Wednesday, August 20, 2025

Five Reasons for the Coming Correction in Memory

.

 


We can discuss the gory details why each of these is true and how you can see it coming. I am not saying it will happen, I am just saying all of the causes for a correction are lining up.

  • Memory is a cyclical market. This is due to Supply, not demand. Long term demand growth is steady but oversupply two years after the shortage causes the crash
    •  It is a 4 year cycle. 2026 is the year.
    • The issue will be due to too much inventory and modest (not negative growth)
  • AI is due for a correction (no necessarily Crash).
    •  2024, 2025 had 100% HBM revenue growth. AI server sales rose 50-100%
    • All forecasts are for growth half that rate or less in 2026. Still good but very different. Behaviors (“We just need to keep ordering”) will change
    • A demand change to only 25% CAGR growth would cause months or quarters of excess inventory. Listen for term “Digestion Phase”
  •  The worlds largest and most dominant supplier of memory has struggled 
    • Samsung is struggling …. Especially on HBM. Especially with Nvidia.
    • Reasonable performance in HBM would lead to a huge increase in competition and supply
  • Supply is coming online. Companies are adding capacity for HBM and migrating processes on DDR5.       
    •  This will add 20% capacity in 2026
    • There is less flexibility between DDR and HBM to deal with demand changes
    • There is more inventory today than you might think
  • Sold out through 2026 doesn’t mean the units are sold. Those are “plans” or what we used to call reservations, they can be cancelled. Prices can be adjusted
    •  Non-cancellable contracts can be cancelled. In 2022, Micron revenue collapsed 6 months after saying they had long term agreements.
    • You negotiate to have them take some volume even though they need zero
    • Sold out …. becomes working inventory …. becomes 10 weeks …. becomes 26 weeks even when end customers demand did not drop.
Scenarios for HBM:

  • 50% CAGR is  trend …. We have been operating in hypergrowth
  •  FMS 2025 is our forecast. Moderate growth in 2026/7
  • 2026 Correction is if we see a correction as documented above

Mark Webb

Wednesday, August 6, 2025

HBM Memory Future: Technology, Cost, and Market Size (FMS)


We are presenting this week on HBM market, Pricing and Costs


 HBM Memory Future: Technology, Cost, and Market Size

DRAM-302-1: HBM: DRAM's Bright Future. Thursday 8/7 9:45

We will show HBM market growth predictions and Scenarios, Technologies and Costs. Also show how to detect and respond when the market corrects and the hype turns negative

Some Key Points:

1) Market has grown 100% CAGR in 2024 and 2025 to $33B in Revenue

2) CAGR will slow to 20% for next two years and then grow at longer term rate of 25%

3) With capacity additions the market dynamic will change from severe shortage to more sufficient

4) Gross Margins for HBM currently in 70% range

5) Costs will increase due to technology on HBM4 but that increase will be offset somewhat with improvements in manufacturing and scale


Mark Webb

www.mkwventures.com



Monday, August 4, 2025

Future Scenarios for Emerging Memory Markets (FMS 2025)

 

Future Scenarios for Emerging Memory Markets

OMEM-202-1: Emerging Memory Technologies. Wed 8/6 9:45AM

We show the status today for memories (RRAM, MRAM, PCM, FRAM) and what will happen in the near future. Spoiler alert, There is only ONE new memory that COULD be a major impactor in the next 3-5 years. We show how to monitor progress on all memories.

 

Key Takeaways:

  1. The “Classic” emerging memories RRAM, MRAM, PCM, FE RAM are running and being used in low volume (<$100M Sales) markets as discrete products. They are also used in foundry technologies replacing eFLASH
  2. The “New Breakthrough” emerging memories are quite early in their development cycle (we give exact timelines all memories must go through in our product life cycle report). Single cell reports with great latency and 10^9 type cycling levels. Until we see a 1Mbit+ array with all performance and bit error rates, These are still  8 years from any possible volume and most likely will not be shipping any units.
  3. As predicted, the only technology with a possibility sell $1B worth of units in annually in 5 years must be close to (90% the same steps) of an existing NAND or DRAM process. It must also be within 50% of the cost of the technology.
    • This appears to be the plan with FE NVDRAM as reported by Micron in technical papers and more recently by FMC in announcements. All companies are working on this technology
    • FE technology, tied to a DRAM process, with tools similar to other high volume tools
    • The key is to see detailed array data and then a real product we can buy ASAP
  4. A new way to look at this revenue is by submarket, not to lump all emerging memory into one market. We break out these new memory markets

Summary: MRAM, RRAM, PCM and small FE arrays emerged a while ago. “New” memories are far from showing the required large array data. FE DRAM has large memory data and the manufacturability needed to become a high volume product

 Mark Webb

www.mkwventures.com



 

MKW Ventures Consulting at FMS Conference this week


  We are at Future of Memory and Storage Conference in Santa Clara This week. Please text to set up meetings or discussions. 




 We have two Presentations this week plus the Experts Table night.


1) Future Scenarios for Emerging Memory Markets

OMEM-202-1: Emerging Memory Technologies. Wed 8/6 9:45AM

We show the status today for memories (RRAM, MRAM, PCM, FRAM) and what will happen in the near future. Spoiler alert, There is only one new memory that COULD be an major impactor in the next 3-5 years. We show how to monitor progress on all memories.


2) HBM Memory Future: Technology, Cost, and Market Size

DRAM-302-1: HBM: DRAM's Bright Future. Thursday 8/7 9:45

We will show HBM market growth predictions and Scenarios, Technologies and Costs. Also show how to detect and respond when the market corrects and the hype turns negative


In NON-MEMORY areas. We are sharing models with clients for how Intel corporation can rebuild into a new successful company and how our predicted scenarios for Intel Foundry played out. The new Intel CEO will re-make Intel into a new company, not try to reclaim the 1990s Intel. We can review how this will play out for Intel


See all our presentations and data at www.mkwventures.com


Mark Webb

www.mkwventures.com


Friday, January 17, 2025

Forecast for Memory Market in 2025

 

The Memory Market in 2025. A Forecast

We looked at Q4 performance, Q1 guidance and some of the market forces. Plugged them into the memory model and came up with some numbers. There are some big trade offs and as a result, the numbers were surprising. This is for 2025 overall.

Caveat: I do not usually believe the “macro” normally affects memory market, but if the US continues to implement trade restrictions with China, That could single handedly put the entire tech market into a holding pattern. I am a free trader because free trade helps the market grow.



*All numbers normalized to 2024



The Guesstimates:

1)        DRAM Bits overall increase high teens percentage

2)        DDR Pricing decreases mid teens percentage

3)        HBM Bit volume increases roughly 90%

4)        HBM Bit pricing is flat

5)        DRAM ASP is roughly flat (HBM offsets DDR)

6)        DRAM Revenue increases mid teens percentage

7)        DRAM Operating Margin dollars increase mid teens percentage

8)        NAND Bits increase mid teens percentage

9)        NAND ASP decreases Mid teens percentage

10    NAND Revenue is flat to slightly down

 

End result: Expect big 3 memory company margins (dollars) to increase Mid teens percentage over the year… growth almost totally dominated by HBM.

The “cycle disrupting” product is HBM. It sucks up capex, sells for 6x bit price of DDR. Costs 3.3x the price of DDR bits. Potentially sucks up excess DRAM capacity. Adds lots of margin for ever bit converted from DDR. We expect supply to catch up by the end of 2025 and we will see what happens next in 2026

We have all the gory details and reasons in excel spreadsheets to look at the tradeoffs and some possible other  scenarios



Mark Webb

www.mkwventures.com