Implications of Intel losing its manufacturing lead: AMD, Xilinx, Nvidia and Qualcomm.

This is the first time *ever* that Intel has not had a process technology advantage. For its entire history, Intel was 6–15 months ahead of the rest of the industry in moving to a new manufacturing node (i.e. 14 nanometer). Because of Moore’s law and Dennard scaling, this meant that Intel consistently had a 20%ish price/performance/power advantage over its competitors.

Even though Moore’s law and Dennard scaling are slowing down, the fact that TSM is at least 9 months ahead of Intel to the 7 nanometer node means that Intel, for the first time ever, is at a 20% price/performance/power disadvantage based solely upon manufacturing. This is a seismic shift in the semiconductor landscape. I believe the importance of this development is highly underestimated; especially now that AMD is fully fabless. A short review of microprocessor history:

Intel focused much of its development effort on the 64 bit Itanium in the early 2000s. This was a mistake as software developers largely refused to rewrite their code to take advantage of Itanium. This let AMD take significant share with its 32 bit Opteron family; which peaked at roughly 25% share of server CPUs in 2006/2007.

AMD’s Opteron had an architecture advantage because Intel’s “A team” had focused on designing Itanium rather than the 32 bit Xeon. However, at the time AMD was at a disadvantage to Intel from a manufacturing process technology perspective which helped offset Xeon’s architectural shortcomings.

More importantly, AMD was not fabless at the time and could not ramp up capacity fast enough to meet demand. This allowed Intel to use volume purchasing agreements to disadvantage AMD in the following way:

“AMD simply does not have the capacity to meet all of your demand. If you switch a significant portion of your volume to them, we will raise prices on you now and never forget the betrayal vs. if you keep us at a majority of demand despite our disadvantages, we will cut prices.”

Think about the market power implicit in that stance. i.e. The ability to raise prices significantly on customers who switched demand to a competitor. The market power came from the fact that AMD fabs could not fulfill more than 30 to 35% of global CPU demand.

Bringing on a new fab took years and bringing on fab capacity sufficient to support 50%+ share would have been a bet the company risk for AMD.

Today, AMD’s EPYC server CPU may have a modest architectural advantage but it has a process technology advantage that compounds this. And given that AMD is now fabless and semi demand is slowing, AMD will have no problem meeting EPYC demand as there will be less “competition” for leading edge wafer supply.

As a result of all this, I am positive on AMD over the next several years. Near term may be choppy due to GPUs/Crypto hangover. CEO Lisa Su is excellent and the potential for Intel to recruit her away is the other primary short term risk.

This thread focused on AMD, but almost every company that competes with Intel is advantaged by the fact that TSM is now 9–18 months ahead of Intel on process technology. Every fabless company now has a 20%ish performance/power consumption advantage over Intel after being at 20%ish disadvantage for the last 20 plus years.

This is positive for Nvidia, Xilinx and Qualcomm in addition to AMD. Also helps all the cloud players building their own semiconductors, i.e. Amazon’s ARM based server processor will be more competitive than it otherwise would have been.

Intel losing its process technology advantage for the first time ever and Apple hiring John Giannandrea were two 2018 developments with profound long term implications for the technology industry.

• This was originally published as a thread on Twitter on 1/19/19.