news intel acknowledges pressure on its 14 nm production capacity

Intel acknowledges “pressure” on its 14-nm production capacity

Lately, rumors have been swirling that Intel is facing the potentially enviable prospect of more demand for its products than it can supply. This morning, interim CEO Bob Swan confirmed that thanks to strong growth in Intel's data-centric businesses and unexpected growth in the client PC market for the first time since 2011, the company is facing "pressure on [its] factory network." As a result, Swan says the company is prioritizing production of its Xeon and Core processors "to serve the high-performance segments of the market."

As a result of that strategy, Swan concedes that supply of Intel products is "tight," especially in its entry-level PC chips. To take the edge off this supply pressure, Swan says that Intel plans to increase its capital expenditures by $1 billion compared to its projections at the beginning of the year, to $15 billion, in order to boost production capacity at its 14-nm fabs in Oregon, Arizona, Ireland, and Israel.

Swan also notes that Intel continues to make progress on its 10-nm process. He claims that 10-nm yields are improving, and that the company still expects it can begin "volume production" on its troubled next-generation node in 2019. Increased 10-nm production would presumably lessen pressure on the company's 14-nm production capacity, as well.

In closing, Swan says the company will "stay close, listen, partner and keep [customers] informed" about its ability to meet demand for its products. He notes that of the companies he's addressing in his letter, "many of you have been longtime Intel customers and partners, and you have seen us at our best when we are solving problems." He says that "the actions we are taking have put us on a path of continuous improvement." We'll be watching to see how those efforts bear fruit in the months and years to come.

0 responses to “Intel acknowledges “pressure” on its 14-nm production capacity

  1. There’s been no ‘winding down’ occurring. Intel has one fab doing 10nm work side by side with 14nm (as they use the same equipment) and that has been the case since 14nm went into volume production. The problem is Intel have historically had a huge overcapacity (from spinning up for a mobile device market they were never able to make real inroads into) and are only now starting to actually hit capacity limits with the modem order with Apple. It’s not wafer throughput that’s the bottleneck here, but test & package. As test & package cost is close to the same regardless of die size, Intel would probably want to dump the modems and dedicate capacity to higher margin CPUs. They can’t do that, because even Intel cannot afford to burn fruitzilla, so the lowest margin of their own chips end up getting the short stick: consumer desktop chips and low-power laptop chips.

  2. He’s not wrong. A lot of mobile SoC designers backed off of 7nm for the cost, and AMD certainly couldn’t have funded it on its own. Apple being an early backer and adoptor is a big part of it being economically feasible at all.

  3. Dude, seriously. You have your preferences and that’s fine, but are you really going to tell people that there’s no sales pitch here? That’s the whole point of putting this press release out in the first place.

  4. No, no, no. It’s up to the SEC to determine. This is nothing to do with marketing. Lying is a Muskable or Holmesable offence. They’re dead serious with statements like:

    “In fact, our data-centric businesses grew 25 percent through June, and cloud revenue grew a whopping 43 percent in the first six months.”

  5. It’s literally the first paragraph of this article on this here august website… Not some sort of wild assertion on my part…


  6. It makes sense to me. A competitor that has 3% market share instead of 0.3% market share still isn’t much of a competitor, but I’m sure Intel has people doing whatever they can to stop exactly that sort of thing from happening here.

  7. I disagree with your assertion that Intel is increasing production in response to demand. Intel is maintaining high end shipments at the detriment to their desktop sector due to chipsets and other devices moving to 14 nm.

  8. I wait patiently for something more than a vapid comment from you. Now go back to bro-stalking SSK.

  9. What are you saying “no” to about? Do you disagree with the notion Intel is increasing production of Xeon and Core products to satisfy increased demand and in the process making tonnes of moolah?

    What do you think earns Intel more $$$ – a wafer of Pentium Golds or a wafer of Xeon Golds?

    Would they have behaved any different even without a competitive AMD? This is basic, rational business behaviour.

  10. As someone in the industry…no.

    Intel is seeing increased competition (up from basically zero) in the enterprise segment and they are deathly afraid of missing shipping dates and/or constraining supplies – either one would give AMD a big opportunity to step in.

  11. it has impacted sales: [url<][/url<] those are numbers of a German webshop, a favorite among German PC enthusiasts. The total impact is probably a lot smaller than these numbers would make you believe

  12. Care to explain why Intel’s marketing has been trying hard to downplay the Epyc platform since its introduction with literal FUD campaigns?

    Intel shareholders most certainly care about retaining the company’s massive marketshare in the lucrative SMB/Enterprise world. They would do anything in their power to retain it from any threat no matter how trivial it is. The desktop and budget markets are inconsequential by comparison.

  13. More smoke :


  14. I’m just reading the lines, not the whitespace between…

    The logic is incredibly simple. Intel are making as much as they can from their fabs BECAUSE demand for their higher margin Xeon and Core stuff is INCREASING. They’re selling more of the good stuff. Why should they be concerned about losing marketshare? They’re not.

    I mean, good on AMD for having good compelling products and finding their place in the sun. But the story for Intel is that – right here, right now – it’s raining cash.

    But I’ll get downvoted by logic-proof AMD fanbois…

  15. Here’s something to whet your appetites..


  16. You aren’t reading between the lines. Intel is gravely concerned about losing precious marketshare in the SMB/Enterprise markets. That’s why they are prioritizing their limited 14nm production towards those markets and ignoring the mainstream and budget stuff.

    They know that Skylake-X HCC and XCCs have an uphill battle against the current Epycs and upcoming refreshes in SMB/Enterprise markets. Cascade Lake refresh is going do little to stop this.

    The manufacturing side is what is really killing Intel. They simply cannot deliver the quantities need to satisfy the demand from their SMB/Enterprise customers. It is from the inherent problem of producing massive pieces of monolithic silicon. Intel was already shifting gears towards “chiplets” on their SMB/Enterprise SKUs once the manufacturing problems became apparent with Broadwell-E and Broadwell-EP (Skylake-X was already taped out by then).

    They weren’t expecting AMD and ARM to come out competitive solutions so soon and really hoped that 10nm process would have iron out without too much hassle. It is going to be a rocky road for Intel in the next few years within the SMB/Enterprise world.

  17. Did you even read what’s at the top of this web page? Written by this here august website?

    “(Interim CEO) Swan says the company is prioritizing production of its Xeon and Core processors “to serve the high-performance segments of the market.”

    “As a result of that strategy, Swan concedes that supply of Intel products is “tight,” especially in its entry-level PC chips.”

    In plain layman’s terms, Intel’s margins are going up. They harvest the gold. The quality fruit. They be making the big bucks. The payola. Pay dirt, baby.

    No doubt I’ll be heavily downvoted by people completely unable to handle the truth…

  18. [quote<]AMD will of course benefit from more demand in the market but they might find it's in the less profitable low-end space - cheap laptops with 1366x768 panels, budget gaming rigs, web surfing stuff for grandparents etc. [/quote<] You couldn't be any further from the mark. It is the SMB and enterprise segments is where AMD is going to be snagging the most market share from. The Epyc platform is simply too attractive for customers in this market to overlook. Intel's reputation and prestige in this segment was severely damaged from the recent discovery of hardware-level exploitations. Ironically, the ubiquity of IGPs on Intel's mainstream CPUs is what going allow them retain their stranglehold on the portable and mainstream desktop markets. It also where the same exploitations have little or no impact on.

  19. The article is wrong in its premise Intel is in trouble. Intel can’t make enough product to satisfy demand. That’s an enviable position to be in. They get to pick and choose which part of the market they want to satisfy which will obviously be higher end, higher profit parts. Server CPUs, processors for ultrabooks, workstations and gaming PCs.

    AMD will of course benefit from more demand in the market but they might find it’s in the less profitable low-end space – cheap laptops with 1366×768 panels, budget gaming rigs, web surfing stuff for grandparents etc.

  20. In connection to this, you guys might find this article interesting:


  21. quite a big change from their usual song [url<][/url<] i guess they aint spendin' cheese.

  22. x86 applications run like 2013 Atoms. They run noticeably slower than Goldmont. Yes, you have UWP apps, but that was true back with WinRT devices as well. Not to mention again, Surface Go uses a Kabylake Y chip, a sucky one, but a Kabylake nonetheless.

    Chrome OS isn’t taking over the PC. Android and iOS has their own market. And its *much* harder for a hardware manufacturer to take over because software and users take priority over specs.

    No need to state further. Time will prove itself. And the biggest barrier to ARM on the PC was always Intel covering their asses by making low power chips(like Goldmont Plus) better. It won’t prevent companies from keep trying though. Such pressure also keeps Intel from being lazy.

  23. If that’s such a great idea, then why is Intel using its own fab space?

    [url<],37819.html[/url<] (which I linked previously and got down-thumbed for)

  24. Again, Chrome OS and android both came on and are taking over with a smaller ecosystem. I don’t agree with you.

    [quote<] Laptop space - AMD since Windows 10 on ARM is going to be another DoA when you are getting compatibility issues and 2013 Atom performance. [/quote<] What did you mean by this?

  25. It’s not perception, its reality. You can’t go after an entrenched ecosystem with an inferior product. Even 1% of applications not working properly is reason enough not to go with it. The existing ecosystem by sheer momentum is “forced” to continue development and support because for them its bread and butter. The newcomer has to prove themselves regarding this, because until it becomes a significant business for them its merely a toe in the water type of thing. If you look at the big picture, the customer is actually not that stupid.

    “You stated 835 performs like a 2013 Atom, and it doesn’t, even when running win32 apps. It’s far quicker. ”

    By the way, I said nothing of the sort comparing it to the 2013 Atom. I said the Pentium Silver is comparable in Integer performance to the SD 845. 2013 Atom is actually not that bad, and I had a Bay Trail Tablet. But Pentium Silver is leaps and bounds better.

    Surface Go is using the Pentium 4415Y, which is an extremely underpowered Kabylake part. They should have used the Pentium Silver chip instead.

  26. Just to be clear, I also think WOA will bomb, but mostly because of the perception of issues being a problem rather than there actually being issues, similar to what occurred with Windows Phone. From your own links:

    [quote<] I'll say this: In general, I encountered few lags on the NovaGo. Sometimes it stuttered when loading a webpage that's animation-heavy, and it slows down when I have more than three dozen Chrome tabs open. But I've also been able to do some light editing on Photoshop CC while writing this review and having Audacity open in the background with no major delay [/quote<] it's not 2013 atom performance. I own a Goldmont PC, a Surface Go, and it's nothing like the 2013 atoms you reference [i<] which I also own. [/i<] You stated 835 performs like a 2013 Atom, and it doesn't, even when running win32 apps. It's far quicker. [quote<] the Windows on Snapdragon ecosystem is more capable than Chrome OS [/quote<] Chrome seems to be doing fine, so that in itself shouldn't kill the platform. 64 bit apps aren't available yet, and while I agree that's a pain in the butt, it's not really a major concern for me given what these devices are for. [quote<] I didn't run into major issues, knowing I might not be able to install something I need was like constantly looking over my shoulder for a stalker. [/quote<] So her issue wasn't that she actually had real problems, it was that she perceived she might one day. Basically, that's exactly what i said: perception of issues killing the platform, not real issues. They both tried to install VPN apps, which Microsoft has specifically stated won't work. You can configure VPN's, but because of hardware limitations can't install x86 vpn apps. They're also not great for gaming. Given that those are clearly not what these devices are for, they're Chromebook competitors, that's unsurprising. I agree they're too expensive. OEM's have claimed they need to make up the r&d from these as they're currently low volume, but i think it'll help kill the platform. The snapdragon 1000 looks interesting, and we'll see how it goes, but I honestly think that the criticisms of this platform have more to do with perception than reality.

  27. Good point. Packaging costs also tend to dominate as die sizes become too small because packaging costs are fixed, so I don’t think companies like Intel try too hard in going below a certain size, because of diminishing returns.

  28. You are welcome. Techinsights is a good source to find die sizes for such chips, but you may not get it for everything.

    Others require digging into datasheets. So it can take time to find, if you can find them at all.

  29. Compatiblity issues.

    [url<][/url<] [url<][/url<] Compatibility issues always exist because you simply can not account for every implementation detail. The issue existed with Alpha's FX!32 emulator to run WinNT back in 1995, and IA32 hardware and IA32EL to run 32-bit code on Itanium. Intel hasn't stood still since 2013. Their latest efficient core implementation is Goldmont Plus with the Pentium Silver naming that substantially outperforms the previous generation(itself which substantially outperforms the 2013 Atoms). Goldmont Plus in Geekbench is actually comparable to SD845 in native performance when looking at Integer performance. Of course a non-native SD835 is going to be much slower. [url<][/url<] All that is made worse because the device goes for $800. Why not just sell it at $400 to make it competitive? Because I bet part of the extra cost is spent on premium system components that use less power so it can have great battery life, which is the only advantage.

  30. The problem with the modem supply contract is not wafer throughout, but test & packaging. The dies are teeny weeny, but still need most of the testing time and floor space as (e.g.) a laptop CPU, and almost identical time and floor space for packaging (die onto substrate, not the cardboard box!). Intel have sent out comms that a new Vietnam test & package plant is coming online for the U series chips, but if they are limited in capacity for now their priorities are the moderns (you do NOT want to piss off Apple by reneging on a supply contract) and the high margin big dies (biggest bang for the packaging buck).

  31. Are you sure that GPUs and CPUs are low margin parts? I suspect that phone SOCs have razor-thin margins and I don’t see much else being made at leading-edge process tech. Of course, pro parts would be a priority (EPYC will be the first 7nm part for example), but $1000 nVidia GPUs are not cheap either.

  32. Market prices do not always align with MSRP. Local prices where I live are, roughly translated to $, 8700K at $490 and 2700X at $330. The 8700K is definitely the better processor in most cases, but the difference in price is 50%.

    Can’t say if it is because of shortages. The 8700K used to be $380 here, which was pretty reasonable in relation with the 2700X.

  33. Which compatibility issues?
    Performance running win32 apps is better than atom already. On uwp apps it’s i3 speeds.
    You’re also forgetting that android is the world’s number one OS, not Windows. As generations of kids learn on Chromebooks exclusively we will see further shifts away from X86. At least, that’s what I expect.

  34. Mark Bohr is one of those Intel guys that I have gotten quite fond of and respect, mostly because of his cool videos on YouTube. Sadly I think he’s in the middle of all this process tech turmoil at Intel.

  35. Yeah. Every time someone buys an Intel processor for desktops, laptops, workstations or servers, AMD doesn’t have an alternative product that people may look at. /s

  36. Brah, I accidently spent my pc money on buying an I5 7400 for 90 cad. My kids had a 6100 and we were starting to feel it. I was going to pick up the annual game pass, but now I gotta see how it goes. You know Kate, man.

  37. Keep in mind Intel’s 22nm process is only 30% denser than TSMC’s 28nm process. It was with 14nm that started their absurd density focus(to be more fair, it probably helped in some areas, but not on their bread and butter). The density focus likely caused the 6+ month delay on their 14nm and many years for their 10nm.

    The die sizes for their 22nm and 14nm chipsets are roughly equivalent. Going to 22nm more than doubles the size.

  38. [quote<]what did you end up building?[/quote<] nothing yet, I'm undecided. Maybe the launch of the coffee lake refresh CPUs will relieve the shortage for a bit in my region and make Intel CPUs attractive again, maybe not. [quote<](if the post is even true)[/quote<] I didn't join this community in 2004 just to just start telling lies on the internet in 2018. It's good to keep your wits about these days but not everyone on the internet is a shill

  39. It’s probably a fine choice, but it just doesn’t seem optimal to use TSMC when Intel has 22-nanometer fab capacity available. Which they apparently do, if Tom’s is to be believed.

  40. Intel’s average sold die size is ~100mm2. This includes server, desktop, laptop. They might be slightly greater, maybe 110mm2. Volume-wise vast majority is small and low cost parts. Then again, quad cores are only 126mm2 in size and hex cores at 150mm2.

    In comparison, the 32nm chipset generation is at 47mm2. The 22nm generation chipset is roughly at 58mm2 and the 14nm at roughly 56mm2.

    The XMM 7460 built on TSMC’s 28nm process had a die size of 70mm2. The previous generation had a die size of 66mm2.

    The difference is unlike for CPUs, they ship a single variant of the LTE modem, so whatever the die size for the XMM 7560 is becomes the average die size. The 7560 is probably quite a bit smaller, but at 120-140 million units shipped its a huge amount Intel needs to supply.

    (We should get analysis on the XMM7560 soon)

  41. Dear Intel,

    how about releasing a bios patch that unblocks 1xx / 2xx chipsets from working with your 8th and 9th gen CPUs?

    that way we don’t have to waste money on the new board with the same features, and you free up production capacity for your expensive cpus?

  42. In a sense this is right. Arm/Apple/Qualcomm (esp Apple) are the ones willing and able to pay for leading-edge process tech from TSMC.

    Take away those customers and TSMC would likely still be on 16nm.

    AMD gets to benefit from the investments TSMC has made in response to those bigger customers.

  43. the i7-8700K went from about €380 to €500 here in a month. I admit that “prohibitive” is a strong word but it definitely completely changes the value proposition of the CPU that I want

    here’s a graph of the price for where [b<]I[/b<] could buy the i7-8700k: [url<][/url<]

  44. I agree completely that AMD and Nvidia are at the mercy of TSMC and Samsung’s engineering prowess. If either fab is unable to get 10nm (or the touted 7nm) up to high volume in the time frames they’re currently publicizing, there’s going to be major repercussions. While Intel has effectively lowered expectations with continual delays, it really does seem like it’s full steam ahead for AMD and NV. If the production forecast gets pushed back I think there’s going to be a big reaction in the markets.

    With that said, I don’t necessarily agree with your second point. The margins the fab makes is entirely up to them – AMD and NVDA can’t exactly go shopping around for a better deal. There’s only three real players left, so if TSMC or Samsung are getting poor yields, they can pass those costs on to A/N. It’ll be up to A/N to then determine what kind of margin they make selling to the public and OEMs. Furthermore, since this is truly the bleeding edge and these chips MUST use the latest node size, I think these chips will get first priority for the fabs. Less important chips (in the grand scheme of things) can be farmed out to dozen or so fabs that are pumping out 28nm chips. Of course it’d be nice for everything to be on the latest nodes, but I think the chips in a D-Link router will get the short straw vs a complex bleeding edge CPU.

  45. Yeah I’m curious what Firestarter ended up doing (if the post is even true). Presumably Intel lost the sale, so you went with a Ryzen processor.

    Based on what I’ve read, the desktop APU’s are pretty weak and not competitive at all with Intel processors. The best Ryzen APU (Ryzen 5 2400G) is a 4 core/8thread with 3.6ghz base/3.9ghz boost. TDP is limited to 65 watts so I don’t see much OC’ing in its future.

    So the most logical decision to me is to skip the APU and buy a Ryzen CPU (good performance) plus a dGPU. Problem with this decision is that dGPU’s have had astronomical costs in the recent past. With a standard Ryzen there’s no way around the need for a dGPU…I suppose you could scrounge an old one from a long time ago to get by, or hunt down some used super low end card for $50. Outside of that scenario, I imagine you’re buying a dGPU that costs several hundred bucks…which means I find it hard to believe a $50 swing in Intel’s prices would really push you away.

    Intel processors simply haven’t had the price inflation that RAM or GPU’s have had. So….Firestarter, what did you end up building?

  46. On the plus side, yields should be excellent, they’ve had plenty practice at 14nm…..
    On the flash side-not so good, rumors QLC at under 50%……

    Recently read review on Crucial BX500, and all the way thru kept thinking why-why?
    Came to the conclusion/guestimation, product was probably meant to be QLC, but
    bad yields killed that idea…………………………………….

  47. My google-fu may be lacking but I couldn’t find it – does anyone know what the die size of this modem is (or similar competing cellular modems)? I imagine they’re tiny compared to an x86 die, but just how big is the size differential?

    On a similar note, when Intel switches most of their volume production CPU nodes to 10nm, will the chipsets follow suit again? Or will the chipsets stay at the mature 14nm size? And is it easy to swap a 14nm CPU production line to a 14nm chipset production line, or is it much more complicated than that? If they’re expanding the 14nm production now, I assume some of the new lines that just came on will stay at 14nm and produce chipsets rather than immediately upgrade to 10nm. And again, are chipset dies similar to CPU dies in size? Much larger? Much smaller?

    I’m curious if a given wafer produces say 100 8-core CPUs, how many z390 chipsets could that same wafer make? How many 5G cellular modems could that wafer make?

  48. Apple is competition for Intel in Apples own devices chips, but nothing else. That was about 5% of Intel revenue iirc, a sizable hit but one they have years to deal with, so not something that would be a strong competitive headwind.

    And if you think stronger chips would impact Intel by way of losing PC sales, just look at Android vs iOS. The vast market won’t shift over to Apple costs for better chips.

  49. Potential future competition, ARM cpu are still not revelant in Laptop, Desktop and servers, Intel closed it’s mobile division. I would say AMD is pretty much the only direct competition currently !

  50. Yes they are.

    Desktop space – AMD
    Laptop space – AMD since Windows 10 on ARM is going to be another DoA when you are getting compatibility issues and 2013 Atom performance. Ryzen mobile may be behind in battery life, but you are getting far better CPU and GPU performance for a few hundred dollars less.
    Server space – AMD because AMD runs on the same ecosystems Intel chips run on.

  51. None of them. A few of them went up a few bucks from the previous generation over the years, but nothing crazy.

    2600K – 317 2500K – 216
    3770K – 313 3570K – 212
    4770K – 339 4670K – 242
    6700K – 350 6600K – 243
    7700K – 340 7600K – 217
    8700K – 359 8600K – 257

    So, unless Firestarter is talking about the X/E series of CPUs, which have always been “prohibitively expensive” I’m just not sure what CPUs are being referenced.

  52. You’ll be one of the (relative) few. They don’t want to lose you, but they won’t wince at it either.

  53. they can acknowledge all they want, fact is that right after I decided I was definitely going to upgrade my PC and sold my old one to a family member, their CPUs became prohibitively expensive for me

    Intel is going to lose a lot of customers this way, including me

    edit: prices went way up in Europe: [url<][/url<]

  54. 14nm supply problems are twofold:
    -10nm chip delays meant chipsets are no longer a process gen behind. Previously, chipsets were a fab gen behind CPUs
    -The XMM 7560 LTE modem featured in the latest iPhone models use Intel 14nm. The predecessor 7460 modem used TSMC’s 28nm process. The amount of wafers that needs to be dedicated to satisfy Apple’s requirements are mind-boggling, considering its a single company.

    Each should increase Intel’s wafer requirements by roughly 30%. Add the two you are looking at Intel’s 14nm generation requiring 70-80% more volume than previous generations.

    Also interesting is that they wanted the next gen XMM 7660 on the 10nm process. If they were really planning on doing that it could also explain some of their 10nm problems.

  55. It’s not that crazy. TSMC is their preferred fab for products that aren’t necessarily beneficial using Intel’s own process. Most of their network products use TSMC as a fab.

    TSMC is actually a good choice as they are the only pure play foundry capable of producing the latest process at volume without much trouble.

  56. Jim Keller and raja are taking the two-pronged approach by targeting CPU and GPU divisions concurrently.

  57. Intel may have production troubles now, but it’s only a matter of time until AMD/Nvidia fall in the same trap. TSMC and [i<]maybe[/i<] Samsung are the only high power, bleeding edge chip manufacturers left. If either one so much as misses a breath, big, low margin parts like consumer GPUs and CPUs will get the short straw.

  58. Meanwhile, another price cut or fire sale from AMD as they try to clear unwanted stock…

    Womp! Womp!

  59. I think it’s got more to do with producing chipsets on the same process as their CPUs to comply with rules that Intel itself helped create (TR [url=<]wrote about it[/url<]). The result was going to be that TSMC would make Intel's chipsets which seemed crazy. Instead, Tom's Hardware says Intel is going to [url=,37819.html<]shift chipset production back to 22nm[/url<].

  60. Pretty sure the problem is Intel didn’t think they’d be this far behind in getting their new fab up and running. So they began winding down using the old process and, whoops, they hadn’t gotten the new one up yet.

    Man, Intel has fallen pretty far pretty fast in terms of their fab advantage.

  61. No, the problem with 10nm is that they play [url=<]this number[/url<] in the 10nm fabs. Not the right message to send there Intel!

  62. i really didn’t. I figured dosbox was sincere about the marketshare part. I definitely did not miss the Meltdown/Spectre reference.

  63. lol what, the 5-10 percent AMD MIGHT take over the course of the whole year? yeah, i’m sure they’re just shaking in their boots.