Monday, July 21, 2014

Pyrrhic victory: Intel forces its way into mobile, but might destroy itself in the process

Greedy... That is all...

Earlier this week, we covered Intel’s determined push into mobile phones and its willingness to spend billions of dollars in a loss-making push designed to ship Bay Trail to as many customers as possible at price points that compete directly with ARM-based Android tablets. Left undiscussed was the question of whether the tablet market will support a business model Intel finds attractive. It’s a question that’s increasingly important given the rapid commoditization of tablets and smartphones.

Consider how much things have changed since Intel first showed Medfield. Two-and-a-half years ago, ARM’s big.LITTLE had only just been announced. Tegra 2′s dual-core Cortex-A9 was the fastest chip on the market. People were still talking about Windows RT as a major priority for Microsoft if the company wanted to be relevant in the tablet and smartphone space long term, and Android’s support for major features like multithreading was still rather crude. There were no 64-bit ARM cores, whereas Intel already had a full 64-bit implementation in hardware. Even more importantly, it had a great deal of experience in optimizing software for x86 processors and providing the tools developers would need to make Android applications sing on Intel platforms.
Intel’s Medfield smartphone — that went absolutely nowhere

Fast forward to the present day, and the market Intel is fighting to enter has changed more rapidly than Chipzilla anticipated. Intel chose to stick with dual-core with Hyper-Threading while the ARM SoCs jumped to quad-core. It chose to keep its modem manufacturing at TSMC, where it has faced repeated, unspecified delays. For all that Intel leads the world in semiconductor manufacturing, its XMM 7160 (that’s the company’s 4G modem actually shipping in any consumer hardware) is built on 40nm, while Qualcomm’s Gobi 9×35 platform is sitting on 20nm. Intel’s XMM 7260 is delayed again with wide availability not expected until 2015 — and uses a 28nm radio and 65nm baseband compared to Qualcomm’s 20nm for both components.

Things don’t look much better on the software side. Between 2011 and 2014 Android added ART, quad-cores and better multithreading went mainstream, and big.LITTLE evolved from a clunky power-saving measure into a heterogeneous compute model. Intel, according to our sources, has struggled with GPU driver support — it’s no accident that the Atom-based tablets that are finally shipping in volume are mostly using Moorefield, which has a PowerVR GPU, rather than the Intel HD Graphics that launched with Bay Trail.

Missing the market

These headwinds would matter less if the tablet and smartphone markets were still growing at their previous rates, but sales have dropped off sharply even as prices trend downwards. Samsung is facing stiff competition from mainland China manufacturers, while Qualcomm faces upstarts like Allwinner, Mediatek, and Rockchip. Intel’s own Rockchip deal will help it establish some traction in these markets, but again, it’s leaning on TSMC’s manufacturing to make that happen.

Let’s assume, for a moment, that Intel’s SoFIA starts to take off late this year and helps its margins out. Meanwhile Broxton and Airmont (14nm Atom) arrive in 2015 and bring costs down further. The XMM 7260 finally gets qualified and starts shipping in real volume and the Rockchip deal strikes gold. By the beginning of 2016, Intel has its entire house in order and is ready to compete up and down the stack.

The problem is, tablet ASPs (average selling prices) are expected to have dropped even further by the time this happens, as are smartphones. Intel will will be slugging it out with companies that have already established themselves at the top of the Android stack for much longer than it has. Growth rates, meanwhile, are slowing across the entire market.

Even Apple products aren’t immune to this trend. Lower ASPs mean less opportunity to squeeze premium pricing.

This chart is older, but it illustrates the point. The SKUs Intel is fighting for are the SKUs at the far left.

It’s simply not clear if Intel can pull its 1980s come-from-behind trick in the future mobile market. In the early 1990s, with the entire PC market still emerging, Intel was able to establish itself as a premium brand and defeat its challengers long before the value fell out of the PC space — which is another way of saying that it was able to hold on to the profit margins Dell, HP, and Lenovo couldn’t grasp. In 1996, while Intel dominated the x86 space, it still faced competition from a handful of consumer manufacturers (IBM, IDT, Cyrix, AMD) and the RISC server market still had some life in it yet.

Intel locked up the margins in the PC business when PCs still cost 4-5x what they do today. It’s already missed its chance to do the same thing in tablets

By 2001, most of Intel’s CPU competition had been swept aside. Obviously Sun and IBM didn’t go out of business, but neither regained its market share. Intel established itself as the premium CPU provider for everything from desktops to servers, and it held that position against AMD thanks to a variety of tactics. The point is, the battle for desktops, laptops, and servers was essentially over by 2003 thanks to Intel’s creative methods of limiting AMD’s sales potential — even if we didn’t know it at the time.

Contrast this with the situation Intel faces in 2016, even if we assume an ideal product mix and strategy. We’ll assume that the company’s Rockchip partnership pushes its x86 products into the mainland China market and that these parts are profitable. That’s important — but by 2016, Intel’s 14nm Airmont follow-up will be facing competition from Qualcomm’s first custom 64-bit ARM chip plus whatever Apple has cooked up for its own efforts in this space. If Nvidia is still doing its own designs, it’ll be fighting for competition there as well — at a time when all three of those companies will be doing their best to stave off the commoditization of their own products.

Intel faces increasing headwinds

Does this mean Intel has lost its chance to be a major player in the mobile market of the future? Absolutely not. Intel still has the most advanced fabs with the top overall staff, the best long-term track record, and some of the best chip architects in the world. This wouldn’t be the first time Chipzilla engineered a potent come-from-behind strategy and wound up dominating a market it entered as a bit player.

Given the company’s performance of the last few years, however, I think a measured reevaluation of its likely role in the larger mobile ecosystem is in order. In 2011, Intel had some clear expertise that ARM SoC developers hadn’t yet matched. Today, there are still benefits that Intel could bring to the table, but Android and ARM have done a fine job of advancing under their own steam. Bay Trail and Moorefield have demonstrated that the company can compete with a section of the ARM ecosystem , but not that Intel can outperform it.

Whether or not Intel needs to clearly outperform the other SoC vendors (and command the associated premium) depends entirely on its cost structure going forward. There are a great many questions about how Android and Windows margins will intersect and whether we’ll ever see ARM devices powering Windows tablets to any great degree. Navigating these waters will be tricky for Intel, particularly in concert with the kind of design overhaul that will allow it to sell its tablet and smartphone SoCs without resorting to contra-revenue tactics — i.e. losing money on every single device sold.

Source: ExtremeTech

Your VBKid

0 Codes:

Post a Comment