- The acquisition of ARM by Softbank means that it gets access to much-needed resources to ramp investments and compete fiercely with Intel.
- Intel's decision to ditch the tick-tock cycle makes the rise of ARM-based server chips more probable.
- Customization services from AMD might as well contribute to the rise of ARM in the server space.
Intel (NSDQ:INTC) has been the unrivalled leader in the x86 server chip segment for almost a decade now. It’s repeated use of cutting edge manufacturing technologies, fabrication processes, node shrinks on an aggressive timeline and robust improvements in chip architectures with each tock cycle have allowed the chipzilla to command a 99% market share in the segment. But things may not necessarily always stay the same forever. There is a reason to believe that Intel’s hegemony in the server segment will be challenged vehemently by ARM (NSDQ:ARMH) and that the latter might well succeed in nabbing a piece of the server market away from the chipzilla.
ARM’s new big parent
Intel is currently at the helm, not by chance, but by making strategic investments. The chipmaker spent several billions each year in capex to make sure that its fabrication process, chip architectures and nodes were more advanced than any of its competitors. Its smaller rivals couldn’t match those investments due to their financial constraints and smaller scale. This served as a virtuous cycle for Intel; it brought more chip business for the chipzilla in return and kept competitors at bay in both desktop and server segments.
But any kind of lead, in any business, lasts only if the industry leader continues to make efforts to maintain its edge over the competition. In the semiconductor industry, think tank quality, R&D and capex spending are some of the very critical determinants of success. We’ve seen in the past that if any company cuts corners in any of the areas, competition tends to catch up with the dominant player sooner or later.
This appears to be the case with Intel. The chipzilla laid off 11% of its total workforce back in April. It also retired its tick-tock strategy earlier this year in a bid to lower its fab-related capex so that it could focus on acquisitions. While all this makes a strong short-term financial case that would boost Intel’s profitability, it does raise a few question marks regarding the sustainability of its manufacturing lead in the chip segment.
The first threat comes from ARM itself. The company was acquired by deep-pocketed Softbank recently. The CEO of the megabank sounded quite bullish about the server chip segment and the prospects of ARM in his recent interview. ARM will now get access to a bigger pool of resources from Softbank to ramp up investment in capital intensive technologies and its progress in the sector might well give Intel a run for its money.
Of course a lot will depend on ARM’s execution and the level of risk that its partners such as Qualcomm, ARM and Cavium would be willing to take. But at the end of the day, if Intel chops of its fab-related capex spending, while its competitors ramp up investments in that area and come up with aggressive designs, Intel’s dominance and pricing power in the segment will slowly start to erode away.
Slowing down on node improvements
I’ve mentioned this point earlier in the article but it’s actually another big threat that needs to be addressed. Intel no longer follows the tick tock cycle and has rather adopted a model that I like to call the tick-tack-toe approach. Up until 4-5 years ago, Intel used to introduce node shrinks every 2 years. It’s aggressive timeline of smaller node releases pretty much ensured that its chips were faster than the competition. This, in turn, brought desktop and server business to Intel and kept competition at bay.
But Intel decided earlier this year to ditch the tick-tock model and proceed with a 3 generation refresh cycle. This essentially means that a node shrink would now take anywhere around 3 years, up from the previous standard of 2 years. Granted that it saves costs in the short term but it also puts Intel’s manufacturing lead in jeopardy. The chipzilla was ahead of its competitors by 18-20 months in terms of fabrication process up until 2 years ago, but with a 3-year cadence now in place, there is a reason to believe that its lead will be completely exhausted by 2018.
Intel is expected to commence full-fledged mass-production of 10nm chips next year which means that it would stick to the node at least until 2020. On the other hand, both GlobalFoundries and Taiwan Semiconductor are expected to commence the volume production of 7nm chips by 2018. This would pretty much level the playing field for all the three fabrication companies, which means that Intel is going to have a tough time defending its pricing power against ARM-based server offerings going forward.
Customization must for the win
Next threat in line is that Intel offers minimal customization for its server chips. It mostly offers SKUs for retailing for small enterprise clients and that’s it for them. If they have special needs, they currently have no other option but to just go with expensive Intel SKUs that have unneeded functions.
On the other hand, AMD (NSDQ:AMD), which also happens to be a partner of ARM, offers a vast customization service to its enterprise clients. If a client wants a particular function embedded in its chip, AMD would do it. This lack of flexibility would drive business away from Intel, over to ARM clients, in my opinion. Not all large-scale enterprises and hyperscale networks have the same needs after all.
Intel has enjoyed its unrivalled hegemony for almost a decade now, but the competitive scenario will change sooner or later. I’m of the opinion that due to the aforementioned reasons, Intel’s pricing power and gigantic 99% market share in the server segment will steadily diminish over the next 5 years.