Why Intel wants to be everyone’s chip maker
This is the first in a series of columns on the strategic direction of several major technology companies that have faltered of late. We’ll start here with Intel, follow in a couple days with Cisco, followed by Microsoft, then see where it goes from there.
At Intel’s annual shareholders’ meeting last week the company talked about moving strongly into mobile chips and selling its stillborn OnCue over-the-top video streaming service, but the most important story had to do with expanding Intel’s manufacturing capacity. This latter news is especially important because if you look at the square footage of 14 nanometer fab facilities Intel says it will be bringing online in the next two to three years it appears that the company will shortly have more production capacity than all the rest of the semiconductor industry combined.
Not just more 14 nm production capacity, we’re talking about more total production capacity than all Intel competitors together.
This is fascinating news for several reasons. First, at $5+ billion per fab you don’t add 3-4 new ones without a darned good reason for doing so. Second, it takes so long to plan and build these plants that this part of Intel’s strategic plan had to have been in the works for years before the company ever mentioned it in public. So it’s not like new Intel CEO Brian Krzanich moved into his office and said, "Let’s build some new fabs".
That’s not to say Krzanich has been without impact on the company. He clearly (and properly) killed OnCue, which was Intel’s third try at building a media empire and wasn’t any more thoughtfully done than the first two failures. Watch for Intel to sell OnCue to Verizon for $500 million as planned, but with sweetheart financing from Intel Capital so it doesn’t cost Verizon a penny upfront.
This is not a criticism, Brian: I’d do the same thing.
Increasing the fab capacity makes sense, too, though maybe not as much sense as it made a few years ago when the idea was first presented. Then PC sales were still growing and excess production capacity promised the sort of low prices that could finally kill the hated AMD.
Only today AMD isn’t that much of a threat. There are other players like Qualcomm in the mobile space and even Apple that are far scarier than AMD.
So what’s a Krzanich to do? He’s making the best of a difficult situation.
Understand that Intel is far and away the best semiconductor manufacturer the world has ever known. Its only real competitor in manufacturing is TSMC in Taiwan and even that’s not a close race. Intel has the best technology, the best yields, and because of the way things work in the semiconductor industry, it has the lowest manufacturing cost per chip.
What it doesn’t have, however, is the best mix of chips to build. Desktops are in decline, the market is all GPUs and mobile with a huge flash RAM opportunity on the horizon -- all areas where Intel is at a disadvantage.
So Krzanich has Intel looking into building chips for others, entering the foundry business. On the face of it this move looks really, really stupid. But the more I think about it the more sense it actually makes.
What’s stupid about competing with TSMC and others in the foundry business is the profit margins aren’t good at all. Typically a larger fabless semiconductor company will pay around $6,000 for each foundry-processed wafer, yet the same wafer filled with Ivy Bridge processors can easily generate $400,000+ in sales for Intel. That’s the whole idea behind vertical integration -- to take all the profit from every stage of the business.
Admittedly there’s plenty of R&D and other expenses that need to be covered by that $400,000 wafer, but the idea of making it all yourself to reap all the benefit makes sense… except if you don’t know what to make.
That’s Intel’s problem. It can build the darned things better than anyone else but it doesn’t necessarily have the right product mix to build for the current market. And even if it gets its product mojo back tonight that won’t have much effect on its business for another two to three years.
Brian Krzanich can’t wait two to three years for new mojo. He needs mojo right now.
And that’s why Intel is suddenly interested in the so-called foundry business. It is soon to have excess 22 nm and then 14 nm production capacity and will be able to easily undercut any other manufacturer at those feature sizes, all the while offering superior performance. Traditionally this would lead to a market bloodbath with most Intel foundry competitors dying and production costs eventually going back up as companies fail and capacity is cut back. But I don’t think that is what’s happening in this case.
If Intel drove the per-wafer price up to $12,000 or even $20,000 it wouldn’t make enough difference to Intel’s bottom line to be worth the anti-trust risk. They’ll still be cheaper, just a little bit cheaper.
I believe Intel is entering the foundry business mainly as an industrial intelligence operation.
As a chip company exclusively manufacturing its own designs Intel competes with most of those fabless semiconductor companies, but as a foundry -- especially the cheapest best foundry around -- those same companies will open their kimonos to Intel (with strict NDAs in place first, of course).
But this time the NDA doesn’t matter in this case because Intel’s purpose isn’t to steal trade secrets -- it is to find companies to buy. Once you buy the company the NDA dissolves.
Intel needs new product lines sooner than Santa Clara can design them itself. More importantly, having rightly lost some confidence in its ability to predict and lead the market, Intel needs a few astounding ideas from outside and this is by far the easiest way to find those.
Suddenly Intel manufacturing engineers will have a view of the chip market they never had before. Find the best products that are close to market, open the checkbook and buy them up. What better way to find those new products that really work than by manufacturing them in the first place?
If I am right, Intel is emulating Cisco’s 1990s strategy of buying ahead of the next technology wave, though in this case leveraging its superior fab technology to figure out that next wave.
It might even work.
Reprinted with permission