Friday, January 27, 2023

Intel Earnings. What does the future REALLY look like?


Intel announced earnings for Q4 2022. The earnings were not great and the guidance was horrible.

Q1 2023 Projections

Revenue 11B, GM 39%, EPS of (-.15) . This INCLUDES the impact of Intel change to depreciation from 5 years to 8 Years in order to make earnings look better…. For the first 5 years. Then it becomes problematic


Intel continues to lose ground. Pat blames macro but Intel is performing much worse than the macro and worse than its main competitor. It is possible that the Intel (and AMD) sell in to the OEMS is much lower than OEM sales due to inventory burn off. .... But Intel said three months ago …. 1 month into the quarter…. That Q4 was low point. They were wrong, so we need ask whether Intel is losing share faster than expected. A look at GAAP impacts and businesses like the graphics group shows additional concern.

Intel’s plan is to spend a lot of money on product development while eliminating jobs. They claim they will build many fabs (6-8??) over the next 3-5 years. This is a great plan but it is not clear that any company can go from trailing in technology to leading while losing money, losing cash, losing market share, and losing employees.

So the question is: What is the Intel claimed future vs the realistic future?


What Intel Claims: 5 nodes in 4 years: Intel still claims to be on track. Since Intel 3 will not ship until Q4 2023 in volume at the earliest, It appears intel plans to ship and ramp 4 nodes in 2 years. This seems impossible, even though two are half nodes. Aside from process, that would require a collision of new product announcement and ramps.

What will happen: IF Intel executes flawlessly and they achieve they goal, the new nodes will ramp slowly and will not be more than 10% of the wafer starts at launch. This might be a good strategy, but it will lead to multiple parallel products on parallel nodes.... which is always difficult to manage.

If nodes or products are delayed, the roadmap and process ramp will fall apart. and this becomes a problem since Intel 4/3 are learning nodes for EUV and Intel20 is learning node for GAA. Short answer they must execute flawlessly (We can provide milestones and probability of success to see if this is happening and impacts if it does not.) But Intel 7 might be a dominant node like 14nm

What Intel claims: Intel will be number two foundry in world in 2025+. Intel want to build 6-8 factories to become a dominant foundry (in parallel with Arizona, Israel, Oregon internal factories). Intel has design discussions with multiple foundry customer and some test chips planned or running. Intel is building two factories now in Arizona and is doing site prep on Ohio. They have plans for Germany/Italy.

What will happen: The Plan was made to get major customers to sign with Intel and show they are serious (good idea). However it was high risk to begin with due to spending and the fact that Intel is not a major foundry today. Now that Intel financials are dropping, cash flow is negative, and the fact that you need to build it and spend billions before getting revenue.... it is not mathematically clear how this is possible. 

We have the details what the actual roadmap and timing for each fab will be (call us) ....  but when the smoke clears, expect ONE  factory in Arizona to tool out in 2024 and the rest to be slowed until Intel achieves revenue from foundry. Intel needs to get to be number five in the world (not an easy task) before being number two. IF everything goes well, Intel will add 1 wafer fab every year after 2025. This is about 25% of what they have promised multiple government entities.

We can discuss impact of Brookfield, subsidies and possible JVs to add more scenarios.


What Intel Claims: Cutting spending by 3B this year and 8B in 2025. If the plan is to change the accounting so that depreciation is now 8 years (Increasing cost dramatically years from now), then this is possible. If starting less silicon and lowering revenue is the way planned, this is possible. If taking all the write offs, reorgs at once, saying it is non-GAAP, then saying expenses are lower later…. This is possible. BUT actually reducing spending 3B while getting revenue to 2021 levels seems difficult. 

 *What will happen: Intel will cut headcount and some spending to claim 3B is reduced spending but they will not be able to show it on P&L as other spending went up. Expect big differences in GAAP vs non-GAAP (We can show the impact of layoffs on spending including severance pay and fab cost reductions). 

A positive note: What Intel needs to do to be successful:

Slowly move into foundry. Do actual foundry work before adding more fabs. Major customers are not going to run significant wafer starts at unproven foundry. Also, customers are allowed to cancel orders so have backup customers to fill factories like TSMC does

Get product roadmap competitive and hit all milestones. No changes or excuses

If you don’t succeed in graphics, stay out. Intel strengths (methodical and dominant leader) does not match graphics market. Focus on what you do well … processors and architecture for Datacenter and Client.

Be aggressive on Datacenter, paranoid on PCs. Datacenter will grow forever. PCs might hit 300M TAM.... but there is a good chance they will not.

Do this and return to be a profitable cash machine who leads in PC and Datacenter architecture.  

 We have lots of details and milestones so you can track Intel progress. Call or text for more info

Mark Webb

Monday, December 5, 2022

Emerging Memory Revenue Through 2032


Emerging Memory Revenue. Optane drops off and slowly MRAM grows. 

Emerging Memory Revenue will actually drop over the next few years as Intel Optane sales slow. Discrete revenue from MRAM  plus small contributions from FERAM and RRAM will slowly replace it. 

Emerging memories need to focus on two areas:

1) Embedded where they are BETTER than other memories. Replace ALL eFLASH now!

2) Niche markets where the speed of MRAM etc is BETTER than other memories. Again, NOR/eFLASH is a great target. Along with Caching for storage and LLC for compute systems.

There is one other possibility we continue to monitor this week. a new memory where the difference with existing memory process technology (DRAM, NOR, NAND) is less than 20%. If you can "bolt it on" it can work. this is how FLASH came from EPROM and how NAND came from NOR. IEDM is  great place to review this and ALL memory companies are trying their best to implement this.

Unfortunately, it is not financially possible to replace DRAM or NAND with "new memories". And most foundries and major memory companies already have their emerging memory candidates for the future.

Emerging memory is GREAT for new applications, NOR replacement and Cache. It is poor for trying to replace DRAM or NAND.

See our presentations at, Markets, revenue, and the lessons to be learned from Intel Optane.

Call or text to discuss!

Mark Webb

Wednesday, November 30, 2022

An Intel Foundry Business Scenario

What will Intel foundry business look like?

Intel broke ground on New Arizona Factories a year ago and followed with announcements for factories in Ohio, Germany and expansions at all existing factories. If we add up the Capex, we are talking $100B in the next 5-6 years and commitments to $200B in next 10+ years. This is all in addition to the ~20B per year to support new technologies in existing Fabs

Intel Stated that they will release 4 Nodes in 5 years and will ramp foundry to have billions in revenue in 2025 and compete with Samsung/TSMC in 2030

Intel announced partnerships and financing with investors and is looking for massive government support

This is great aspirational leadership by Pat Gelsinger. "We are the greatest Semiconductor company ever, we can dominate Foundry" Unfortunately there are some challenges.

1) Intel claims foundry revenue of $600M per year. This is not wafer foundry work. This includes mask making and tool sales. Actual wafer sales are less than $350M per year. Which means Intel will need to grow 10x to pass HH Grace in revenue to be in 6th place .... and it will be half the size of SMIC at that point. 

2) Intel is purchasing Tower which is a great company and will provide excellent mature foundry capability. But Intel wants to focus on leading edge which Tower never works on. 

3) Intel has committed the Fabs so customers can believe Intel could have tons of capacity. But to get that capacity intel needs to spend 20B+ before wafers start. If you build it, they will come. If they don't come, Intel will collapse. If you don't build it, they definitely wont come. Paradox

4) The is no chip shortage on advanced foundry. If Taiwan get invaded, the market either continues under PRC politics or Electronics collapses (Foundries exist other places... but systems and packaging assembly.... its mostly in China and Taiwan. If there is a China war, Intel could end up being the victim due to no PC, Server, networking upgrades (we all use old stuff)

5) Intel was counting on spending all its huge earnings on capex. Then the PC market dropped and Intel lost Datacenter Share to AMD. Cash Flow is VERY negative.

There are other challenges to Intel and insights that we can discuss in person or on phone.

So what will actually happen:

1) Intel will clear land and potentially build shells for new Fabs (6-8 of them). This will happen at a slow pace and trickle into 2024

2) Intel will try to convince customers to not only sign with Intel but actually pay money up front for capacity. Signing without volume commitment is easy. Apple qualifies and signs with suppliers all the time and then awards supply to others. Everyone needs a backup. Intel is a good backup. Why someone would pay in advance is not clear.

3) Assuming Intel delivers on Intel 4, 3, 20, People will commit small amount of volume to Intel sometime in 2024. The sum total will be about 25K wafers per month in 2025 up to 60K+ wafer per month in 2026. This is less than one fab. This is well below Intel forecast but is actually quite good for a new foundry. As usual, that is the issue.... separating reality from hype.

4) One Fab will ramp in 2025 with others delayed and on hold. 

5) Intel will ramp other sites and fabs IF business continues to grow. Intel will cancel/delay indefinitely over half the fabs and or sites. Intel needs to grow 10X faster than TSMC or Samsung. TSMC and Samsung Factory ramps in US are good benchmark to measure Intel against

What would my my scenario wrong and Intel successful???

1) Intel actually executes on Intel 4,3,20 on time with leading cost. It is better Faster Cheaper than TSMC.

2) TSMC/Samsung is unable to deliver for companies at the leading nodes on time

3) Government passes laws requiring people to buy from Intel. 

These are just scenarios and far from what Intel claims will happen. BUT...

That's the beauty. WE provide milestones so YOU can decide if Intel is on track. You can judge whether people are committing to Intel or not. We even throw in timeline tracking of Intel factory builds and pictures and tool installs 

As a Employee of Intel for over two decades, I would love to be wrong and see Intel dominate technology like they did in the old days (ie when I was there).  

Set up a meeting and we can discuss how this will play out and we can provide you with milestones to track. We are at IEDM all week.... call or text

Mark Webb

Emerging Memories Enter the Trough


Emerging Memories have "emerged".... into the Trough Of Disillusionment. It was a fun ride until now.

Less papers on new memories, Optane cancellation, and multiple options allowed by CXL for NAND and DRAM and shared resources have stamped out the enthusiasm for emerging memories. 

What we know: 

1) Optane was the most successful emerging memory in terms of development and revenue. 100s of Millions in revenue was more that all other emerging memory combined.

2) The applications were niche. Too small and narrow to allow widespread development, Capex, and support. Slower than DRAM, More expensive than NAND didn't work. Optane/3D Xpoint was abandoned

3) MRAM has small markets (<$100M/year) and great applications for discrete chips. Companies have viable business models TODAY and moderate growth. it could reach $800M by 2031 if all goes very well.

4) MRAM (and RRAM) are great for embedded memory and should replace embedded flash in all applications and should enable optimization on SOCs. The memory is <10% of the chip area typically and we can't track revenue just like we don't track revenue for eDRAM, SRAM, or registers. 

5) Other Memories (DNA, NRAM, FeRAM, SOT, etc) are 10 years from volume. None of these will replace DRAM or NAND. Only possible caveat is IF new memory is added existing NAND or DRAM architecture with <20% Fab process changes. Many are working on this but there are no wins yet.

CONCLUSION: So with Optane ramping down, Emerging memory from discrete will DROP over the next 5 years. After that if MRAM (and RRAM) grow as expected and applications emerge, Emerging Memory could become a 1-2 Billion dollar business buy 2031.While embedded revenue is not meaningful, we do expect 10% of foundry parts >14nm could have embedded MRAM or RRAM in them. This will not be for leading edge chips. If you allocated Area of chip*number of chips*price per chip one guesstimate would be $600M in revenue by 2031. 

Emerging Memories could hit $2-3Billion in revenue if we include all sources and all embedded by 2031. 

Intel invested big, took the risk for all of us. It didnt work out. Follow on memories may not be as lucky as Optane.

What will happen and allow growth:

1) Niche markets are great, grow moderately with solid applications 

2) Spend capital if the market takes off. 

3) Look to replace NOR, not NAND and DRAM

We have scenarios for each memory, what it takes and where the researchers are.... and what they need to do to be successful ..... Set up time with us to discuss next week

Mark Webb

Thursday, November 17, 2022

Mark Webb/MKW Ventures at IEDM Dec 3-8

 We will be at International Electron Device Meeting Dec 3-8. Please contact us to set up 1:1 meetings 

We will have updates on some diverse topics:

1) Future of emerging memories ... will anything emerge? how can we monitor this? (see FMS Presentation and we have updates since then)

2) Intel Plans for Foundry and whether Intel can compete on technology and cost. Intel wafer cost compared to other companies. Intel technology roadmap challenges

3) Intel Fab builds. Which fabs will get built and which ones will not get built (Hint, they are not all getting built)

4) The Chips Act: Where will the money go and who will benefit. 

5) TSMC: What will happen in Taiwan (Geopolitical scenarios) and what will happen in US.

6) Shortages, Surpluses, Lowering wafer starts, building new factories. What is really happening on wafer capacity and new fabs (Hint: Global economy is NOT the biggest issue)

7) Back to memories: Who is the real leader in DRAM and NAND memory.... Technology, cost, growth.

8) China Companies, Factories in China, Trade restriction realities. China needs to US. The US needs China.... what is going to happen here?

Mark Webb

Thursday, September 29, 2022

Micron FQ4 Earnings and What to Expect in 2023


I tried to warn people. The guidance is far worse than analysts predicted

Micron FQ4 2022 Results

Micron reported results today for FQ4. The results were inline with the previously lowered expectations.

The guidance was not expected by most analysts. Micron forecast FQ1 earning per share with a loss in GAAP and expectation of 0.04/share Non-GAAP (not accepted accounting principles). This is one of the largest drop offs in micron history.

As presented to our clients in presentations over the past month, the price drops and inventory adjustments are far worse than Micron and others predicted

As mentioned before, every company claims to have long term agreements… but these do not affect actual pricing or prevent crashes. They are mostly a sales pitch to analysts. Micron is not seeing stability.

Micron is on leading edge with technology currently but will drastically slow ramps of new technology. This is very wise and is  the BEST way to control bit growth. This is not good for equipment companies.

Micron cited numerous macro events. But inventory adjustment happens every 4 years regardless of news and the claimed shortages in chips caused un-necessary inventory build ups at end customers. The cause of this was the boom of the past 12 months. It will happen again in 4 years regardless of the world news.

Micron predicted strong second half of F2023. There is no data to support this but historically we expect 3 quarters of downside to get to target inventory, 2-3Q of average after inventory is corrected and then “the good times” with price increases and a boom.


How to see when this will turnaround and how you can monitor?

1)     Look at inventory for suppliers and customers. Suppliers report their inventory

2)     While Micron and Hynix are nice companies, the fact is that pricing and inventory is dominated by Samsung. Check Samsung models and Samsung capacity actions

3)     Look at pricing models. We can provide inputs on pricing. If pricing drops, there is excess supply. It is a commodity market and differentiation is minimal.

4)     As in the past, customers will run their inventory too low and then there will be a shortage and the upturn will happen (in 2024 timeframe)

5)     In the last upturn, MU Price increased months before the market recovered …. It anticipated the recovery…. But this made it difficult to set a target price AFTER the market recovered. End result was sell side analysts having targets of $130+ when all of the good news was fully booked in at $95. We expect the same overstatement of price targets to happen again in late 2023 into 2024

6)     We provide monthly updates on DRAM and NAND both from markets and from technologies and costs. Contact us for more information

Mark Webb

Thursday, September 22, 2022

Memory Downturn... this time is not different... or is it?


Every time there is a memory downturn, people come up with 3-5 news items from the past 6 months to say why it happened. War, covid, hyperinflation, elections, FBI raids, interest rates. except we can predict the downturn accurate to 6 months ....  2-3 years ahead of time (I have an excel macro that predicts this based on lead time, inventory, and capacity lead times)

Every 4 years, the market downturns. This always happens after a strong increase and the crash is worse when there is a special cause to the increase or when someone says "this time is different"

DRAM memory sales increase at 15-18% like always .... steady state.

Server and PC demand tick up a little. need to react ASAP

Supply chain issue appear in the news

Buyers increase inventory and builds. but they show increasing demand so weeks of inventory looks OK

wait six months.....

Buyers decide we MIGHT have a slowdown... since life is cyclical. they cut to normal inventory and normal growth. 

Buyers decide we definitely have a slowdown. "why am I paying $$ for memory when the price will drop 20% in next six months? we already have 3 months of inventory!" .... "Cut orders 50% for the next 4 months. That will get inventory on track and save us lots of money." 

Memory suppliers say "but wait.... if we give you 20% off, will you buy some now? we gotta sell them somewhere or we will crash.

Bits crash, prices crash. and if the price is dropping 10% per quarter AND the supplier has inventory, buyer says "Why am I buying anything? I can do just in time ordering at super low prices" 

This lasts until there is a small uptick (6-12 months) and Customers realize they have no inventory. and the opposite happens. 

So how does one deal with this as a 

1) Memory buyer?

2) Memory supplier?

3) Investor?

There are 3 REAL differences from previous cycles that influence next steps. 

Call/text us know to discuss what will happen in the next 6-12 months. We will post follow up articles over the next 2 months

Mark Webb