Did 3D XPoint Costs Reach Break-Even?

Dollar sign superimposed over a 3D XPoint Optane DIMMDuring Intel’s latest earnings announcement the company provided information to indicate that 3D XPoint, which Intel sells under the name “Optane”, may have finally reached break-even: It may no longer be selling at a loss.

How would The Memory Guy know?  Well, in fact, I don’t, but I can make an informed guess.

The chart below shows profits for NAND flash makers from 2016 through the most recent quarter.  It’s a rough way of comparing major NAND flash manufacturers’ business.  As of August 14, all major NAND makers have reported their earnings for the second quarter of 2020, and that appears at the end of this chart.

Column Chart to compare NAND maker margins

There are a lot of side notes for the chart:

  • The margins shown for Samsung, SK hynix, and Micron are for DRAM and NAND flash combined
  • Operating margins are shown for most companies, but WDC only reports the gross margin for its flash business.
  • Intel’s margin is for its Non-volatile Solutions Group (NSG) and includes NAND flash and products based on 3D XPoint memory.

In brief, this chart is far from perfect.  Still, it’s a useful way to compare the companies.

The important take-away from this chart is that Intel’s NSG margins have been negative every quarter except for 4Q17, 3Q18, and last quarter, 2Q20.  During 2017 & 2018 the other NAND flash makers had impressive profits.  Why didn’t Intel?

The Memory Guy guesses that Intel was profiting from its NAND flash business as much as its peers, but the 3D XPoint Optane effort was sucking all of those profits out of the business, and more, to cause losses.

In fact, if you estimate that Intel’s NSG group’s NAND profit was equal to the average of its competitors, then you can calculate XPoint losses of about $2 billion for 2017, another $2 billion for 2018, and $1.5 billion for 2019!

If we zoom way in on this chart to show just the past two quarters we can see that Intel’s NSG profits for the second quarter are suddenly on a par with all other NAND flash makers.  This is the first time this has happened.

Column Chart to compare NAND maker marginsIn the second quarter of 2020, Intel’s NSG posted a profit of $322 million, giving it an operating margin of 19%.  This is close to the operating margins of the other companies in the chart:


Company Margin
Intel 19%
Samsung 30%
Micron 18%
SK hynix 23%
Kioxia 5.5%
WDC 31%
Average (w/o Intel) 21%

Since Intel’s NSG margin is now comparable to its peers, then it’s reasonable to assume that the group’s NAND profits are no longer being offset by Optane losses.

Even before Intel and Micron announced 3d XPoint memory five years ago The Memory Guy argued that a new memory would lose a lot of money until it ramped to volume.  We now have concrete proof of that since Intel has lost well over one billion dollars per year in 2017-2019 to bring 3D XPoint to the break-even point.

But there’s something confusing here.  When 3D XPoint memory was first announced partners Intel and Micron boasted that it was: “10 times as dense as today’s Conventional Memory,” meaning DRAM.  If it’s that dense, and if it only has to sell for half of DRAM’s price, then why did it only now become profitable after three years of billion-dollar losses?  That’s because of the economies of scale.  This is something that is explained in depth in a new report Objective Analysis and Coughlin Associates recently released: Emerging Memories Find Their Direction.  This mechanism is the major impediment to the acceptance of new memory technologies, and will be the single most important consideration as new memories vie to displace today’s standard technologies.

The economies of scale have allowed Intel to finally reach the break-even point, and from now on Optane is likely to either continue to break even or to make a profit.  This is enormously important if the company ever wants to spin off the Optane business.  A spin-off seems very likely since Intel has exited nearly every memory business it has participated in since its founding: SRAM, DRAM, EPROM, EEPROM, NOR flash, Bubble Memories, and PCM.  The only two left are NAND flash and 3D XPoint.

To get a solid understanding of how the economies of scale impact the market for new memory technologies please look into purchasing the report mentioned above: Emerging Memories Find Their Direction.  It can be purchased online for immediate download.  Find out how by clicking the link.  It’s just one more example of the ways that Objective Analysis helps our clients with their strategic planning efforts.

15 thoughts on “Did 3D XPoint Costs Reach Break-Even?”

      1. Thanks. So why did you list those separately in your comment about their spinning off virtually every memory business they have participated in?

        I.e., “A spin-off seems very likely since Intel has exited nearly every memory business it has participated in since its founding: SRAM, DRAM, EPROM, EEPROM, NOR flash, Bubble Memories, and PCM. The only two left are NAND flash and 3D XPoint.”

        1. Steve, That’s a fair comment.

          When Intel & STMicroelectronics created Numonyx in 2008 Intel tossed in their NOR flash and NOR-compatible PCM chip. That’s what I was counting as Intel’s exit from PCM.

          The announcement of 3D XPoint memory in 2015 came after Intel had stopped selling any PCM for 7 years.

          There’s room for debate about whether or not that is a real exit from the market!


  1. Would you happen to know the names of the businesses they have spun off? Or does “exited” just mean left the business without selling or spinning it off?


    1. Most of these businesses were not spun off. Resources were simply reassigned and the product was phased out.

      SRAM, DRAM, EPROM, and EEPROM all fit that description. So does Bubble Memory, but I don’t remember if Intel ever even commercialized that technology. PCM and NOR were indeed spun off as Numonyx in 2008.


    1. Lugang, The Kioxia cost argument was presented at IEDM or ISSCC by SK hynix a few years earlier. It misses a key point: 3D XPoint costs don’t have to compete against NAND flash, all they have to do is to be much lower than DRAM costs. Over the next decade I would expect for the price of 3D NAND to drop much faster than DRAM prices and for 3D XPoint prices to remain at half of the price of DRAM.

      I was actually surprised when the second-generation of 3D XPoint was announced as a 4-layer 20nm design – I thought it would be 2 layers and 15nm, since Micron’s Lehi fab already had lots of experience making 15nm NAND. Adding layers to a crosspoint memory is costly and difficult, as Kioxia pointed out.

      The PCM that 3D XPoint is made from can be scaled much smaller than DRAM or NAND flash, so that is the most likely path for future cost reductions. Back in 2004 someone from ECD (the real creator of ovonic memories, or PCM) showed me a photo of a working 3nm PCM bit. Back then 3nm was unimaginable. Now it’s a topic of much discussion at CMOS logic fabs. There are six process nodes between today’s 20nm 3D XPoint and a 3nm process.

      I have a lot of faith in 3D XPoint’s technology. The economics are the hard part! If Intel can make this part consistently profitable then it has a long life ahead.

      Thanks for joining the conversation, and for sharing the Tom’s Hardware link!


      1. Thanks Jim, your thorough explanations make a lot of sense.

        It’s interesting that Micron made no mention about 3D XPoint in their earnings call two days ago – I guess, given the somewhat dampened market outlook in the near term, they are still not in an environment to achieve the economies of scale required.

        Look forward to your next articles.

        1. Lugang, Thanks for the support!

          There was one mention that Micron made about 3D XPoint in their earnings callthis week. They said that the Lehi fab had $130M in under-utilization charges last quarter. That’s about the same as saying that they had to throw away $130M worth of 3D XPoint for one reason or another.

          I am sure that a part of that is their lack of a product that uses the chip, and another part is that some of the material didn’t work, or the fab wasn’t running at full capacity, or other similar start-up issues.

          We will never know the details, but from a very high level we know that they are still using Lehi to make it and they aren’t selling any of it yet.


Comments are closed.