In 1991, my boss handed me a photocopy of an article: “The Computerless Computer Company”, by Andrew Rappaport and Shmuel Halevi, published in the Harvard Business Review. At the time, my boss and I were Motorolans – and this thinking didn’t quite fit our model of Six Sigma and surface mount manufacturing lines and accelerated life testing and real-time single-board computers.
But we could hear the train coming. In that article was a very prescient statement, one that got us wondering what the future might look like:
Implicit in most discussions of U.S. high technology is the assumption that a true computer company is one that manufactures systems hardware. We disagree. A computer company is the primary source of computing for its customers.
The context of this 1991 article was a world dotted with iconic hardware brands like Apple, Compaq, Dell, HP, IBM, Intel, Sun Microsystems, and the former Motorola including its microprocessor operation. In the wings waited a new breed of company, with different ideas about what computing would mean.
ARM had just been founded as a stand-alone entity from its Acorn roots, working with Apple on a processor for the Newton. Qualcomm had just gone public on the strength of its CDMA technology, and Nokia was about to embark on its first digital handheld GSM phone. TSMC and other foundry operations were still in the early days, before fabless became fabulous.
Software was also making a big run, riding along with every PC and Mac. Microsoft had made a huge splash with Windows 3.0, and Lotus 1-2-3 and WordPerfect were still household names. Oracle was becoming the standard in enterprise databases, and a tiny RTOS company named Wind River was making headway in embedded circles. A student named Linus was about to announce his operating system project which would later become known as Linux.
Amazon, Facebook, and Google hadn’t even been imagined yet. Jeff Bezos was working on trading systems on Wall Street, and wouldn’t pull his business plan together for another three years. Mark Zuckerberg had just turned seven, probably with a short list of real-world friends. Larry Page was entering his freshman year of college, a few years from a chance encounter with Sergei Brin.
We know what happened between then and now. Computing is no longer about just spreadsheets and word processing, but about browsing and streaming and blurting. The onslaught of mobile devices and their apps has transfixed the PC establishment, and for many people the primary source of computing isn’t a computer at all. The new computer companies sell phones, and TVs, and books, and cars, and thermostats – connected to clouds with more information and people with even newer ideas.
What is now a trickle of “things” looks to become torrential, if an entire industry can restructure and respond to the next change: how to stay in business building chips. The shape and size of a computer has changed radically, and the system-on-chip and microcontroller are now the focus, with supporting technology like displays, touch sensors, batteries, sensors, and wireless connectivity.
Everyone – the fab heavy, the fab light, and the fabless – is challenged here. Intel, with the delicate problem of filling billion dollar facilities running mostly cutting edge processes, has to be very careful not to replace one $230 Core i7 with one $23 Quark, or even two or three – it has to be more like 10 or 20. At the other end, fabless firms can create a huge range of ARM-based chips for the IoT, but often run into one of two risks: low average selling prices and margins and the strategy of “we’ll make it up in volume” which may never show up, or complex, higher performance devices with fewer design wins, a long time-to-money, and a more difficult customer support model.
I disagree with one conclusion in the “computerless” article: Amazon, Apple, Google, Qualcomm, and Samsung went vertical, building their own chips, their own software, their own devices, and their own content stores and retail channels. This wheel creates volume consumer demand, reduces the risk of development, and makes for a more manageable supply chain. To Rappaport’s and Halevi’s point, vertical does look a lot different in the disintermediated, social-powered environment of today.
Intel, on the other hand, is now holding a lot of risk in their bag in this new reality. They have chip capacity – with Fab 42 waiting to come online when it becomes clear what to do – and software ideas, but so far no winner mobile or IoT chips. For Quark to succeed on the IoT, they desperately need to spur device demand, with the latest idea the diminutive Edison module targeting wearables and the maker community at-large.
Other established semiconductor firms, while emitting positive signals about future prospects, are secretly hoping the IoT doesn’t run away and blow up their comfortable embedded models with longer lifecycles and predictable growth. The reality is if established firms like Atmel, Freescale, Microchip, Renesas, and TI were all asked to grow shipments 10x in five years, it would be a huge stretch.
Yet, we seem to be assuming these IoT chips are going to come from somewhere in enormous volumes. Let’s assume for a minute hockey-stick demand does actually show up – how does the industry respond?
The “computerless” end-run devised by ARM targeted the fabless startup and the vertically integrated, and those firms enjoyed wonderful growth but are now discovering the joys of capacity, lead time, yields, margin, allocation, and other issues which will slow them to a more normal pace. A company like Qualcomm, using TSMC fabs and technology like their AllSeen software, is probably best positioned to become a leading IoT supplier based on their mobile success.
A manufacturing pivot from Intel rallying around a more microcontroller-ish Quark has some hope of succeeding, if IoT demand can be crystallized on their ecosystem. Intel has the resources to make capital investments and a sales/marketing push on a scale none of the existing microcontroller companies can match, and a leg up with a $5B fab already on the ground ready for advanced tooling on relatively short notice when demand picks up – if they have the right part.
Some firms are getting braver in freeing themselves from old school thinking to get into this game. Xilinx has shown tremendous creativity in Zynq, merging ARM cores with programmable logic for many applications. The MIPS core has found new life with Imagination Technologies, while Cadence (Tensilica) and Synopsys (ARC) are finding a lot of homes with non-general-purpose cores.
But the new “computerless” companies are finding hardware is not an easy business. I doubt a firm like Google, showing disdain to stay the commodity hardware course in jettisoning Motorola, would be willing to fill the semiconductor gap. Microsoft is also very likely to disembark from some or all of its hardware soon, with a new CEO in place, and that may knock an AMD recovery off course. Cisco is dabbling in chips, but is also an unlikely candidate to fill the massive IoT chip gap.
The prediction of “computerless” certainly was on target, but “chipless” won’t work at all for the IoT. Can you hear the train coming? Someone will have to take on new semiconductor technology and a new model of economic viability on a bigger scale than we’ve ever seen, probably leaving behind a lot of older stuff in the process.