Intel has enjoyed quite the run since the company introduced the Conroe and Merom 65nm processors in 2006. At that time, Intel was taking a beating from AMD, and the company needed a “Hail Mary.” For the last ten years, Intel has enjoyed a mostly unchallenged position in the processor market, and it has served the company and its investors well. However, ever since the company has struggled to deliver 10nm and AMD’s resurgence, some of the company’s technological weaknesses have started to become more apparent. Intel’s 7nm delay isn’t the first time the company has had to delay a process node, but it is the most crucial time for the company to make such an announcement with AMD and Qualcomm attacking them from both sides (highest performance and lowest power). I also believe this latest market selloff was the message that investors don’t know what to trust. So, naturally, it has been quite easy for everyone to pile on Intel and claim that they are “done.” However, things are much more complicated than that; they always are.
Keep in mind that not all “nanometers” are created equal. There was a day, let’s say 28nm when most of the transistor gaps were 28nm. This is no longer the case as marketing teams now arbitrarily name different technologies by some fake “nm” designation. Therefore, Intel’s 10nm looks very similar to TSMC’s 7nm high-performance process. Keep that in mind as many uninformed make this direct comparison off the simple “nm” designation.
Intel had some issues with 10nm and is finally shipping 10nm across most of the company’s portfolio of products and, during its most recent earnings, has shown itself to be profitable doing so, too. Not only that, but Intel had demonstrated that it could really get some miles out of a process node as the company did out of 14nm when they ran into 10nm delays. So, during that same earnings call, Intel cautioned that their upcoming 7nm process node for processors might be delayed by up to 12 months, which obviously puts the company at a competitive disadvantage. This is especially true when you consider that their competitors are already at 7nm, but at the same time, Intel is competing quite well with AMD on peak performance with 10nm parts. AMD has taken the multi-threaded crown from Intel on high-end desktop and workstation but still has significant inroads in the platform space, including Wi-Fi 6, NVMe SSD storage, and Thunderbolt.
Intel’s challenges in data center in the last few years have led to AMD’s growth in data center just as AMD finally introduced the Epyc line of products that gave Intel a run for their money. AMD is not alone in challenging Intel in the server space, as Intel also has competition from accelerators from NVIDIA and Xilinx as well as ARM server vendors like Ampere, AWS Graviton, HiSilicon Kunpeng, and many others. However, there are still going to be a lot of customers who will continue to buy Intel processors because of their market share and familiarity with the product and pricing. Incumbency matters here, and the past decade, Intel has been the “safe bet.” This has always been AMD’s challenge, and Intel has much ,more profound relationships and longer-standing relationships with OEMs, cloud players, and enterprises, even if Intel has has had some challenges supplying them with enough chips. Intel has seen three straight quarters with giant, $600M upsides from cloud and carrier on its 14nm process, and AMD has not seen any of that business, yet. I believe that if Intel can continue to improve their supply situation, the company will continue to grow and be profitable, especially with technologies like 5G increasing demand for infrastructure spending, as we noted in our paper a couple of years ago. Servers and HPC are always growing, which allows for room for AMD and Intel to both experience growth, but 2021 will be when we start to see the consequences of Intel’s delays, according to Serve The Home in the datacenter. However, Intel has a lot of additional compute accelerators with the acquisition of companies like Habana and Altera and accelerating AI capabilities into their existing processors, which should allow the company to extend performance in ways other than just increasing clock speeds and core counts. I believe that in areas like AI and 5G, Intel has the competition beat, which are both very high growth areas.
On the consumer side, Intel has already built-in some mitigations to these delays as a result of the 10nm delays. More specifically, packaging technologies like FOVEROS, which allow for chip stacking, are enabling chips like Lakefield where different types of cores are being utilized together while separating CPU cores from other components of the processor to optimize for yield, performance, and cost. A key element in this design is the utilization of new Atom cores codenamed Tremont, which we wrote a paper about last year. Combining low-power cores with improved performance with high-performance cores optimized for lower power is how I believe that Intel will be able to weather the storm of not having a process node advantage for the foreseeable future. Additionally, many applications are still optimized for Intel’s processors, which means many applications will still favor Intel processors and their higher clock speed. Like in the data center, Intel has a much more mature AI play with their processors that I believe will pay dividends in the future as more software starts to embrace AI. After all, the CPU is usually the lowest common denominator for most software developers, including AI developers, when they don’t always know what GPU, DSP, or accelerator a platform will have.
On the topic of Intel using TSMC, it is not quite news that Intel already uses foundries for some chips like some chipsets, network accelerators, and some Altera FPGAs. However, Intel wants to expand its capacity at TSMC to be able to manufacture more chips and take advantage of leading process nodes when possible. However, I don’t see Intel actually making much of a move over to TSMC or getting much capacity because TSMC has limited capacity and has to balance that capacity across multiple customers, including Apple, Qualcomm, Nvidia and AMD and even departing customer Huawei. Intel may still look for capacity elsewhere, like Samsung, which is an alternative to TSMC. I honestly do not see Intel getting too much capacity from TSMC. I believe that too many of TSMC’s customers have been Intel’s adversaries for so long that it would hurt their relationship with longer-term customers to give too much capacity to Intel. Intel is never going to fully move over to TSMC, and there’s no way TSMC could handle their capacity, so the reality is that Intel is going to move over to TSMC what it can and need to move over there until Intel gets its 7nm house in order. Fundamentally, Intel and TSMC are still competitors, even if Intel is a TSMC customer, and it isn’t in TSMC’s best interest to cater to them more than any other of their existing customers because Intel is likely to dump TSMC as soon as it can. What is interesting is that I believe the company may try to manufacture the company’s first discrete GPUs at TSMC because it is expected to be leading-node and probably somewhat low-volume, which would lend well to a new chip. Additionally, TSMC has lots of experience manufacturing discrete GPUs for Nvidia and AMD, so TSMC probably has some valuable libraries that could prove useful.
Intel has suffered from a bit of an identity crisis over the past few years after spending over a decade on top. The company has hired a lot of brilliant engineers over the past few years and has reorganized the structure of the company’s top engineering talent. One of those significant reorganizations is that the company has decided to part ways with Murthy Renduchintala, which means that five different engineering organizations will now report directly to CEO Bob Swan. I believe that this move is designed to signal that Murthy is being held accountable for the 7nm delay. Many people are being promoted into these new leadership roles, while Raja Koduri and Dr. Randhir Thakur will continue to head their organizations. That said, there are still a lot of questions about Intel’s future, and we hope to see them get answered in the next 12 months as the company navigates yet another challenging environment.
Note: Moor Insights & Strategy writers and editors may have contributed to this article.