This morning, Intel gave an indication of how it plans to continue development in 5G technology after it shut down earlier internal efforts and allowed Apple to absorb their remains. In today’s announcement, Intel said it is partnering with MediaTek in “the development, certification and support of 5G modem solutions for the next generation of PC experiences.” Apple, which switched earlier this year to Qualcomm’s 5G modems, had been Intel’s most important customer. MediaTek is one of Qualcomm’s principal rivals in modem technology.
It’s been something like five years since I had a close look at Intel. Somewhere early in the regime of former Chief Executive Officer (CEO) Brian Krzanich, my lines went dead. Other industry analysts reported around then that they weren’t on the good list anymore, either. Then, recently, the new analyst relations team called me up and said it wanted to reengage. The simple explanation is that Krzanich didn’t believe in analysts, and his successor, Bob Swan, does. Thus, last week, the company hosted a two-day event for a group of us at its headquarters campus in Santa Clara, CA.
In his initial remarks at the conference, Swan explained that future growth for Intel would come from 5G, the Internet of Things (IoT), and something he called “mobility as a service” (shorthand: robotaxi). I had been wondering about the 5G part, and even the IoT part (because so many of the trillion or so connected perimeter devices expected to ship in the next half decade will need 5G connections). Today’s announcement gives at least a partial explanation of how Intel expects to play in that market.
Swan is a financial guy, having been Chief Financial Officer for various companies before taking that role at Intel. As such, he represents a departure from Intel’s usual bench. With the exception of Paul Otellini, who was a marketing guy (and produced mixed results), chieftains before Swan have all been semiconductor engineers.
But his skills seem to be timely, and he wields them carefully. According to one Intel executive I spoke with during an informal reception, Swan understands he must take financial risks and knows some of his projects will fail, but the risk is calculated, and he’s not letting fear of failure prevent his making the necessary bets. And understand, here, the bets we’re talking about run into the low tens of billions of dollars, the cost these days of an updated semiconductor plant.
Intel is the most important of the few remaining integrated device manufacturers (IDMs). An IDM designs, makes, and sells its own semiconductors. Samsung and Huawei make and use some of their own chips, but Intel is the only volume merchant producer. Most other chip companies, like Qualcomm, NVidia, and AMD, design semiconductors but have them made by a foundry like TSMC, Samsung, or GlobalFoundries.
Now, it might be coincidence, but the pause in the conversation Intel had been having with the analyst community did occur at roughly the same time the company was going through some rather uncomfortable experiences, notably, its long struggle with the 10nm node, a milestone that the foundries blew through handily. In the past, Intel had enjoyed as much as a two-year lead on the latest node in the ever-shrinking-semiconductor race. There were some arguments about how gate width is actually measured, and whether someone was playing fast and loose with the designations of the nodes leading down to single-digit nanometers. But in the current moment, when Intel has chosen to reengage with analysts, the company is arguably behind.
For the old Intel, this news would be devastating, as the firm pegged its reputation on being process node leader to the degree that it would bury skepticism about architectural ungainliness with an assertion that any inefficiencies would be solved merely by the existence of the next node, which would always be smaller, faster, higher capacity, cheaper, and cooler. What’s not to love?
But the new Intel is all about facing reality and pulling up its socks. Swan called the delayed 10nm transition “challenging” and “painful,” but promised a “predictable product schedule” in the future as the company gets “back on a two-year cadence going to 7nm and 5nm.”
Owning past missteps, he went on to explain a fundamental change in how the company views itself in the future. Over the years, various antitrust bodies have examined Intel through the lens of its holding a 90% share of the PC market, making the company appear monolithic and incumbent. But Swan sees the company as addressing a larger potential market, in which it has a much smaller share and faces strong competitors.
Beyond selling x86 processors into the PC and server markets, Intel expects also to jockey for markets in various types of memory and storage technologies, programmable logic chips, customized chips, wired and wireless communications technologies, silicon photonics, visual processing, and, yes, even graphics, all surrounded by enabling software, tools, and templates.
Beyond servers, storage, and PCs, the company now plans to become a force in networking, connectivity, the IoT, and related software and services. Intel’s commitment to the IoT is rather important. In terms of big markets, Intel is one for two, having won the microcomputer market and lost smartphones. The next big opportunity is the IoT, which isn’t one thing, however, like PCs or phones, but many things. And there’s good and bad news in that diversity. The good news is there will be lots of opportunities. The bad news is that there will be lots of competitors.
With its MediaTek partnership, Intel is better positioned to address a large portion of the IoT opportunity and related artificial intelligence market. It will eschew the little devices at the perimeter. Those are better suited to ARM’s low-power computing technology. But it will do battle at the edge node or gateway, a small piece of the cloud that sits near where data is gathered (and often used). ARM’s licensees, including Qualcomm, also plan to bid for the edge node market, but Intel has an inside track on cloud servers, particularly those with wired connections.
With respect to the mobility-as-a-service or robotaxi opportunity, Intel is focused on the largest, most coherent, and nearest-term IoT market: self-driving cars. To this application, Intel brings its most successful acquisition to date, Mobileye, an Israeli company that uses multiple vehicle-mounted cameras to map terrain and guide a moving vehicle. With visual data alone, Mobileye has achieved results as good or better than systems supplemented with radar or lidar. These results bode well for fully automated driving, as completely redundant systems make for better safety.
Toward the end of the analyst event, I had an opportunity to sit down with Tom Lantsch (with a last name that has one of the lowest vowel-to-consonant ratios on the planet). Lantsch, the senior vice president and general manager of the IoT Group at Intel, is in a singularly good position to weigh in on the coming battle for the edge gateway: he used to work at ARM. He left not long after SoftBank bought ARM in 2016. He likes the cards he’s holding now.
Lantsch made it clear that Intel won’t be going after most of the tiny devices on the perimeter, but sees a world in which the x86 and ARM camps coexist. “We’re collaborating with those perimeter devices efficiently,” he said. “Very few ARM licensees will try to compete in my space” he predicted, noting that VMware, the king of virtualization in the server world, “doesn’t really support ARM,” and that no commercial grade of Red Hat, key open source enterprise software, exists on ARM. He also pointed to the Movidius acquisition as one Intel answer to the perimeter device market. Movidius makes low-power visual processors the latest of which can perform AI operations right at a remote camera site.
The Intel of today is humbler, chastened by the rough patch it just went through. But the company retains plenty of its old swagger and displays a hunger to compete in the large markets of tomorrow. Certainly, it will garner much of the server capacity required to handle the data generated by the expected trillion perimeter devices, but it is also likely to make inroads in specialized applications at the perimeter that involve visual processing and AI, particularly in vehicles. Expect a lot of competition and dynamism at the network edge.