As the helmsman of the Ever Given,
currently recently stuck sideways in the Suez Canal, can attest, turning a massive ship is no easy task. And there’s no bigger vessel in our industry than the S.S. $INTC, otherwise known as Intel Corporation.
Intel is probably the only semiconductor company in the world that the average person can name. That’s thanks to years of “Intel Inside” advertisements and the company’s dominance of personal computers over the past few decades. Emphasis on past.
It’s no secret that Intel has stumbled lately, and the company’s unquestioned dominance of all things computing is under threat. It’s possible that its best days are already behind it. But is Intel really taking on water, or does it just need a minor course correction? Is the Good Ship Intel sinking or just listing a bit to starboard?
Taking over command is one Patrick P. Gelsinger, a former Intel engineer returning to his previous employer after a 12-year absence to take on the biggest job of all. Gelsinger helped design the ’386 and ’486 microprocessors, the chips that took Intel into the 32-bit world by dragging the old 8086 kicking and screaming into the 1980s. The ’386 was a huge technical feat and a complete success. That alone would have been enough to gold plate anyone’s resume. Now he’ll try to do the same for the entire company.
Last week Gelsinger gave a speech/pep talk that outlined his strategy for righting the ship and steaming ahead. Naturally, it was optimistic. That’s expected of any new CEO. But amidst all the optimism was some real news. Gelsinger doesn’t just expect to do the same old thing as his predecessors and somehow get different results. He’s promising big changes – changes that might actually work. But, even if they don’t, it will mean a different Intel for all of us. It’ll be either a spectacular success or a splashy disaster.
First of all, Intel isn’t in terrible shape, despite what many short sellers and doomsayers may portend. The company has a $250 billion market capitalization (in case you’re thinking of buying it) and its stock is trading at a 5-year high. It made about $6 billion in profit just in the last quarter. This is not a company circling the drain.
Having said that, Intel has certainly lost its edge in manufacturing technology and new design wins. So, the future isn’t looking as bright as it usually does at Intel. The company that used to lead the world in fabrication technology is now one or two generations behind mere third-party foundries like TSMC. Its most obvious competitor, AMD, is gaining market share in part because the AMD/TSMC pairing produces better silicon than Intel can with its in-house technology. That’s a completely nonintuitive reversal of fortune.
Intel’s historic dominance has been based on two things: the demand for its x86 microprocessor architecture, combined with the company’s outstanding manufacturing technology. The two complemented each other. Microprocessors based on x86 (as opposed to, say, ARM, MIPS, or RISC-V) are insanely complex, which would normally make them slow and power-hungry. But Intel compensated with exceptional silicon fabrication. It manufactured its way around the problem. Even though Pentium, Core, and all of their siblings were creaky, byzantine designs, they were still faster than more modern, cleaner competitors. RISC challengers tried for decades to unseat Intel and failed every time.
But now both halves of that two-pronged strategy are failing. Demand for x86 chips ain’t what it used to be, and not just because Apple switched to M1. Other architectures are making inroads into servers, PCs, laptops, and especially mobile devices. And Intel made some bad decisions that, in hindsight, seriously hurt its manufacturing technology. Intel now lags, not leads, in semiconductor process nodes. An über-complex chip built on a second-rate fab line isn’t a good combination. Enter Pat Gelsinger.
His strategy is also dual-pronged. First, to open up x86 to licensing as IP. (Heresy!) And second, to rent out Intel’s production facilities like an independent foundry. (Horror!) Other people will be able to make Intel microprocessors, and Intel will make other people’s microprocessors. He’s unbundling Intel’s architecture from Intel’s manufacturing.
There’s a third part, too. Intel will increase its use of outside third-party foundries for its own top-line CPUs. In other words, Intel chips won’t necessarily be made by Intel. While that seems like a big deal, Intel has been using third-party foundries for ages. It’s already one of TSMC’s biggest customers, behind Apple but amongst companies like MediaTek, Qualcomm, nVidia, and AMD. Weirdly, Intel and AMD processors might soon be built side by side on the same production line.
Licensing the x86
Intel has never licensed its x86 CPU designs (well, not for a really long time), so the idea of licensing a real, genuine, official Intel CPU core is kind of exciting. Customers will be able to take something like a Core i5 CPU and combine it with their own IP to create a custom SoC, much as ARM or RISC-V licensees do today. Welcome to the 1990s, Intel. Enjoy your stay.
Licensees could also remove or tweak some features, like changing cache sizes or eliminating peripherals from the standard Intel parts. You could create your own version of the Core i7-9750, for example, but with different features to suit your application.
There are bound to be limitations, though. Gelsinger wasn’t clear about which CPU designs would be available for licensing, or how much leeway licensees would have, or where such parts could be fabricated. It seems likely that any CPU you license from Intel would need to be built on Intel’s existing production lines. Intel’s CPUs have always been exquisitely fine-tuned for Intel’s own process peculiarities, so sending them out to TSMC (for example) is a nonstarter. It’s also unlikely that Intel would spend the time and effort required to port its CPUs to other foundries’ nodes. CPU licensing will probably be tied to Intel’s foundry services. “If you want our CPU, you gotta build it here.”
Gelsinger didn’t say that Intel would license the CPU architecture, just the CPU core designs. That’s pretty much the way most CPU IP companies work, including ARM. Nearly all ARM licensees get a specific CPU core design, with all the features and functions locked in. No tweaking allowed. A relative handful of ARM licensees (like Apple and Qualcomm) have a much broader and more expensive architectural license that allows them to design their own microarchitectural circuitry around the standard instruction set. No word on whether Intel would allow this.
It’s long been possible to clone the x86 architecture with no outside help from Intel. It’s not easy, but it’s doable. AMD has built an entire business around this, and many smaller chipmakers have done so since the 1990s. Each successive generation of Intel processor gets harder to clone – that’s deliberate – but it’s still possible to make a “clean room” x86 chip. Licensing from Intel would be much less work than creating a compatible CPU from scratch. The bigger question is, why would anybody want to?
Given the choice between spending, say, $1 million for a license to ARM or a license to x86, which would you choose? Depends on what you’re making, of course, but I’d wager that most would pick ARM over x86. Or choose RISC-V and save the $1 million fee. Opening up its x86 CPU cores to licensing is a big deal philosophically, but I’m not sure it’ll make much of a difference to anyone’s bottom line.
Intel Foundry Services
Then there’s the foundry half of the strategy. Intel will agree to manufacture chips on its fab lines for third parties, something that it’s done before only in a very limited way. This new foundry service will operate as a separate P&L, signaling that Gelsinger is serious. Back when Intel’s fab technology was second to none, this would have been a big deal. Now, it’s somewhat less remarkable, and less attractive.
That said, Intel’s current state of the art isn’t exactly terrible. It’s just not quite as advanced as it used to be. Not everyone needs to use the latest process node. There are plenty of chips being manufactured in slightly older-generation processes, so Intel has a big potential customer base, which Gelsinger characterized as a $100 billion addressable market.
But again, Intel’s internal processes have always been proprietary and therefore somewhat outside of the mainstream when it comes to EDA tools or cell libraries. You can’t just shift your production to Intel from another foundry. That’ll take some reengineering. How, exactly, is using Intel as a foundry service going to be any better than using TSMC, UMC, GlobalFoundries, SMIC, or any of the other foundries that have specialized in that business for decades?
One differentiator is that many – but by no means all – of Intel’s fabs are on American soil. That’s important to a certain class of customer. But it’s also not unique. TSMC and Samsung, to name just two examples, are expanding into the United States with two new fabs in Arizona, figuratively just down the road from the two new fabs Gelsinger says he wants to build in the state.
Becoming an independent foundry with a separate balance sheet would put Intel in competition with the very foundries it says it will increasingly rely upon for its own products. How successfully will Intel be able to negotiate with TSMC to build its CPUs while simultaneously trying to steal foundry customers from them?
Gelsinger even theorized about someday making chips for Apple. Ironic, since Intel is currently running an anti-Apple advertisement. That ad was presumably already in the works before Gelsinger took charge.
Newton’s First Law
Objects in motion tend to stay in motion, and deflecting their course requires energy proportional to the object’s mass and its velocity. Intel is a massive corporation, as these things go, and it will require a lot of energy to change its course. The company didn’t get this big by mistake; it grew through years and years of success. Changing all that will not be the work of a moment.
With massive size comes massive inertia. There is the “Intel Way,” and it’s proven remarkably successful. The company has grown so predictably and so consistently that people have spent their entire careers inside Intel without ever having a bad quarter or seeing a significant layoff. That constant reinforcement makes people complacent, maybe even cocky or arrogant. Or, to be more charitable, it engrains certain habits and assumptions that will be hard to dislodge.
Gelsinger’s new strategy will make some longtime Intel employees very uncomfortable. That doesn’t mean they can’t be persuaded or cajoled, or even fired. But it does mean the transition won’t be easy, swift, or free of drama. Imagine that you’ve spent your whole career protecting the precious intellectual property embodied in every x86 microprocessor. Or that you’ve enjoyed working only with other Intel engineers in shepherding a design through the fabrication process. Now that all goes out the window because the new CEO – the third in as many years – says he wants to make a change. How many senior employees will drag their feet? How long before the crew is really on board and rowing in the same direction?
I like Pat Gelsinger. He’s a brilliant hardware engineer, a successful manager at VMWare and EMC, and a nice guy. But he’s got a Herculean task ahead, and he can’t do it alone. Not all engineers make good CEOs, but neither do they all make bad ones. Anyone who can make an old x86 processor do the things he can has got some serious talent. Such a massive change of course is going to churn up turbulence, but I wouldn’t bet against him.