Look what the internet of things has wrought! Monday, Broadcom, which was bought in 2015 by Avago in a $ 37 billion acquisition, said it would spend up to $ 103 billion buying Qualcomm. Let’s not forget that Qualcomm is trying to close a $ 47 billion acquisition of NXP that should happen some time next year. Meanwhile, Intel and AMD have surprisingly decided to team up to rival Nvidia with a new graphics chip.
These partnerships and potential deals are an excellent example of the challenges that chipmakers face as computing and connectivity moves everywhere and becomes more commoditized. These are challenges caused by the growth of the internet of things.
The Broadcom takeover offer is an example of consolidation in several markets (communications, embedded computing and mobile) as prices for these components drop and markets shift. Meanwhile, the Intel deal signals Intel’s acceptance that general purpose compute can’t do everything as computing expands to more devices, and if it wants to succeed it has to embrace other architectures to retains its pricing power.
That’s the big picture, but there’s also the mundane facts of day-to-day life as a chip company driving these deals. Making chips is expensive, both in R&D and then in getting the parts designed and manufactured. As consolidation occurs, companies can combine R&D and business lines across many different companies, creating greater economies of scale. In chip-making and design that scale does matter.
Additionally, more and more companies are designing their own chips, whether it’s Apple in its mobile products or Microsoft for its servers. They do this because they have enough scale, and because the tiny tweaks they can make in silicon can differentiate their hardware or services in ways that leave the competition in the dust. Thus, the original chip vendors are left with a market that isn’t exactly shrinking, but one where if a customer succeeds, might graduate from their products.
Let’s hit the Broadcom takeover offer for Qualcomm first. For the last few years, the average selling prices of many of these chips make by Qualcomm, Broadcom, NXP and others have been heading lower and lower. While companies are selling more of them, they are also selling them at lower cost and at lower margins. This is good for the internet of things because it means adding intelligence into a device becomes cheaper, but it’s a double-edged sword.
Essentially as software started eating the world, the value now accrues to software, while the hardware that makes it possible becomes cheaper and almost interchangeable. That puts pressure on the chipmakers. Additionally, they too are getting more and more into building software to make popping their silicon into existing devices easier. A company like Whirlpool doesn’t want to spent its time designing boards or tweaking protocols. It wants to buy a product that “just works.”
That’s good for the customer and helps the market expand because you don’t have to be a firmware expert to design these chips into your products, but it’s expensive for the chipmakers, many of whom have more software engineers on staff than chip designers.
For Qualcomm there’s another challenge at play. Its efforts to swallow NXP (and CSR in 2014) were all about getting more chips that fit into automobiles, RFID networks and smart home devices because it was seeing its customer base for smartphone processors stagnate. It was attempting to move from the mobile world deeper into the embedded world — which is what NXP did when it acquired Freescale.
As companies like Apple, Samsung, and now, Google, design their own chips for their phones and devices, Qualcomm’s core application processor business is under threat. That’s why we see it seeking new markets such as drones and robotics that also require a bunch of brains at efficient power consumption.
As part of Broadcom, which also makes application processors, Wi-Fi, Bluetooth and other baseband chips, there’s a huge opportunity to combine communications product lines for servers, mobile and embedded devices. Broadcom, as part of Avago has deep ties in the embedded market through earlier acquisitions of HP’s Agilent and massive ties in the networking world with LSI, PLX Technologies and Emulex.
If we’re looking ahead we can even see that Broadcom buying Qualcomm cements its dominance in embedded and mobile, but it also begins to push it further into servers. Qualcomm is one of several companies trying to use the low-power ARM architecture to build servers that would compete with Intel’s x86 architecture that currently dominates. Qualcomm even has a joint venture in China to build such servers.
Before we get to other other big chip news of the day it’s worth adding that if Broadcom does end up with Qualcomm the big question is what happens to Qualcomm’s patents and licensing business? Activist investors have urged the company to sell the licensing division, which is currently part of of Qualcomm’s fight against Apple. Spinning that out could generate cash to cover the purchase, while giving Broadcom the chip businesses it wants. Whoever buys those patents (Apple has a lot of cash) could build their own networking chips for smartphones and connected devices.
Now, back to Intel: Intel and AMD are teaming up to put an AMD graphics core inside an Intel chip for notebook computers. This may not seem like a big deal, but it’s huge. Intel and AMD have been rivals since the creation of AMD. AMD has the only other license to make x86 chips and for decades it has lost money acting as a foil against Intel becoming a monopoly.
Don’t get me wrong. AMD has some awesomely smart engineers who have built technology that leapfrogged what Intel was offering at the time. But execution challenges, and even dirty practices from Intel always dogged it. AMD did see the importance of graphics processors early on. In 2006 it purchased ATI Technologies, which made graphic cards, and ended up with GPUs that would later help AMD stay competitive with Intel as parallel processing became more and more important in compute.
It even sold the mobile graphics division to Qualcomm, which then used it to build better graphics into its applications processors.
Intel is putting the AMD Radeon graphics tech inside an Intel Core chip designed for the notebook market in a deal that signals Intel’s acceptance of its lack of graphics horsepower. Intel tried to design a graphics chip back in 2008 but eventually gave up after realizing its architecture wasn’t competitive with AMD’s or Nvidia’s GPUs.
Mostly the Intel/AMD partnership is about the new Intel recognizing that the heyday of general purpose compute is over and that the x86 architecture can’t do everything, especially in a constrained power environment. Under CEO Brian Krzanich Intel had increasingly embraced the concept of heterogeneous architectures from custom-made chips for machine learning to the ARM architecture of mobile. So why wouldn’t it work with its former arch-rival?
After all, like every chip company in a world where non-custom silicon is everywhere and worth less and less, Intel has to survive. To do this, it has to make its chips work everywhere they can and ensure that they still sell for a premium.
At a macro level, both deals are a result of more computing in more places putting pressure on pricing, power consumption, as well as the shifting market for semiconductor companies who may see their customers graduate to making their own silicon as they succeed. Stuck in the middle, chip firms have to consolidate to survive.