurations reaching as high as 128 cores, along with DDR5, PCIe 5.0 and AVX-512 support. But the strangest feature is SMT4.
Modern AMD and Intel processors generally use two-way simultaneous multithreading. One physical core can therefore process two hardware threads. Hygon reportedly wants four. A 128-core C86-5G configuration could consequently expose as many as 512 hardware threads.
That does not make one processor equivalent to 512 conventional CPU cores. SMT allows multiple threads to share a core’s execution resources, improving utilization when one thread is stalled or unable to use everything available. Server workloads can benefit from that kind of design because they frequently contain enormous numbers of independent tasks.
Games generally don’t. If you buy a 512-thread server CPU exclusively for Counter-Strike 2, the processor is not the questionable part of that purchasing decision.
More importantly, SMT4 would represent a substantial architectural change from AMD’s conventional Zen approach. If Hygon successfully implements it, C86-5G could provide some of the strongest evidence yet that the architecture is moving beyond straightforward evolution of its AMD starting point.
We still need independent analysis before declaring that transformation complete.
Hygon Has One Enormous Advantage Over Loongson
Hygon becomes especially interesting when compared with Loongson. Loongson pursued architectural independence by developing LoongArch. That gives it control over its instruction set, but creates another enormous problem: software. Applications compiled for x86 do not magically run on LoongArch. They must be ported, recompiled or translated.
Hygon largely avoids that problem because its processors remain x86-64 compatible. The distinction illustrates two radically different approaches to technological independence. Loongson controls the language but has to convince software to speak it.
Hygon inherited the world’s most widely supported desktop and server CPU language, but also inherited a complicated technological relationship with AMD. For enterprise customers, Hygon’s approach has an obvious attraction.
Existing x86 applications can often move onto Hygon systems without being rewritten for an entirely different instruction set. Linux already understands the architecture, mainstream compilers support it, and decades of x86 software optimization remain relevant.
Hygon Isn’t Only Making CPUs
There’s another reason Hygon has become increasingly important. The company also develops DCUs, or Deep Computing Units, designed for AI and high-performance computing. The easiest comparison would be AMD Instinct or Nvidia’s data-center GPUs, although that describes their role rather than suggesting equivalent performance.
These accelerators target AI training, inference and scientific computing, and Hygon has again leaned heavily toward software compatibility. Its accelerator ecosystem has historical similarities to AMD’s GPU computing stack and supports familiar programming approaches intended to reduce migration work.
The pattern is becoming obvious. Hygon likes compatibility. Its CPUs retain x86 compatibility. Its accelerators try to minimize software migration. Rather than building an entirely isolated computing ecosystem and asking developers to please come over, Hygon is trying to make the transition as boring as possible.
In enterprise computing, boring can be an excellent feature.
Hygon Has Become a Serious Business
Hygon is also considerably larger than its relative obscurity outside China might suggest. The company reported roughly 14.4 billion yuan in revenue for 2025, an increase of nearly 57% year over year. Growth continued into 2026 as demand for domestic computing infrastructure and AI hardware increased.
Its close relationship with Chinese server manufacturer Sugon gives it another advantage. Sugon can integrate Hygon processors into complete servers, workstations and computing systems rather than leaving Hygon to sell CPUs in isolation.
The two companies even proposed a massive merger in 2025 before cancelling the plan later that year. They remain separate businesses, but their relationship illustrates something we have encountered repeatedly throughout this semiconductor series. A good chip isn’t enough.
You need systems, software, customers, manufacturing partners and an ecosystem capable of turning silicon into something useful. Lisuan has to build that ecosystem around a GPU. Loongson has to build it around an entirely different ISA. Hygon started with much more of the puzzle already assembled.
So Does China Really Have Its Own x86 CPU?
Yes, although the history behind it makes the answer much more complicated than the phrase suggests. Hygon did not independently create its first x86 processor from scratch. Its CPU business began with legally licensed AMD technology, and the original Dhyana processors were unmistakably relatives of first-generation Zen.
But that happened nearly a decade ago.
AMD stopped providing new CPU technology after the 2019 restrictions, while Hygon continued developing successive generations. C86-4G now receives dedicated compiler optimizations, and C86-5G reportedly introduces architectural features that don’t exist in conventional AMD Zen processors.
The interesting question is therefore no longer whether Hygon started with AMD technology. It unquestionably did. The question is when an inherited architecture becomes your own.
If someone gives you a processor design and you manufacture it unchanged, the answer is obvious. If you spend seven years changing its internals, building new platform capabilities, creating your own compiler targets and eventually introducing major architectural changes, the boundary becomes considerably fuzzier.
Hygon began its journey with an unusual shortcut into one of computing’s most protected ecosystems. AMD gave it Zen 1. Then the door closed. Everything Hygon has built since has been an attempt to answer a deceptively difficult question:
What comes after Zen when AMD is no longer there to tell you?
