ARM isn't comparable to Intel, they sell their designs to any of their licensees, including Intel. Intel in comparison, is a virtual monopoly in desktops , servers, HPC et al. Where do you think China can start - MIPS, Open Power or Zhaoxin? None of them are comparable to highly complex designs like Skylake, Ryzen or even Apple's Ax based on ARMv8. Not to mention - the patents.
I was talking about dominant ISAs. Intel and ARM.
Curiously, China can start by introducing ARM-derivative in server space! Lot of companies have tried that at one point or another and failed. China can do that because it has money to throw at this problem. China can devise a processor based on a derivative of ARM ISA implemented on many many cores, like 1000s and build a tool chain around say LLVM. Then sell it cheap to their own companies like TenCent or Alibaba. They can build cloud offering around it, complete with all JVM/XVM ported and an ecosystem around it. It will take time and a lot of money but it is doable.
This is how multi-trillion dollar industries are built. By reinventing an industry. Microsoft in its time did that. First with a desktop OS, then a graphical desktop OS, then with browsers and hence internet. They were setting standards at one time. They failed in mobile space though. They tried that however. China has deep pockets to keep on discounting hardware for long enough.
You must be know already that x86 or CISCy architecture is a poor choice for present day compilers and software stacks. Simple ISA and massively multi-core is the way to go with the way current software is written. No one is committing to it because it is expensive and long drawn battle.
That's the mobile space where China will continue to compete thanks to their economies of scale, 5G is interesting but we're still at least half a decade away from that in major parts of the world. In the meantime, Apple & Qualcomm still lead the mobile world, wrt performance, with Samsung following closely behind.
Chinese play long games. Long drawn out games.
I don't believe we've ever gotten close to matching Haifa or their fabled Conroe team. Chipsets & FPGA are a lot easier than designing something like a modern CISC or even RISC chip. In that, I'd argue designing a modern processor is certainly a harder step than fabrication & everything else that follows. Chip Fabrication itself is highly capital intensive & there's only 4 major foundries around the globe that can make complex chips on 2x nm & below. Packaging doesn't add much value, tape out is handled usually by the designated foundry partners, or in house in case of Intel & Samsung.
Modern processors are not monolithic designs. They are built block by block. I know teams within Intel India which are designing interfaces for front side bus on processor side.
Also quite sometime back, there was a historical project in Intel India called Project White-field (after an area in B'lore)(
https://www.theregister.co.uk/2005/10/28/intel_whitefield_india/). This was an effort to design a server processor, multicore back in 2005. It failed because of Intel's bone headed policies regarding taxation. My cousin was part of the team and after its head left the entire project fell apart. That said, they had completed the design and validated it on FPGAs in pre-silicon validation.
Again covered by massive patents, the only comparable industry titans they have are in Huawei & ZTE but limited to telecom.
Sorry I thought Tainhe2 was powered by Intel Xeon, did that change anytime recently? Talking about Sunway TaihuLight I see, not much info about their ShenWei processor so I can't comment on that, nonetheless it's impressive given what it's achieved.
Sunway or SW2600 takes its 'inspiration' in ISA from DEC-Alpha AFAIK.
https://www.nextplatform.com/2016/06/20/look-inside-chinas-chart-topping-new-supercomputer/