Keeping fab and design under one roof does not require that the fab and the design sides of the business exclusively with with each other.
(I think Intel should keep Altera, too. A paired CPU and FPGA solution has genuine value. One thing that Intel appears to still be good at is software tooling. If Intel were to provide tooling for their FPGAs that was anywhere near as good as their x86 tooling, then Altera could take off as everyone’s favorite platform. Right now they’re at substantial risk of losing new developer mindshare to Lattice because of the IceStorm project. I have trouble imagining that more than a few $M of total resources have been spent there.
As it stands, nVidia is a major player in big chips that do serious computation. Whatever one may think about CUDA, it’s in a whole different league from the garbage currently available from Xilinx and Altera.)
1. Mobile (I include cars, IoT etc)
2. Server/cloud (most new servers will be deployed to AWS, MS, Google, and maybe Alibaba)
3. Graphics / ML (you mentioned Nvidia and CUDA)
How would your suggested option 4 help Intel gain a significant foothold in any of those markets? (or stop the Arm transition in the case of #2)
Value proposition: Intel needs to scale manufacturing and R&D, in a way that doesn't leave them in a performance bind (ie without performance, they can't sell at scale, and thus can't afford to buy performance). Abstracting compute behind an IaaS offering would let them do exactly that while providing an alternate sink for binned chips.
Furthermore, they've got decent HW/SW integration talent in-house, and they could offer a key feature other clouds couldn't: vendor agnosticism. Building {PaaS product} on an Intel cloud seems strategically smarter than building on one of the continually-vertically-integrating clouds.
My bear case for Intel Web Services would be:
1. As IBM, HP and others discovered, the expertise required to run cloud services is very different from what's required to sell servers (or CPUs). 2. To compete with Amazon and MS, you would need to spend several tens of billions of dollars just to get a similar infrastructure footprint.
A variation on your proposal would be for Intel to buy a bunch of smaller cloud players, like DigitalOcean, and CloudFlare. Intel has the huge amounts of cash that these companies need to compete with AWS and Azure, Intel would then be vertically integrated, just like AWS is, and they would get the customer base and expertise from those acquisitions.
DigitalOcean, CloudFlare, etc would definitely have to be involved... but it's a tricky balancing act to avoid their becoming dominant solutions, at which point you would have AWS, except as a combination of companies.
Not sure how to square that circle. Maybe an extension of the Open Compute group + FRAND licensing of Apache-style preferred value add solutions on top, with Intel operating the hardware as a certified implementation? Essentially enabling MVNOs for cloud.
Intel's key distinction from foundry-less competitors, is that they can scale capacity to whatever they want, at least on older nodes. And the more volume they scale to, the more they ammortize their own costs (ie fab version of Amazon's AWS rationale).
The key would be establishing an ecosystem where innovation at the higher levels, and FRAND licensing of same, financially rewarded and incentivized players to invest in the platform.
Also, Intel-only would be limited compared to other clouds offering Intel/AMD/ARM.
It's more likely that Intel will increase market development funding to the existing cloud providers as their market share declines.
It seems like Intel could create a pretty compelling offering. Especially since AWS, Azure, and GCP all seem to be giving them the strategic middle finger.
Their fabs need more customers to justify the astronomical investment. They must expand their foundry business.
I expect the federal government to take action here, probably by giving huge monetary incentives to Intel/AMD for creating a state-of-the-art fab in the US.
TMSC is planning to build a fab factory in the US. https://www.forbes.com/sites/willyshih/2020/05/15/tsmcs-anno...
I think this is partly because they can easily get a license to use ARM or RISC-V in custom chips, but doing that with x86 is pretty much impossible.
Source? I don't think they care, I think they just don't want to be beholden to a single supplier, like they were to Intel for a while.
"AWS Graviton processors are custom built by Amazon Web Services"
and
"Apple Silicon"
It's not just about x86 -> Arm, it's about bringing more control of the design in house. This only gets more important as we move towards more SoC type designs. Imagine, for instance, AWS doing an entire SoC with Nitro, disc controllers, memory etc on one chip. I'm not sure that they could, or would want to do such a thing with Intel and x86.
Edit: Adding my sources ;)
Graviton quote: https://aws.amazon.com/ec2/graviton/
Apple silicon quote: https://www.apple.com/mac/m1/ (for fun, try to CTRL+F this page for "arm")
Option 1: Strip mine the company. This is more of the traditional PE/activist play. Once you stop spending all those R&D dollars, then Intel could extract a lot of profits from the x86 over the next decade or so. But this is only a short-medium term play. All mines run out of gold eventually.
Option 2: Split the Fab and design (x86) business. I'm not sure this helps that much, I think x86 is F'ed even if it could use the TSMC fabs. This is what AMD is already doing. The major important players (Apple, Amazon, MS, Goog) want to go ARM and then maybe RISC-V for their own strategic reasons anyway.
Option 3: Spend unheard of amounts of money to try and win back a process advantage over TSMC. This would entail spending tens of billions of dollars to basically make a bet that might pay off. It would be fun to watch, but success is not guaranteed, or even likely.
Is there an option 4?