Most of what we talk about here is the PC business, or what Intel calls "client". Their other highly profitable segment is the datacenter, where AMD has been gaining share rapidly. There's limits to how fast AMD can ramp up though, since they have to compete for capacity on leading edge nodes with the extremely lucrative mobile SoC market. On the other hand, once performance and efficiency are competitive there's no obvious ceiling to the share they can grab. My contention is that likely even before they can grab a majority of share from Intel, they will divert enough of the growth from the sector to make Intel's financials look bad, and that this will hurt what has been a consistent growth segment for Intel all through the years of PC stagnation.
There's other even larger threats though, from ARM and possibly RISC-V. ARM designs are recently drawing close to Intel on a per-thread basis and typically offer significantly better efficiency.
RISC-V is a long way behind but making surprising gains. I didn't understand what the appeal was until I saw a talk on YouTube, where a major chip vendor remarked that they have various IP cores in a product representing 15 different ISA's in one of their products, all with their own toolchains and licensing/royalties to worry about. Consolidating that down with a common architecture makes sense, potentially even if your main business is selling a CPU with a proprietary ISA like x86 or ARM. So far RISC-V implementations have been limited to the low end but I think there's the potential for growth if big companies that add value at other parts of the stack decide to invest, such as cloud providers, especially those with a desire to limit their exposure to foreign dependencies such as Tencent and Alibaba.
Too, cloud providers are putting more and more value in the "serverless" part of their products, where there's no customer controlled instance to worry about hence no compatibility concerns.
All of this to me suggests a long term decline in the reliance on Intel as the monolithic performance engine behind the datacenter. And worse, I think this would have been true even if Intel hadn't suffered their process setbacks, that just accelerated the trend. Which means it's not necessarily a fix to move their chips to external fabs or catch up with their own process.
There's other even larger threats though, from ARM and possibly RISC-V. ARM designs are recently drawing close to Intel on a per-thread basis and typically offer significantly better efficiency.
RISC-V is a long way behind but making surprising gains. I didn't understand what the appeal was until I saw a talk on YouTube, where a major chip vendor remarked that they have various IP cores in a product representing 15 different ISA's in one of their products, all with their own toolchains and licensing/royalties to worry about. Consolidating that down with a common architecture makes sense, potentially even if your main business is selling a CPU with a proprietary ISA like x86 or ARM. So far RISC-V implementations have been limited to the low end but I think there's the potential for growth if big companies that add value at other parts of the stack decide to invest, such as cloud providers, especially those with a desire to limit their exposure to foreign dependencies such as Tencent and Alibaba.
Too, cloud providers are putting more and more value in the "serverless" part of their products, where there's no customer controlled instance to worry about hence no compatibility concerns.
All of this to me suggests a long term decline in the reliance on Intel as the monolithic performance engine behind the datacenter. And worse, I think this would have been true even if Intel hadn't suffered their process setbacks, that just accelerated the trend. Which means it's not necessarily a fix to move their chips to external fabs or catch up with their own process.
