HBM Market and Revenue (FMS 2024 Presentation)

 

HBM Memory Market at FMS 2024

Jim Handy and I are presenting at FMS (Future of Memory and Storage) on HBM market, AI datacenters and scenarios on what the future holds.

 

“Is HBM Memory Headed to New Heights, or is it just Hype?”

DRAM-103-1: Influence of AI on Memory Technology, Tues 3:10-3:40

Spoiler alert: There is a lot of hype and it is heading to new heights

We show the impact of AI on the HBM market, Multiple forecasts, and memory scenarios that happen with new and fast growing markets

 

 

Some Recent Data:

• HBM current market revenue estimates are 3B-20B+ per year (see graph)
• HBM revenue growth is forecast at 25% CAGR to 150%+ CAGR  
• NVIDIA revenue is growing at astounding rates. But other companies not so much
• Hyperscalers are spending TONs of money on AI servers AND still buying more traditional servers than 2023
• HBM bit price is ~6x DDR memory.  Bit cost is ~3x DDR Memory.  

Some Forecast Challenges:

HBM is now competitive. Solutions from Micron and Samsung joined Hynix dominance. Hynix recent earnings report shows they still are the largest supplier by far and are not planning to give up share

Competitive memory leads to cyclical shifts. This is due to lead times, factory builds, and inventory

• During hypergrowth, everyone over orders to cover possible growth. 
• Cutting CAGR by half (not even flat) will lead to Inventory going from 4 weeks to 20+ weeks. Then, most orders stop and the price drops. 
• The larger the forecast growth, the larger the problem

Update on Forecasts: 

We pulled recent forecasts from sources and compared them. We will discuss in more detail at FMS

(The identities of the analysts ... Including myself and Jim...  are being protected for now)

We definitely underestimated HBM sales in 2023 and 2024. We increased the forecast based on Q2 results. The forecasts show growth from everyone.

Regardless of growth of 50% or 150%, we would expect some correction sometime. Datacenter digestion, overproduction, pricing causing demand destruction, Price war among producers. We will discuss when, how much, and who might win the battle long term

 

Mark Webb

www.mkwventures.com

Intel Capacity by Fab Through 2030 (MKW VENTURES CONSULTING)

 

Intel Fab Capacity Expansion Scenarios (Option A)

One potential scenario based on our model and changes to capex and timing in last 6 months.

• Fab 52 runs production in 2026, Fab 62 year after
• Ohio starts production in 2028
• After delays, this is aligned to graph shown by technology at Foundry Day
• Assumes EUV tools installed in ~2026 in some 10/7 Fabs to support 4/3 Ramp.

• 20A/18A/14A fabs are D1+Fab52+Fab62+OHIO in this model

We have 2 addition scenarios based on TSMC use and whether Intel Foundry starts to ramp. 

Also we can discuss cost and revenue by Fab... including JV SCIP impact

Call or text to discuss.

We will be a Future of Memory and Storage (FMS) in Santa Clara Aug 5-8

Mark Webb

www.mkwventures.com

New and Emerging Memory Technology Status/FMS 2024

 

Status of New and Emerging Memories (Mark Webb, MKW Ventures Consulting LLC)

We have a Future of Memory and Storage/Flash Memory Summit Paper that we are presenting showing how chiplets change the equation on adding different types of memory to products. It allows designers to trade off cost and performance characteristic with more knobs. MRAM, NOR, DRAM, all on the same chip in the desired proportion with different processes.

Our presentation is “Memory Technologies : How Chiplets Change Everything”

FMS OMEM-203-1 Heterogeneous Solutions for Performance Session - Wed Aug 7, 3:00pm - 4:05pm Ballroom C

 

That said, we need to reiterate the opinion, backed by data, that emerging memories will not emerge to any material impact on the DRAM or NAND Market. Unlike some peoples predictions, the emerging memory market will not be $30B or even $3B in the next 5-10 years. It is well below $300M today

 

Emerging memories are possible in 3 areas:

• Niche products (which we will show can be integrated in a chiplet product)

• Embedded In a foundry process from TSMC or Global or Intel

• High volume  IF they require minimal (<10% of steps) changes to a NAND or DRAM flow.

 

Fourth Option:  They stagnate due to technical issues and lack of demand.

Costs for Memory Technologies

Quick Summary:

MRAM (STT): Available today. In embedded, replacement for E-flash if you need advanced process and embedded memory. Niche applications as discrete. Those applications will increase with chiplet use.

RRAM: Available today. May scale differently than MRAM and typically is slower. But has applications depending on cost scaling. Embedded, Chiplets, Niche applications. Some AI weighting use

PCM (1T1R): Available today.This is lower density (<1Gbyte). A technology that 50+ years old …. Typically used by companies who love PCM (ST, IBM, Micron) as they have decades of history of characterization.

Crosspoint technologies (1TnR): Optane fully developed the worlds best and most dense “faster than NAND, cheaper than DRAM” technology. The economics didn’t work and the demand was 10x lower than anticipated. Intel and Micron gave it a huge effort and we all appreciate the attempt. It’s not clear how to get these technologies to high volume IF they worked well.

FeRAM (high density): Was a potential future due to ability to integrate with DRAM flow. However all memory companies worked extensively on Hf based FeRAM products since 2020 and none were productized due to a variety of reasons.

Other New Materials: UltraRAM, New PCM materials, New RRAM materials/companies etc: These are possible but are 10 years from potential production. Unless implemented with 3D DRAM in 2028-2030, these are lower probability at this point. Our presentation on our website of  “Memory technology Product Lifecycle” give detailed timeline of how new memories come to production

 

Summary: There is no universal memory coming. We have SRAM, DRAM, NAND, NOR, MRAM, RRAM, PCM available today. Lets focus on integrating those as embedded or in chiplets.

I will be in Santa Clara Aug 5-8. Call or text to set up meetings at FMS or by Zoom. We will host some Zoom sessions during the conference timeframe. I can address all of the technologies one by one with costs and applications

Lots More info on our website www.mkwventures.com

Mark Webb

MKW Ventures Consulting LLC

www.mkwventures.com

 

Update on Intel External Foundry Revenue from Q1 2024

 

In December, I published my forecast for Intel External Foundry Revenue. These are sales of Wafers to other companies. Intel has shown "foundry" numbers of $200M-400M per quarter using the previous reporting structure.

I frequently pointed out that MOST of the revenue was from Mask operations and some of it was from packaging. So actual wafer sales of products was less that 50% of those numbers. Making Intel ~ the 20th largest wafer foundry in the world.

With the new foundry finances report out, Intel counts sales (at market price) to Internal groups as "revenue". This allows the product groups to have a P&L reflective of what it would be if they purchased from TSMC or another factory. Intel foundry takes the billions in losses from the fact that Intel costs are higher than TSMC prices. 

My worry was that NOW we wont ever be able to check external foundry wafer sales and it would all be hidden. But we are Blessed to have a 10Q from Intel

Intel Foundry Revenue in Q1: 

4.4B for total "sales" including external and Internal.

$27M in sales to external companies. This includes Assembly, Test, and Wafer Sales. $27M. 

In Q1 2023, the sales to external companies was $118M. 4 times the sales shown here

From the Intel 10Q

"Intel Foundry also includes certain third party foundry and assembly and test revenues from external customers that were $27 million in first three months of 2024 and $118 million in the first three months of 2023"

We can discuss what these numbers mean and what this changes in our model Contact us for more Info. 

Mark Webb

www.mkwventures.com

FIVE fun facts on HBM memory

 

                 As with all hot topics, there are facts and myths. Five HBM Facts:

•       HBM volume is increasing. It was 2% of bits and 11% of revenue. This will grow to approximately 10% of bits over the next 4 years based on current models. The supply growth has to be relatively steady (capacity/designs are HBM specific), but the demand growth will be wildly variable. There are impacts from this supply and demand disconnect.

•           HBM cost is ~3x DDR5 Cost (our model is 3.5x but who is counting?). HBM price is ~6x DDR5 Price. Gross margin per wafer, margin per COS, margin per revenue is higher. Due to overhead of HBM specific designs, testing, and manufacturing, Operating margin can be challenging and needs to be managed.

•       HBM is stacked, packaged, and tested as a unit before going to customer or their contractor for assembly in CoWoS/Foveros type assembly. The packaging of HBM is typically done by the memory company or their contractor. Some recent “cartoons” seem to show memory chip stacking as part of CoWoS process. It is not.

•       In Advanced Datacenter AI systems, Most of the memory bits are still DDR and most of the cost is now processor/HBM module (SemiAnalysis did a very nice presentation on all the details).

•       I think everyone would agree that HBM is THE competitive focus of all memory companies. Current APPROXIMATE models show Hynix with 55% share, Samsung with 35% share and Micron with less than 10% share. The battle to change these numbers is on with Micron looking to double its share and Samsung looking to take its historical position at #1. These battles will cause wild fluctuation in market share and pricing and revenue. You have been warned

 

We have all the numbers on price, cost, revenue and bit volumes. Contact us to discuss more

 

Mark Webb

www.mkwventures.com

Intel Foundry Margins Reported Today

 

Intel released its new reporting structure today. Intel Manufacturing and Intel Product Groups are broken out. The breakout assume that Business units pay market price (TSMC price) rather than Intel cost. 

There are tons of details and specifics in the report out that reflect on the issues. High level takeaways:

1) Intel Manufacturing Costs are much higher than TSMC Price. Intel Costs are 2x the costs of TSMC. There is a reason why everyone uses TSMC and why everyone struggles to compete with leading edge foundries. 

2) Intel Product group has reasonable margins. The pull down in margins was due to using Intel manufacturing. Intel Manufacturing loses 5-7B per year

3) Intel HAD to break out the groups. Otherwise no business unit would ever buy from Intel manufacturing since the costs are so much higher. Reminder: Intel BUs were told they could choose Internal or external for Fabs instead of being forced to use Intel. 

4) The numbers are pretty large. Intel Manufacturing loses 5-7B per year. That is the difference between Intel Cost and TSMC PRICE.  Manufacturing Gross Margins are negative (TSMC Gross margins are 50%+)

5) Intel is starting to clarify that the "turn-around" is not really a 2024, 2025 or even 2026 solution. Pat stated that 18A ramps in 2026 (As we documented). 14A is a 2028 solution. Foundry sales ramp in 2027+. Foundry plans to be profitable by 2030. We can discuss this more with specific numbers

We gave the possible scenarios in a blog post last month (see link). These are still valid and perhaps more optimistic than we thought last month. Some specifics on what can be done:

1) Intel is not an efficient manufacturing company. Fabs are too small, they run too complex processes, they spend too much on engineering samples. This can be helped by doubling the volume with sales to external companies and bringing TSMC wafers home. This is why Intel is doing foundry. Need 2x the volume ASAP.

2) Intel will have leading edge processes again. Its one thing to be inefficient operationally with a leading tech (like in the 2000-2012 time). Being inefficient with n-2 technology is a death sentence.

3) Intel has now clarified the problem for all the public to see. Intel Manufacturing is too expensive. Old technologies have negative margins, New technologies have negative margins. This will eliminate the "business units are too demanding" excuse. Note: I am very well aware of the problems between Intel Fabs and Intel BUs. I also understand why BUs like TSMC. We can discuss this in detail and how Intel can fix this.

4) More volume, more efficient fabs, clear acceptance of the problem. The road is still extremely difficult but this is a start. Intel presented this as an opportunity "we can earn billions more per year if we just match other foundries" 

5) How this impacts the foundry business development will be interesting. 

Future blogs have the numbers and quantitative impact of the improvements the CFO mentioned. by group and by technology impacts. We also have breakout of Fab capacity. 

Call for more information

Links: 

Intel breakout: https://www.intc.com/filings-reports/all-sec-filings/content/0000050863-24-000068/0000050863-24-000068.pdf

Our website foundry Scenarios: www.mkwventures.com

Mark Webb

www.mkwventures.com

Intel Foundry Strategy and Future Scenarios Feb 2024

 

In December 2023, we published the Intel Revenue forecast for external wafer sales, gave a breakdown on how customers plan to ramp the foundry. The forecast is still valid (it assumes Intel executes perfectly on all plans) but since then we have a better understanding of Intel's strategy and scenarios that could unfold. The scenarios are based on Intel's strengths and weaknesses which are quite different than TSMC and quite different than what we expected 2-3 years ago.

Background:

In 2019-2021, it became clear that Intel was a distant follower to TSMC in technology and that they needed to catch up or just outsource everything to TSMC/Samsung/others. Intel BUs complained about technology delay and cost and wanted to work with TSMC.

• It seemed like Intel would move to outsource, but Pat changed the plans based on discussions in 2021. Intel would allow BUs to choose Internal or TSMC. They would (and still do) come up with dual sourcing options  until final decision later in the product development lifecycle

• Intel cannot lead in tech with the small scale of current Intel  (Times change, Intel is the third priority for equipment companies). Equipment vendors do much of the process and all of the tool development. You need scale to get their support. So Intel needs to offer foundry services to roughly double the scale of Intel wafer output. Intel needed to go “all in” on being a leading foundry.

• Pat [hypothetically] said:  …Business units say manufacturing is the problem. Manufacturing say BU is the problem. Fine …. Each of you can do what you want…. BUT we will make major decisions based on your execution.

Hence, where we are today: Intel is ramping TSMC on chips for processors of all types. Some leading products are 100% TSMC. And Intel is promoting foundry for others at the same time. 5 nodes in 4 years (not really, but that is a different report).

The BUs are extremely happy with this. So far multiple products have been moved to TSMC and the flexibility in using N5,N3,N2 is something they love. TSMC price is about the same as Intel's cost, so BU margins will increase.

** But how does Intel compete cost effectively with TSMC and ramp foundry and pay for all these fabs? **

We overlooked a some things until our IEDM discussions with various people in December 2023.

• Intel still wants to win and be better than TSMC. It seems unlikely… but it might not matter.

• The US government buys chips for internal products and DoD items. No strategic DoD product has TSMC parts in it. TSMC does not meet the criteria. As a result, those products have technologies that are not close to leading edge. IBM (past), GF and other defense approved companies make chips for those products but they are nowhere near leading edge. They would love to use leading edge but they need a DoD approved US company. While DoD parts are relatively low volume, the government could expand this to any Government supply chain (they track detailed supply chain and factories for all parts). IRS, Social Security, etc. TSMC cannot fill this today and it would require massive regulation to even have Samsung US or TSMC US support it. Trust me, I have done the audits with government products before, it can be extremely painful.

 Also, While Intel is not set up from a scale or from a cultural perspective to be a leader in cost,  US Government pays cost plus and incredibly high prices for products. Intel could have half filled fabs and still have great margins. You can see this at some government suppliers today

• One more thing could have been predicted but was missed. Leading edge is too expensive and complex. So many foundries…. GF, UMC, SMIC, Grace, Tower have no ability to provide leading edge or even 2 generations behind technology. Intel can partner with them, provide “more modern” technologies, provide scale etc. All companies not named TSMC or Samsung could GREATLY benefit from partnering with Intel and this allows them to compete with Samsung and TSMC.

Based on the above strategies. Intel could outsource most of its silicon to TSMC to keep the BUs happy and STILL be a leader in foundry just based on being the “US Fab company” and “advanced fabs to other foundries”. These customers are much more compatible with Intel than selling to Apple, AMD, Nvidia, Broadcom.

This is a different foundry model but one where Intel has a strength and can potentially dominate. This all may or may not work. We have quantitative milestones you can track to see if Intel is successful. 

The Three Potential Foundry Scenarios are:

• *Intel Foundry Success*: Intel has competitive processes at competitive prices and ramps up to be another dominant leading edge foundry. Intel is leader and Intel BUs use Intel processes. Revenue and profits grow dramatically

• *Intel fills TSMC gaps*: Intel supplies all other foundries, Intel supplies government. Both have few other options so they pay the price needed. Intel is 2nd source to TSMC. Revenue grows steadily over then next 10-15 years.

• *Intel is IDM2.0  …. IBM2.0*: Intel struggles to ramp government work and factories. Intel’s foundry partners decide it’s not worth it to work with them and the processes are unsuccessful. The fabs are given away, or cancelled, or underloaded. Eventually Intel foundry is absorbed.

We have more details on each and in the next few years, the probability of each scenario will change. We have updates on the probability and what tactics, models, and strategies Intel is using. More importantly we provide milestones so others can track progress.... and we track the impact to P&L and Capex

Foundry Day Update (BREAKING NEWS):  All of the presentations and commitments support the background we show, the strategies, and the scenarios. 

Mark Webb

www.mkwventures.com