This market research report was originally published at Yole Développement’s website. It is reprinted here with the permission of Yole Développement.
The current semiconductor growth trend does not resemble those that came before. Taking stock of the past cycles, at Yole Développement (Yole) we see a common theme: a certain application finds its place in society, and the OEMs and major semiconductor companies race to fill the massive, growing demand. The winners are those OEMs that offer the right product with the right specifications at the right price, and the chip vendors with design wins reap the benefits. See Intel during the PC boom of the 1990s or Qualcomm during the smartphone era of 2010-2015.
Today, instead of a key end system (like PC or smartphone) we have a key technology (AI) and many possible implementations that carry huge potential. Some AI implementations will follow familiar market characteristics; AI in datacenter, for example, has found a formidable acceleration engine in the general-purpose GPU, and those players who build GPUs certainly benefit. But AI has the potential to change so much more in all facets of life, whether in Automotive and Mobility, or Industrial and Retail, in Medical… really any application where contextual value can be added to sensed data is a candidate for deploying AI technology.
This analysis is well detailed in the Processor Quarterly Market Monitor powered by Yole. Q1-2022 update of this Quarterly Market Monitor. New material examining the trend of multi-chip solutions (like chiplets) in the CPU and GPU space… More info.
Some players are well-aligned to this dynamic (see Nvidia’s datacenter growth), but others like Intel have yet to really cash in on the new trend.
It is worth a look at how Intel has evolved in just the last 4 years. If we compare the Intel of today against the Intel of 2018, we see a company with similar revenue actualization in the core compute businesses centered on computing in PC and datacenter, but the strategic positioning of these two companies is quite different. The Intel of the late 20-teens looked to capitalize on the growing data center market as the dominant provider of MPUs. At the time, Intel still carried enterprise storage ambitions with the NVM business dominated by NAND, as part of a strategy to provide more of the datacenter hardware.
Today, the processor business is still critically important, and the desire to offer a broader DC hardware portfolio continues with acceleration solutions like IPU, accelerating network services, and Ponte Vecchio to accelerate AI and HPC workloads. However, the capital intensity and commoditized nature of the NAND business soured Intel’s appetite and thus that business was sold off to SK Hynix. Shortly after, Intel has embarked on IDM 2.0, with a renewed focus on the technology roadmap and new investments in the foundry business. Recommitting on the technology roadmap is way to strengthen the core microprocessor business, but also adds the new ambitions of Intel Foundry Services as a natural pivot given two realities: 1) global foundry capacity has been exposed to be too lean 2) foundry demand will keep growing as the AI+edge boom turns into the next semiconductor wave.
This brings us to consider the recent announced acquisition of Israeli-based foundry company Tower Semiconductor. In a move that surprised many at the time, Intel shared plans to buy Tower for approximately $5.4b. The synergies were not obvious given the diverged position of the two companies. Intel is a major provider of leading-edge processor & logic technology as an IDM and is looking to pivot to providing foundry services. Tower is small by comparison, existing as a pure foundry player with technology at the large-dimension end of the tech integration spectrum, more focused on specialty technology like analog and CIS. The two seem to have little overlap and left industry observers scratching their heads.
However, in hindsight, the deal makes good sense. Intel no doubt has great expectations around the growth of their foundry business. Intel is strong in the product development and manufacture of leading-edge CMOS products, which is a key ingredient for a successful foundry. But succeeding in foundry also requires expertise in foundry customer service and the kind of business planning that foundry customers expect. This is the business plan risk-mitigation that Tower can immediately bring to Intel.
Perhaps more important than mere risk-mitigation is how the combination of leading-edge CMOS and specialty analog can be key to the next major phase of semiconductor growth. Unlike the smartphone- and PC-centric semiconductor cycles of the past, the current cycle is based on the growth of AI applications at the edge and in the datacenter. This explosion of AI applications requires low power compute (at the edge) and high-performance compute (in the cloud/DC), plus the integration of evermore sensing at the edge, which requires the kind of technology in which Tower specializes.
The kinds of companies that will strike gold in this next phase of semi growth will come from a variety of business models. The edge AI applications that take off may come from one of the many small niche-focused startups or as an exploratory project from some of the untrenched big players. Of course, the ecosystem enablers like IP vendors, material vendors, and foundry services providers, stand to benefit, as long as they are aligned with the winning solution.
This is where Intel’s latest strategic moves start to really make sense. Revenue streams from Intel’s core CPU business have plateaued and the strength of PC-centric business is under some threat from non-x86 based architectures. So, to capture more market growth, especially in the coming edge-AI powered environment, Intel needs to position themselves in the critical path for the computing and sensing solutions that can win. Certainly, Intel has in-house activity in this realm, taking the neuromorphic computing project Loihi, for example. But Loihi might just catch fire for a few applications geared toward neuromorphic AI, or it may remain as a useful and interesting research vehicle. Intel has found edge AI success through M&A, with MobilEye as a prime example, which was acquired in 2017 for $15b and estimated to IPO at more than three times that value. However, if Intel can be the IP and foundry service provider to the next 10 or 100 winning edge-AI applications, then they will see revenue growth as this new wave of semi revenue builds. Bringing Tower into the fold is more likely to place Intel into that critical path of companies looking for a range of foundry and IP solutions to address this growing market.
The Intel of 2022 finds itself on a path, while wearing several different hats. Certainly, the cloud & data center age would not have blossomed without Intel, and that business will still carry the company and the industry forward. However, these core businesses are coming under some threat and Intel is looking to capitalize on the next major semiconductor growth in the edge & AI era.
Will Intel offer a compelling foundry and IP solution to capture the attention of the next fabless winners? Will Intel succeed in transitioning into the foundry model of customer service? Will companies who operate at some level of competition with Intel set this fact aside as they look for manufacturing services? In a market of tight foundry supply, some of these questions lose importance, but as new fabs rise to meet the demand, at Yole, we can expect to see whether the strategic posturing has put Intel on the right path, and how much traction Intel’s new foundry services have gained.
John Lorenz
Senior Technology and Market Analyst, Computing & Software
Semiconductor, Memory & Computing Division, Yole Développement
