ntel CEO on PC market: green, shoots and leaves
May 13, 2009 10:10 PM CT



SANTA CLARA -- On Tuesday, Intel CEO Paul Otellini took the stage for the opening keynote of Intel's investor conference, and told the assembled audience the following about the chipmaker's order and billings pattern: "what we've seen so far is a little better than we expected." As he began speaking those words, laptop keyboards across the auditorium immediately erupted in a chorus of clacking, which continued as he went on: "I'm more firm in my belief than even six weeks ago that we will see seasonality in the second half."

Half of those typing were reporters, who were eagerly filing "green shoots" stories about a possible turnaround in the economy; the other half were investors who were entering "buy" orders for INTC, which had jumped 4 percent in price by the time the afternoon was over.

Otellini's keynote was a fitting opening for the entire conference, in which Intel hopes to give some sort of closure to the past three years of pain—internal pain from restructuring and share price declines, and external pain from the global downturn. Describing the former, Otellini recapped how Intel's payroll has shrunk by 25 percent since 2006, due to a wave of factory closures, layoffs, and consolidation that has so far put almost $4 billion extra in Intel's bank account. He then moved on relatively quickly from talking about how the company has shrunk to talking about how it will grow.

The primary growth driver that Otellini sees for Intel is, unsurprisingly, the Internet. Specifically, Otellini pointed to the continued global growth in IP traffic and quipped, "there's no recession on the growth of the Internet." And he's right, of course—unless you're talking about Sweden, global IP traffic is the one curve that hasn't suddenly reversed direction in the past year.

Intel plans to ride this curve by selling the silicon that goes at each endpoint of an IP connection—any IP connection, not just connections between traditional "PCs." This means that Intel will have to continue to grow up, into high-performance computing and mission critical servers, and down, into the "embedded" market.
Shifting segments

Otellini declared the standard desktop PC "about dead," but he then went on to describe a segment that's less dead than it is disaggregated into multiple parts: "big, boring, and beige" (Otellini's description) has given way to "targeted microsegments": enthusiast PCs like Alienware, the corporate desktop, lifestyle PCs like the iMac and HP Touch, and nettops.

When touting Intel's stepped-up efforts in the high-end of the computing market—mainframes, servers, and HPC—Otellini took a moment to poke a finger in Sun's eye. "As the fate of SPARC is up in the air, it gives us an opportunity to push Itanium and Xeon into that installed base." The Intel CEO said that Intel hopes to see HPC accounting for 25 percent of its Xeon business by 2012.

Moving down-market, Intel sees netbooks, handhelds, "embedded," and consumer electronics each as $10 billion businesses in 2011 timeframe.

You'll notice that I put quotes around "embedded," because, as I've pointed out before, Intel's definition of embedded is markedly different from that of most of the rest of the industry. And even then, the company isn't very consistent in its use of the term.

At any rate, Otellini made it quite clear that Intel would use Atom and its successors—which currently numbers some 8 SoCs under development—to push directly into territory occupied by ARM, MIPSs, and PowerPC. He didn't give much of a rationale for Intel in this space, though, beyond the typical stuff about Intel's installed software base and fab muscle.
The shift to software

"Software is becoming increasingly important to us, and it's an increasing part of our spend," Otellini told the audience. As a professional Intel-watcher, I can vouch that this is definitely true.

As a result of trying to scale processor performance with increases in core count instead of clockspeed, Intel has had to become more and more of a software company. Nowhere is this more apparent than in the company's forthcoming discrete GPU, codenamed Larrabee. Indeed, I've previously described Larrabee as a software GPU that runs on an accelerator add-in board. In this regard, the Larrabee launch will be the first high-profile test of just how well this shift into software is working out for the company.

It's worth noting that this software-centric trend puts the chipmaker—with its massive and growing investments in Linux—increasingly at odds with Microsoft, especially in mobile. But Intel can't worry about that right now; the chipmaker has to keep plowing money into x86 Linux projects to keep Linux projects on ARM and other architectures from catching up. This dynamic is great for Linux, but not so great for Microsoft.
TSMC: it's about third-party IP blocks

When I covered the announcement of the Intel-TSMC fab deal, my take on it was that it was fundamentally about Intel farming out the less profitable, higher-risk part of its chip business so that it could focus its own fabs on producing higher-margin products. That is still my take, but a number of readers vehemently disagreed, insisting that Intel did the TSMC deal in order to make it easier for customers that want customized, Atom-based SoCs to integrate their own IP blocks with Intel's Atom cores.

Otellini sided with our readers, reiterating that "this is not an outsourcing move for Intel. We're not doing this for capacity, and we're not doing it to help Intel internal production." He went on to explain that the TSMC deal was done so that Intel could attack new markets that have players with large volumes and who want to bring their their own IP to an SoC. Such companies are often already TSMC customers, so that TSMC already has a library of such blocks. It would be too expensive for these customers to port those blocks from TSMC to Intel, so Intel is just porting to TSMC so that these folks can do their custom products easily and cheaply.

So that's the official line, and it makes a lot of sense. It's also not incompatible with my own take, but perhaps I'm overthinking things. We'll know that I'm probably out to lunch if 2010 sees a slew of announcements of Atom-based, bespoke SoCs from various SoC customers (like Apple?). Conversely, if 2011 rolls around and TSMC is turning out Intel-designed SoCs with the occasional one-off from another customer, then that would suggest I was right.


Source: http://arstechnica.com/hardware/news...and-leaves.ars