Intel Increasing Use of Third-Party Fabs as 14nm Shortage Continues

(Image credit: Intel)

In a sign that Intel's pervasive shortage of 14nm production continues to be a nagging issue, Intel issued a press release today that states it is increasing its use of third-party foundries. The company also notes that the shortages persist even though it has increased production in the "double digits," with no specific end date given. 

Sources close to the matter tell Tom's Hardware that the increased production from third-party foundries consists of products other than processors, enabling Intel to focus more of its fabrication capacity on CPUs. 

This follows reports earlier this year that Intel planned to outsource production of its 14nm Rocket Lake CPUs to Samsung, which we also confirmed was not the case. Instead, these contracts deal with simpler low-margin chips that ship in high volumes, like chipsets, and other products already fabbed outside of Intel's own foundries.

We've already seen signs that Intel has outsourced some chipset production to Samsung, which the company has done in the past. Intel has already made other adjustments to free up 14nm capacity, like stepping back some of its own chipset production to the 22nm process.

Intel's disclosure aligns with several statements the company has already made, with a blog penned by Intel's Ann Keller in December 2018 pointing out the practice: 

"In addition to expanding Intel’s own manufacturing capability, we will continue our selective use of foundries for certain technologies where it makes sense for the business. The use of foundries has been an Intel practice for nearly two decades. As we invent more products for a broader set of customers, you can expect us to be strategic about the application of Intel’s differentiated manufacturing capability and the selective use of foundries."

Intel currently uses third-party fabs for several of its products, tapping TSMC's 16nm for Nervana NPP-T, and TSMC's 7nm for its Mobileye and Barefoot chips. 

The 14nm chipsets exacerbate the manufacturing challenges in Intel's foundries. In most cases, the company has to create one chipset for each processor, so easing that production load, not to mention the overhead in its packaging and test facilities, would free Intel up to produce more high-margin 14nm processors.

Intel's latest blog post doesn't specify when the shortage will end, but does note that the company is busy allocating more of its resources to boosting both 14nm and 10nm production. That isn't very comforting to OEMs and hyperscalers waiting for shipments of chips, but it also indicates that AMD's continued push into several market segments apparently isn't hurting Intel too badly in terms of volume: The company is immediately selling every chip it can punch out. Instead, the impact of competition from AMD is more pronounced on Intel's balance sheet

Intel's full statement follows: 

November 20, 2019 

To our customers and partners, I’d like to acknowledge and sincerely apologize for the impact recent PC CPU shipment delays are having on your business and to thank you for your continued partnership. I also want to update you on our actions and investments to improve supply-demand balance and support you with performance-leading Intel products. Despite our best efforts, we have not yet resolved this challenge. 

In response to continued strong demand, we have invested record levels of Capex increasing our 14nm wafer capacity this year while also ramping 10nm production. In addition to expanding Intel’s own manufacturing capability, we are increasing our use of foundries to enable Intel’s differentiated manufacturing to produce more Intel CPU products. 

The added capacity allowed us to increase our second-half PC CPU supply by double digits compared with the first half of this year. However, sustained market growth in 2019 has outpaced our efforts and exceeded third-party forecasts. Supply remains extremely tight in our PC business where we are operating with limited inventory buffers. This makes us less able to absorb the impact of any production variability, which we have experienced in the quarter. This has resulted in the shipment delays you are experiencing, which we appreciate is creating significant challenges for your business. Because the impact and revised shipment schedules vary, Intel representatives are reaching out with additional information and to answer your questions. 

We will continue working tirelessly to provide you with Intel products to support your innovation and growth. 

Sincerely, Michelle Johnston Holthaus 

Executive Vice President 

General Manager, Sales, Marketing and Communications Group

Paul Alcorn
Managing Editor: News and Emerging Tech

Paul Alcorn is the Managing Editor: News and Emerging Tech for Tom's Hardware US. He also writes news and reviews on CPUs, storage, and enterprise hardware.

  • hannibal
    So Intel is using Also TSMC... good plan, now amd have even less production capasity to use :) the cpu / gpu shortage of amd products is getting worse.
  • InvalidError
    hannibal said:
    So Intel is using Also TSMC... good plan, now amd have even less production capasity to use :) the cpu / gpu shortage of amd products is getting worse.
    AMD almost certainly has a multi-year wafer start agreement with TSMC that guarantees a minimum volume and likely some margin for extras. TSMC getting even more customers and an even longer backlog has no impact on that. As far as AMD is concerned, it will only delay supplemental orders beyond the agreement's maximum should AMD decide it needs to place any. If AMD foresees continued increasing demand, it could also seek an increase in its commitment's limits that would get phased in over the next six months of TSMC backlogs.

    It is mainly companies that want occasional batches instead of on-going production that get screwed over by long backlogs as more fab lines get dedicated full-time to large customers like AMD.
  • thGe17
    As a note: The NNP-T1000 is a bad example for outsourced production. When Intel aquired Nervana, the design was (almost?) finished using TSMCs 16 nm process (CLN16FF+), therefore it was obvious that Intel would continue this path. (Otherwise Intel would have had to redesign the whole chip with its own tooling and libs, wasting a lot of money. Not funny with 27 billion transistors in size.)
    On the other hand, the NNP-I1000 for inferencing workloads will be produced in Intels 10nm process.