This gives some insight in chip shortage in few industry
Most of the discussion here is about intel & AMD ā¦ But the biggest threat both of them face is the replacement of x86 architecture with ARM architecture in laptop/notepad cpus & the increasing prominence of GPU based datacenters in AI/ML applications .
Already apple has replaced intel CPUs from MACbook with thier M1 processors based on arm architecture.
In Windows based laptops , already Qualcomm have a arm based processor (not successful) & promised to have nuvia based processor by 2022 end.
Microsoft is already working on their own ARM based laptop for a few years.
ARM based laptops have a higher battery cycle ( battery lasts for a day or more after a recharge ) & provides almost same performance ( apple laptops have almost same performance as that of intel).
This shift to ARM based laptops is a big risk for valuations of intel & AMD as it will no longer be a 2 player market.
ARM will kill both of them in the next 10 years if they canāt move to better instruction set. Server computing is getting consolidated w/ large cloud companies, who all have their own ARM processors. For laptops Apple has already moved to ARM. Other OEMs will make the move too to not be left behind. Qualcomm already has a product. Where does that leave Intel and AMD?
Datacenter
I think I have given enough coverage on ARM inroads. But do note that if anyone read the beginning posts - I have tried to focus on datacenter - which is the cash cow for
current leaders - Intel/NVDA. That is the area where everyone is trying to get a hold on. ATleast any serious CPU/GPU company end goal is datacenter. I should have named this thread with ādatacenterā.
Let me clarify:
Short term incursion possiblity - AMD. Hence I focus a lot in this
Long term - ARM is a serious threat. Hence I have devoted some time/space for it. But I have informed about what to expect and at what stage ARM is. Not too much day today noise comes in this space. Also note that NVDA ARM acquisition is in process and ARM CEO has come out with some sorry picture if acquisition does not happen.
I am not talking laptops/desktops etc.
Just that my recent posts have been about INTEL/AMD since current fight is among these two in CPU space. Also, the NVDA acquisition story being still in process has prevented me from writing much about arm. Ampere has come up with great ARM server platforms and ARM based supercomputer fugaku is 3rd time top (Revised every 6 months). Apple could well be planning on deploying their ARM ISA CPU in their cloud infra.
Now, with the clarification for ālack of focus on ARMā out of the way let me come to the most important part I was expecting from readers in the know how of this industry. Low NVDA discussion. Not sure why someone did not complain about this
Some info to munch on
- NVDA profit last year was double ARMās revenue - ARM financials - https://group.softbank/en/system/files/financial-report_q4fy2020_01_en.pdf?file=pdf/ir/financials/financial_reports/financial-report_q4fy2020_01_en.pdf - page 38.
- ARM is facing/will face competition from RISC-V. Things get murky here and it becomes opinionated since things have only started to play out now.
- Also, do read the following interview of ARM CEO
The combination of Arm and NVIDIA is a better outcome than an IPO. The level of investment that will be needed to lead in AI will be unprecedented. Weāve been down this road before when we predicted a major market shift in 2016. We knew that we needed to invest heavily in our products, talent, and technology to take advantage of the opportunities ahead. The initial investments came when SoftBank acquired Arm; it enabled us to build new technologies that expanded our reach into data centers, the automotive and networking industries, all while retaining our leadership in mobile.
Now is the time for us to take our scale to the next level to address the technology challenges ahead. We contemplated an IPO but determined that the pressure to deliver short-term revenue growth and profitability would suffocate our ability to invest, expand, move fast and innovate. Combining with NVIDIA will give us the scale, resources and agility needed to maximize the opportunities ahead. This deal is the best opportunity for Arm and our customers and will enable the UK to be a meaningful industry player in the age of AI.
- They have put in a lot of money in āfutureā plans. Yet to fully monetize?
- Softbank will squeeze ARM if it goes to IPO
- CEO hinting AI race needs huge investments and not easy with IPO. He is not even talking about CPU licensing/royalty in datacenter in the future. Read on to understand more.
When people talk about arm, they talk about how ubiquitous its presence is going to be. But they do not understand that ARM sells CPU IP. It does sell some other IP - GPU/interconnect. But unlike nvda/intel, it does not yet have a full platform capability that is widely used. There are some solutions like Neoverse, but nothing compared to two behemoths.
Now add this up with what ARM CEO is saying - I suspect/feel royalty is drying up or something like that - I am not sure.
Finally - My picture of data center current and future is CPU shifting data in and out. GPU doing to hard work. Hence ARM does not occupy a lot of my mindspace. Offcourse, after NVDA aquires ARM, it becomes a different ballgame as it becomes part of a power AI/compute platform. Do read CEO interviews to understand the TAM capability here.
I hope I have given a perspective with major players covered. And no area missed (offcourse only datacenter)
Expecting NVDA results to follow similar trend.
By now, itās no secret that AMD is smashing Intel in retail desktop CPU sales, even after Intel started offering discounts on several of its 11th and 10th generation offerings. Team Redās processors dominate Amazonās best-selling charts and are now preferred by workstation builders. And Epyc processors are stealing Xeonās thunder in the data center, albeit slowly.
Beyond the x86 market, the Mercury Research report also includes an interesting bit about Arm-based CPUs, which are now estimated to account for 7 percent of all desktop CPU shipments, primarily thanks to Chromebooks and Appleās M1 Macs.
More and more articles about amd server market inroads, which is pretty much what made me start this thread an year ago. Server market.
https://seekingalpha.com/article/4456428-the-market-is-wrong-about-amd
The fastest-growing business for AMD is not the Computing and Graphics business, although it generates the largest revenue share, but the server business which relies on the relatively new EPYC processor series. AMD has made significant inroads in the server market since 2018, at the expense of Intel (INTC), and AMDās market share has seeing continual growth in 2021. Based off of information provided by highly respected research firm Mercury Research, AMD had a 9.5% market share in server CPUs in the second quarter. AMDās server market share increased 0.6 PP from the first quarterā¦ which itself saw the fastest growth since the second quarter of 2006. AMDās accelerating growth in the server market is predominantly driven by strong customer uptake of EPYC processors
.
I think one should look at semi equipment space to gain from chip shortage. One such company is ASML which has monopoly on EUV machines used for making advanced chips with better form factors. In fact itās solely responsible for keeping Mooreās law relevant. All chip designers/manufacturers including Intel/TSMC even Micron are clients of ASML. As per ASML, EUV machines (each machine costs $150-$160MN) are most complex systems ever built. Unfortunately EUV machines are in crosshairs between US and China tech war. China desperately needs this system to win tech war as per XIās stated policy. But US/Dutch govt have banned shipping these advanced systems to China so that US may remain leader in tech innovation. Chip shortage has created earning visibility and enough tailwind for ASML to compound revenue/EPS growth with minimum 20% clip in foreseeable future. Any good news on China ban may add more spice to its already multibagger return.
Intel is very likely making a comeback with alder lake.I expect they can last longer if a price war comes in (they are cash rich). Share price is yet to reflect this it seems. Combine this with their own fab (if the issues are fixed now) and you get a very potent company as history proves it.
Personally, the move to invest back in growth instead of generating FCF is a good thing. If what I gathered from my conversation with intel employees is right, they are going chiplets (tiles) (read above the thread to understand) in a big way and they are investing back into their own fab capabilities. Both are good things. World can use fab alternatives and also tiles should ease up physical constraints when laying out the design and make better use of fab capabilities.
Chiplets is not a new concept. The SoIC 3D fabric from TSMC leads the world in packaging of chiplets. The data rate to the interconnects leads the industry. Secondly, migration of customers from one fab to another is not as easy it sounds. Samsung and TSMC has been working with EDA vendors and customers for decades in fine tuning this ecosystem. Intel all the while has been a closed house. To build up this ecosystem and to make customers switch, it will need to highly subsidize so much on margins and if at all it will happen, it will happen long into future. Also Samsung or TSMC will not be mere spectators here. They are already leading Intel by a generation
A quick question. CEO mentions this is a long game. The rate at which intel is going, I do not see them matching tsmc (atleast in the nodes game) by 2023 assuming they skip a node. Is that even possible? or are we looking at 2025 and beyond?
Pat Gelsinger is a great CEO, but itās better to keep hopes under control. EUV machines any one with capital can buy, but mass production with good yeild is something else. Intel have changed their node naming strategy to keep things in perspective with other fabs. But overtaking TSMC in near future is close to impossible. 2025, I donāt know. This is what I donāt like about Intel. Big talks with long timelines before they do anything. They did it for 10nm. And the whole industry knows the outcome.
[Disclaimer: I have worked with both these firms and its only my personal perspective and data shared if any lies in public domain]
Thanks Neeru, wonderful and very useful insights. What are those companies that are going to grow their their base in 5-10 years from today ?
Global Foundaries (IPO is on the horizon)
TSMC
Samsung
Intel
Please share your picks and educate us why .
Many Thanks
I think we should bet on pick and shovels theme here. Companies like āASMLā will keep benefiting with the capex getting planned across semiconductor ecosystem. To be precise, there is absolutely no other firm which could manufacture high precision EUV machines which makes this transistor scaling possible. Any large capex for semiconductor fabs will bump up sales and profits for ASML. Being a monopoly, they have huge pricing power. You can also look into companies which are core part of semiconductor ecosystem like āApplied materialsā. EDA companies also will see traction due to need to improve PPA of advanced chips. CDNS and Synopsys fall into this category. Be it Intel, TSMC, Global Foundaries or Samsung, no one can play it without these guys
AMD datacenter snowball rolls onā¦
AMD - data center sales doubled yoy AMD Reports Third Quarter 2021 Financial Results :: Advanced Micro Devices, Inc. (AMD)
NVDA - 35% yoy NVIDIA Announces Financial Results for Second Quarter Fiscal 2022 | NVIDIA Newsroom
INTEL - 10% yoy Intel Reports Third-Quarter 2021 Financial Results :: Intel Corporation (INTC)
Intel appears to have a response in Alder lake. AMD appears (news leak from multiple sources online) to have cancelled on zen 3+ and started focussing on zen4. Competition is getting getting heated up but expect AMD to continue the momentum in datacenter atleast for an year (based on management commentary).
@neeru3855 Any views on ASM Technolgoies
I have started reading about the company, do they have any of the capability that can fit into Semiconductor value chain ?
I can see they are planning for 50cr Capex.
Thanks