Intel: Fighting for its Life… and Losing
It has been a bad year for Intel. Despite the Silicon Valley Giants near record-high stock price, Intel could be on the defensive and playing catchup in the industry it once dominated.
A little background: Intel is a semiconductor chip design and manufacturing company with a variety of products including but not limited to CPUs, SSDs, desktops, and laptops.
So, how exactly did Intel manage to bungle utter dominance in the semiconductor sphere?
At its core, Intel is losing the battle in terms of design. But first, here are some definitions:
ISA (Instruction Set Architecture) - an abstract model of a computer; governs how hardware handles program execution.
CPU (Central Processing Unit or processor) - the brains of the computer; directs the rest of the computer to carry out a program.
ISA assembly - unique language that the ISA operates on; how the program tells the CPU to execute an input.
In 1978, Intel pioneered a revolutionary type of ISA called x86 and has been riding the success of this architecture ever since. These x86- based CPUs have been an industry standard for higher processing power sectors: PCs, laptops, and servers. However, the Arm ISA, invented in 1985, has dominated the lesser processing power sector of the market: mobile devices (tablets, mobile phones). Each architecture assembly language has its own unique idiosyncrasies, meaning they are not interchangeable. But the main difference between these two architectures is “x86 traditionally targets peak performance, [while] Arm [targets] energy efficiency” (Bilal 2020).
However, it has become increasingly apparent that x86 architecture is going out of style due to Arm showing significant progress in becoming the preferred architecture for all sectors. In fact, Ark Invest’s Big Ideas for 2021 outlines that Arm-based processors will be the dominating architecture by 2030 (Ark Invest 2021). This is mainly because Arm-based processors now have reached parity with x86 chips in processing power and at a much more efficient energy consumption rate. However, the shift to an all Arm-based CPU world has slowed due to the inability of application usage across the two different architectures. Since x86 has been the industry standard for years, developers need to either re-compile their applications to be compatible with Arm architecture, or have both types of code compiled into a new application (Gabeci 2020). These delays have not stopped some of the industry leaders from initiating the shift regardless. Apple and Microsoft, both long term partners of Intel, have recently shifted their chips from x86-based to Arm-based CPUs. Amazon, who used Intel CPUs for AWS server applications, has also shifted to developing their own Arm-based chips. Another benefit to developing chips in house, for many companies, is that they are able to make chips to precise custom specifications rather than wait for Intel or AMD to innovate. Additionally, they can also cut out the markup Intel likely issues on any of their chips, as they need to manufacture and design their own chips. If a company decides to just give the manufacturer a design, they could reduce transaction costs.
Consider Taiwan Semiconductor, a powerhouse squashing Intel in the manufacturing sphere. They were founded in 1987 as the world’s first dedicated semiconductor foundry or “fab”, meaning that they do not design their own chips, but rather produce other companies’ designs. Today, TSMC is one of two companies to produce the world’s most advanced chips (5nm) and are years ahead on Research and Development for 3nm and 2nm chips (Chen and Shen 2020; Owen 2020). Since 2014, TSMC has dominated over 50% of the global semiconductor market by maintaining around 500 clients, including some of the biggest tech companies in the world: Apple, AMD, Qualcomm, Nvidia (Sun 2020; Zafar 2021). Make sure to remember them.
Now, back to Intel’s problems, starting with Apple. Apple’s successful release of their M1 chip in late 2020 was one of the first instances of performance parity between an Arm-based chip and a x86 chip. However, this was a double blow as Intel lost Apple’s business as their chip supplier and Apple gave manufacturing duties to TSMC (Hruska 2020). Another reason that Apple finally decided to shift to their own in-house designed chips is because Intel has been lagging in manufacturing capability of leading edge technology.
Apple simply did not want to wait around.
More on that later. Furthermore, Apple also switched away from Intel in 2019 for 5G modems and then bought Intel’s modem business for one billion dollars. This purchase meant Apple would start developing their own modems and likely send business to TSMC (Gurman 2020). However, these transactions are all relatively low in overall impact on Intel’s bottom line, as Apple only makes up 1-2% of total revenue. Nevertheless, Apple’s willingness to shift away from Intel processors to Arm-based ones seems to be a trend that will continue throughout the PC industry.
Another player who has also tried to shift to Arm-based processors, albeit significantly less successful than Apple, is Microsoft. In 2020, Microsoft introduced their new product: the Surface Pro X with Microsoft SQ1 processor (Floyer 2020). The Microsoft SQ1 processor is based on and built in collaboration with Qualcomm, marking another shift from Intel to a direct customer of TSMC. The PC rollout was not smooth, but Microsoft have released a statement acknowledging the problems on their website (Microsoft 2020). Addressing these issues means that they have a clear path to continue rapid product development in the near future. Microsoft has also announced plans to transition all of their own data centers and PCs to Arm-based chips (Leswing 2020).
Microsoft’s announcement of shifting their data center processors to Arm-based technology was a continuation of a battle on another front for Intel. While Intel is slowly bleeding on the side of PCs as Apple and Microsoft start the shift to Arm, the obsolescence of x86 architecture could come much quicker in the data center and server market. There is one main reason: energy costs. Though Arm-based chips have only recently achieved comparable performance compared to x86-based chips, Arm-based chips have always been significantly more energy efficient. Why would a company running a data center want to spend more money on electricity when it could get the same results with less electricity? They would not. This rationale has not only driven Microsoft to transition away from x86 technology, but Amazon and Nvidia as well.
Back in 2019, Amazon launched a processor, Graviton2, that had 40% greater performance at 20% less power than comparable x86 chips (Dignan 2019). These chips were Arm-Based and had huge implications on the entire data center market, even without a broad release. AWS, according to Amazon, is “the world’s most comprehensive and broadly adopted cloud platform” given its wide range of abilities specifically, “comput[ing] power, database storage, content delivery and various other functionalities” (Menon 2017). Amazon Web Services is Amazon’s largest profit sector, so by designing their own chips they are vertically integrating their product, creating “hardware and software in a seamless experience” (Dignan 2019). They are able to control their margins even better, as Intel was essentially a middleman and cutting into potential profits. Amazon AWS switching to Arm-Based processors is a big deal.
If AWS, the largest cloud computing data center is confident in switching away from Intel’s x86-based CPUs, it would not be surprising if the other giants like Google Cloud and Microsoft Azure follow suit.
In April 2021, Nvidia took it even further, by announcing an Armbased chip called Grace and becoming a direct competitor to Intel’s server processor chips (Nellis 2021). Even with minimal information, investors were optimistic of Grace, as Intel’s stock plummeted 4.18% while Nvidia’s skyrocketed 5.62% (Nellis 2021). Jensen Huang, CEO of Nvidia, said Grace CPU was the “final piece of the puzzle” in Nvidia establishing the “basic building block of the modern data center” along with their GPUs and Network Processors. A GPU or Graphics Processing Unit, uses parallel processing to “break complex problems into thousands or millions of separate tasks and work them out at once” (Caulfield 2009). Nvidia has also been running relatively unopposed in a rapidly growing GPU sphere. Originally used for video games in order to get higher refresh rates and better graphics, there have been more and more applications for GPUs increasing prominence such as cryptocurrency mining and Artificial Intelligence development. This rise in GPUs and Nvidia’s dominance has led to a 82% market share, prompting Intel to release their own GPU in 2021, named the Intel Xe line (Mujtaba 2021; Gapo 2021). However, Intel is years behind in R&D while Nvidia continues to pump all of their manufacturing business to...you guessed it, TSMC.
Besides design, Intel is also lagging behind in manufacturing. A general rule of thumb when talking about semiconductors is that the smaller the distance between nodes the better, as smaller transistors use less power and perform more quickly. Right now, the industry leaders TSMC and Samsung are able to mass produce chips at 5nm technology capturing 55% and 16% market share respectively (Tsung 2020). However, Intel is still producing their chips at 14nm and are about to release 10nm chips this year (Alcorn 2021). Last July, Intel’s prospects of catching up with TSMC and Samsung went further south as they announced that their 7nm chips would be delayed until 2022 (Fox 2020). At that point, it seems as if TSMC and Samsung will be 2 generations ahead, likely being able to mass produce 3nm chips. As many industry leaders shift towards Arm-based chips, more companies will demand leading edge technology to ensure they can have the best products. This means “fabless” companies (companies that outsource their manufacturing) will turn to TSMC, and sometimes Samsung, to produce their chips, allowing them to grow further ahead of Intel in R&D for these leading edge chips. Furthermore, Intel’s utter lack of ability to deliver on their 7nm chips have made them temporarily a partly “fabless” company as they have to outsource their production of both CPUs and GPUs to TSMC and Samsung (Hwang 2021; Alcorn 2021).
This is getting increasingly embarrassing for Intel. In order to meet demand for their best chips, they must fund their direct competitors, thus weakening their manufacturing ability.
To compound on the problem, Intel’s bottom line will see significant risk factors in the coming few years. Over 60% of Intel’s 2020 Q4 revenue is in their Client Computing Group or CCG. The CCG is generally composed of “PC processor and related component sales,” which is the part of their revenue that is currently at risk in terms of design and manufacturing (Eassa 2017; Business Quant, n.d.). In 2016, Intel held around 80% of the market share of x86 based chips up until 2019. Since then, AMD has made a significant dent, now capturing upwards of 40% (PassMark Software 2021). Besides AMD, guess who stands to benefit from AMD doing well? If you guessed TSMC, you were right, as AMD outsources to TSMC.
At the end of 2020, the founder and owner of the hedge fund Third Point, who own over one billion of Intel stock, sent a letter to Intel’s Chairman of the Board, Omar Ishrak, included a list of changes that he believed would help Intel maintain their position as an industry leader. Among the key points:
TSMC’s and Samsung’s superior manufacturing processes have allowed for AMD to capture “meaningful market share in Intel’s core PC and data center CPU markets” and Nvidia to completely dominate GPU.
Because of the the dependence on TSMC and Samsung for manufacturing, “we fear that America’s access to leading edge semiconductor supply will erode, forcing the U.S. to rely more heavily on a geopolitically unstable East Asia to power everything from PCs to data centers to critical infrastructure and more.”
Stagnation of innovation at Intel cannot be reversed without the “best talent.” Intel has lost many of its most inspiring and talented chip designers and leaders, and our sources indicate that those who remain (several of whom are highly regarded in the industry) are becoming increasingly demoralized by the status quo.
Intel needs to be able to accommodate for manufacturing needs of design competitors; aka take back semiconductor foundry market share from TSMC and Samsung. “Intel must figure out how to serve its competitors as customers” (Loeb 2020).
Loeb’s letter to Intel spurred change at Intel. In February, Bob Swan, long criticized for stagnating innovation due to his financial background, stepped down as CEO and was replaced by Pat Gelsinger, a 30 year Intel veteran (Porter 2021).
So now, the Gelsinger Era begins and things are looking promising for Intel. He recently rolled out an ambitious plan of how to restore Intel as THE semiconductor industry leader. Three major points were emphasized:
Use outsourcing in order to catch up in the manufacturing realm and prevent further delays.
Keep most of the manufacturing in-house - 20 billion dollar investment to make 2 new “fabs” in Arizona.
Intel Foundry Services (IFS) - this would establish a division of Intel manufacturing to help fulfill orders for “fabless” companies and help with the current chip shortage (Duffy 2021).
Gelsinger’s third point directly addressed Third Points letter and seems like a good plan on paper. However, it may completely backfire. The Intel Foundry Services is not a new idea; Intel has tried it in the past, and failed. This could be further exacerbated by the fact that Intel is in a bind. Most of the money in the semiconductor foundry business is made on the leading edge, but Intel is unable to make anything 10nm and under. Intel’s plan does include a huge capital expenditure of twenty billion USD, but this will still not allow them to catch up to TSMC and Samsung’s manufacturing processes due to their lack of technology and research needed to compete with the already mass producing TSMC. Despite a twenty billion dollar capital expenditure for the development of two new “fabs”, this is still less than TSMC’s twenty-five to twenty-eight billion dollar current capital expenditure for 2021. Ignoring technological differences, all of the high volume fabless companies already book their orders with TSMC, and the ones that TSMC cannot produce go to Samsung. With these two giants already able to fulfill the needs of high volume clients, there would be little opportunity for Intel to gain any market share. As TSMC develops better processes and profits from a rise in order volumes, Intel will have no choice but to abandon their foundry plans. In fact, they might have to abandon manufacturing all together.
Only time will tell how Intel is able to address their plethora of problems, but major changes must be made in order to avoid the fall of a former giant.
Sources
Alcorn, Paul. 2020. “Intel Says First 10nm Desktop CPUs Land in Second Half of 2021.” tom's HARDWARE. https://www.tomshardware.com/news/intel-says-first-10nm-desktop-cpus-land-in-second-half-of-2021.
Alcorn, Paul. 2021. “Intel to Outsource Some Key CPU Production for 2023 Chips to TSMC.” tom's HARDWARE. https://www.tomshardware.com/news/intel-to-outsource-some-key-cpu-production-for-2023-chips.
Amazon AWS. n.d. “Cloud Computing with AWS.” AWS. https://aws.amazon.com/what-is-aws/.
Ark Invest. 2021. “Big Ideas 2021: Innovation research by ARK Invest.” Ark Invest. https://ark-invest.com/big-ideas-2021/?source=content_type%3Areact%7Cfirst_level_url%3Aarticle%7Csection%3Amain_content%7Cbutton%3Abody_link.
Bilal. 2020. “ARM VS X86: INSTRUCTION SETS, ARCHITECTURE, AND ALL KEY DIFFERENCES EXPLAINED.” Inter Press Service News Agency - Business. https://ipsnews.net/business/2020/06/29/arm-vs-x86-instruction-sets-architecture-and-all-key-differences-explained/.
Business Quant. n.d. “Intel’s Revenue by Segment (2016-2020).” Business Quant. https://businessquant.com/intel-revenue-by-segment.
Caulfield, Brian. 2009. “What’s the Difference Between a CPU and a GPU?” Nvidia. https://blogs.nvidia.com/blog/2009/12/16/whats-the-difference-between-a-cpu-and-a-gpu/.
Chen, Monica, and Jessie Shen. 2020. “TSMC 2nm GAA process development ahead of schedule.” DIGITIMES. https://www.digitimes.com/news/a20200921PD211.html.
Dignan, Larry. 2020. “AWS Graviton2: What it means for Arm in the data center, cloud, enterprise, AWS.” ZD Net. https://www.zdnet.com/article/aws-graviton2-what-it-means-for-arm-in-the-data-center-cloud-enterprise-aws/.
Duffy, Clare. 2021. “Intel investing $20 billion in new US chipmaking plants as part of turnaround plan.” CNN Business. https://www.cnn.com/2021/03/23/tech/intel-semiconductor-manufacturing-turnaround-plan/index.html.
Eassa, Ashraf. 2017. “What Is the Intel Client Computing Group (CCG)?” The Motley Fool. https://www.fool.com/investing/2017/12/24/what-is-the-intel-client-computing-group-ccg.aspx.
Fox, Chris. 2020. “Intel's next-generation 7nm chips delayed until 2022.” BBC. https://www.bbc.com/news/technology-53525710.
Gabeci, Kevin. 2020. “Intel x86 vs. ARM: Architecture and all key differences explained.” gitconnected. https://levelup.gitconnected.com/intel-x86-vs-arm-architecture-and-all-key-differences-explained-fb54a04788dc.
Gurman, Mark. 2020. “Apple Starts Work on Its Own Cellular Modem, Chip Chief Says.” Bloomberg. https://www.bloomberg.com/news/articles/2020-12-10/apple-starts-work-on-its-own-cellular-modem-chip-chief-says.
Hruska, Joel. 2020. “Apple Books TSMC’s Entire 5nm Production Capability.” Extreme Tech. https://www.extremetech.com/computing/315186-apple-books-tsmcs-entire-5nm-production-capability.
Hwang, Jeong-Soo. 2021. “Samsung wins Intel foundry order; TSMC takes GPU deal.” The Korea Economic Daily. https://www.kedglobal.com/newsView/ked202101210008.
Khan, Farhad A., and Daniel Loeb. 2020. “Third Point’s Full Letter To Intel.” Yahoo! Finance. https://finance.yahoo.com/news/third-point-full-letter-intel-220753103.html.
Leswing, Kif. 2020. “Apple is breaking a 15-year partnership with Intel on its Macs — here’s why.” CNBC. https://www.cnbc.com/2020/11/10/why-apple-is-breaking-a-15-year-partnership-with-intel-on-its-macs-.html.
Leswing, Kif. 2020. “Intel falls on report Microsoft plans to design own chips for PCs and servers.” CNBC. https://www.cnbc.com/2020/12/18/intel-falls-on-report-microsoft-will-design-own-chips-for-pcs-servers.html.
Menon, Sreerenjini. 2017. “AWS Services in simple terms.” ideamine. https://www.ideaminetech.com/blog/aws-services-in-simple-terms/.
Microsoft. 2020. “Surface Pro X app compatibility.” Microsoft. https://docs.microsoft.com/en-us/surface/surface-pro-arm-app-performance.
Nellis, Stephen. 2021. “Nvidia to directly challenge Intel with Arm-based 'Grace' server chip.” Yahoo! Finance. https://finance.yahoo.com/news/nvidia-announces-arm-based-grace-162709963.html.
Owen, Malcolm. 2020. “TSMC 3nm 'risk production' in 2021 paves the way to 2022 mass production.” Apple Insider. https://appleinsider.com/articles/20/07/19/tsmc-3nm-risk-production-in-2021-paves-the-way-to-2020-mass-production.
PassMark Software. 2021. “AMD vs Intel Market Share.” PassMark Software. https://www.cpubenchmark.net/market_share.html.
Porter, Jon. 2021. “Intel is replacing its CEO next month.” The Verge. https://www.theverge.com/2021/1/13/22228702/intel-ceo-bob-swan-pat-gelsinger-technical-financial.
Sun, Leo. 2020. “3 Tailwinds Are Lifting TSMC’s Stock to New All-Time Highs.” The Motley Fool. https://www.fool.com/investing/2020/07/31/3-tailwinds-are-lifting-tsmcs-stock-to-new-all-tim.aspx.
Tseng, Guan-Wei. 2020. “Total Revenue of Top 10 Foundries Expected to Increase by 18% YoY in 4Q20 While UMC Overtakes GlobalFoundries for Third Place, Says TrendForce.” TrendForce. https://www.trendforce.com/presscenter/news/20201207-10587.html.
Zafar, Ramish. 2021. “TSMC Earned $1,634 Revenue/Wafer In 2020 With A 54% Global Market Share.” WCCFTech. https://wccftech.com/tsmc-earned-1634-revenue-wafer-in-2020-with-a-54-global-market-share/.