Intel May Be Finished With Contract Manufacturing
DigiTimes, granted, isn't known for being a solid single source in every instance, but the points made in the article not only echo what's been obvious in the larger semiconductor industry, but they also reflect the general state of Intel's roadmap as well. LG was originally announced as a major Intel 10nm customer, but obviously, the company wasn't going to ship customer hardware before it had the node online for its own use.
DigiTimes sources claim that Intel's manufacturing prices are higher than those offered by TSMC and Samsung, and that the ongoing 10nm delay has tightened supply of 14nm and 22nm products as well. To date, the only prominent win for Intel's custom manufacturing — Altera — became its subsidiary. That's not to say the acquisition was the wrong move for Intel, but buying your customers isn't exactly the classic foundry model.
In our IBM story, we noted that Samsung hasn't earned many foundry customers for its 7nm node, at least not major wins that it's publicly announced. DigiTimes remarks on this as well, stating:
For Samsung, the company has recently announced completing the development of 7nm LPP (low power plus) process incorporating EUV technology, which will soon enter commercial run. But industry sources said Samsung has yet to land orders from any major customer for chip fabrication using the latest process. In fact, Apple has placed all 2019 orders for its A13 mobile processors with TSMC.
Now, given that Samsung also functions as an IDM, it's not quite the same disaster as it would be for TSMC to lack leading-edge customers, and its agreement with IBM give it a prominent win — but given that TSMC's 7nm node is reportedly running below capacity in 1H 2019, it seems foundries may be having trouble filling their lines to begin with. With Intel beginning to roll out 10nm at the tail end of 2019, it's not clear the company has the fab space to spare for foundry customers. It's also not clear if it has potential customers in the first place.
We wouldn't expect much downside to Intel if the company does close down its client foundry, if only because it's far from clear the project ever brought in much business to start with. And keep this in mind as well: Given that Intel is struggling to deal with booming demand for its server products in the first place, there's precious little reason for the company to dedicate fab space to lower margin customers when it could make more money building chips for itself.