When Intel announced that it was delaying its 10nm ramp earlier this year, it was a major miss for the company. Intel has led the industry on process node transitions for decades. Even today, Intel’s 14nm node is believed to be most similar to the 10nm nodes in production at Samsung and TSMC. Intel’s 10nm was expected to be as good or better than the 7nm nodes currently ramping up at other pure-play foundries. Now we know that the delay is a worst-case scenario for Santa Clara, with the ship date set for holidays, 2019. Data center products will follow thereafter.
The 10nm delay was the overwhelming focus of the Q&A portion of Intel’s Q2 2018 conference call. In fact, the company’s overall financial figures were excellent. Compared with the same period last year, operating income rose 37 percent, to $5.3B. Every Intel business segment reported growth and the company’s financials are quite robust.
Note the company’s data center operating income. Data center net revenue is only ~64 percent as high as client computing. Data center operating income (squared in green) is ~85 percent as high as CCG, which reflects the substantially higher profits one can earn in the data center business as a whole. Client computing’s share of Intel’s overall revenue continues to shrink — last year, for the six months ended in July 2017, CCG represented 81 percent of Intel’s total operating income. Today, CCG represents 62 percent of Intel’s operating income. The acceleration away from client computing that began under Brian Krzanich is accelerating.
There are two ways to read these events, and I think there’s fairness to both of them. On the one hand, Intel’s failure to deliver 10nm on time after 14nm was itself delayed is a significant embarrassment for Intel. It’s the first time in decades that the company hasn’t led the industry in deploying a leading-edge node. And Intel is well aware of it — if our own theory is accurate, Chipzilla fired Krzanich for missing the 10nm node and the consensual internal affair was nothing but a smokescreen.
On the other hand, what short-term penalty or risk does Intel face from its failure to ship 10nm silicon in 2018? Nobody expects the new node to deliver an overwhelming performance improvement or power consumption advantage. This delay may open a slightly wider window for ARM to enter the laptop and low-end markets, but ARM is attacking those spaces from the bottom up with low-cost SKUs, not hitting the high-end where Intel is most likely to concentrate with its eventual 10nm launch. Intel’s 10nm delay may be good news for AMD and give the company some additional breathing room with its own 7nm silicon, but we won’t know if that’s true until those chips hit the market. We also won’t know how much of a temporal advantage AMD can gain until we know when it’s going to launch Ryzen 2 — and right now, we don’t know that.
AMD has continued to reiterate that it’s targeting mid-single-digit market share in servers this year, so I don’t expect to see Intel hemorrhaging market share at any point this year. The bigger risk for Santa Clara would be an early 2019 Epyc and Ryzen 2 launch with real gains on 7nm while Intel remains stuck in the doldrums on 14nm. During its conference call, Intel reiterated that it expects a much smoother transition from 10nm to 7nm, noting: “We’ve also made some fairly judicious choices in defining 7-nanometer, learning from our 10-nanometer experiences. And we’re focusing on an optimum balance point between density, power and performance, and schedule predictability. So I think what you’ll see is a more balanced approach across those three vectors.”
We shall see. For now, we’ve got at least one more generation on 14nm to work through.