Last week, we reported on Intel’s manufacturing constraints and the difficulties the company faces in ramping up additional 14nm production. Today, a new report claims that the CPU giant will outsource some of its chipset production to TSMC to ease its own capacity woes on 14nm. That’s the word from DigiTimes, which quotes the usual “industry sources” to say that Intel’s overall 14nm chip supply has fallen short by as much as 50 percent — but simultaneously claims that Intel only intends to outsource the production of its H310 and other entry-level chipsets to TSMC. Sources also claim that Intel’s 14nm process could fall short of demand by as much as 50 percent.
This doesn’t make a lot of sense. It’s one thing to say that Intel’s 10nm delay could be causing it some headaches and expansion problems — that’s probably true — but if demand for the company’s parts was running 2x higher than supply, we’d expect to see much higher prices than the relatively modest increases that have hit the market to date. It would also most likely require Intel to shift far more manufacturing capability to TSMC than merely offloading some chipset work.
Intel’s chipsets, while critical to the operation of its various platforms, are also much smaller than its CPUs, particularly its larger server cores. The fact that more capabilities are now integrated on-die at the CPU level than was once true should also reduce the overall size of these chips, and therefore make it easier to produce the necessary amounts with a relatively low amount of foundry capacity relative to the lines required to support high-end Xeon processors. Our guess is that the 50 percent shortfall, even if accurate, refers to demand for a specific chipset or product rather than Intel’s overall 14nm capacity.
While unusual, this would not be the first time Intel has tapped TSMC for additional manufacturing capabilities. Intel’s 28nm modems are built by TSMC and that foundry handled some of the company’s SoFIA chips and manufacturing for the related Rockchip deal (some of the SoFIA tablet and smartphone processors never went into production, but the Rockchip processors supposedly did). Intel is facing a short-term capacity crunch because the delay to its 10nm rollout has left it with capacity dedicated to that node that isn’t yet online at a time when demand for x86 products in data centers is quite robust. Intel has previously said that it expects to recognize an additional $5B in revenue in 2018 that it didn’t predict back in January due to this increased demand. It’s entirely possible, therefore, that this deal represents a profit maximization strategy for the company. It may be cheaper to push some chipset manufacturing off to a third party and keep more of its own foundry space crunching on high-margin Xeon parts than to shift existing 14nm over to building chipsets while simultaneously trying to bring 10nm online.
The overall impact to Intel’s business from this shift should be minor, particularly if TSMC has 14nm capacity it’s looking to sell. Higher prices on Intel chipsets and overall higher prices from increased demand could represent a pickup opportunity for AMD, but Intel will presumably be looking to manage its own inventory levels to prioritize parts for its most lucrative markets. If that rumor about a 50 percent demand shortfall actually does apply to Intel’s entire product line we’d expect price increases to continue and hardware to eventually become difficult to find, which would open up a commensurately greater opportunity for the company’s rivals, potentially including Qualcomm. But so far, the only products being moved are chipsets, which cuts against this being as big a problem as that blurb makes it sound.