I was already working on a column about AMD purchasing multicore server maker SeaMicro, pointing out what a coup the deal is for AMD, when the story appeared yesterday about an Intel executive claiming the chip giant had been offered SeaMicro and chose to pass on the deal, followed by a SeaMicro board member claiming the Intel exec’s statement was a bald lie. Who is telling the truth here? Who is lying? And does it matter? It is my opinion the answers are that both are telling the truth, nobody is lying, and none of it matters very much. Here’s why…
Remember Bill Clinton saying in a deposition that the truthfulness of a particular statement depended on your definition of the word “is?” I think that’s what is happening here, too. I doubt that SeaMicro sold itself without Intel knowing in advance. At the very least SeaMicro would have tried to get a second bidder to drive higher the final purchase price. Whether Intel formally rejected the chance to buy SeaMicro, they probably did so at least tacitly by, for example, choosing not to counter the AMD offer. Maybe there weren’t papers on the table, either, but I’m sure a question was asked and answered to the satisfaction of both parties, maybe over lunch or in a phone call. And SeaMicro wasn’t insignificant to Intel, which saw the startup as an important enough customer to build a custom 64-bit Atom processor just for SeaMicro.
To better understand the deal let’s look at SeaMicro’s product line, which presently consists of servers using Intel Atom and Xeon processors with up to 256 cores. These are remarkably low power servers ideal in many respects for the data-center-ic (I just invented that word if you were wondering) world of IT in this decade. Intel makes the processors (or has so far) while SeaMicro integrates them into servers using custom chipsets that define the “fabric” of their distributed architecture.
In the simplest sense, then, AMD bought SeaMicro to eventually dump the Intel processors and replace them with AMD processors, which shouldn’t be difficult to do. That’s certainly the way Intel is viewing this deal. But there is much more to it than that.
AMD did not buy SeaMicro to go into the server business or, frankly, to even replace Intel processors, though the latter is likely to happen at some point. AMD bought SeaMicro for the fabric, for those proprietary chipsets that are the true heart of SeaMicro servers — chipsets that will go shortly to 512 cores, then 1024 and beyond.
AMD doesn’t want to sell servers to you and me, they want to sell chips to HP and Dell to put in their SeaMicro-type servers sold to you and me. Expect the SeaMicro server business to disappear before long.
Now why didn’t Intel go for the deal? It’s not like they couldn’t afford the $336 million price. It comes down to several factors, the first of which is NIH — Not Invented Here. SeaMicro has been working for years with Intel and Intel had to have learned plenty from that partnership. If Intel turned down the chance to buy SeaMicro is has to mean there’s a line of similar fabric chips on their way soon from Intel — chips on which I’ll guarantee you Intel has already spent more than $336 million.
But Intel is a proud company — proud and to a certain extent deluded. They see an all-Intel fabric solution as being inherently superior and therefore more valuable even if it isn’t quite ready to hit the market. Intel decided to build rather than buy. But in this instance Intel probably made the wrong choice, as I think they are beginning to see.
It’s this brouhaha about who offered what to whom that raises a red flag for me. Intel would normally not have commented at all on AMD buying SeaMicro. The fact that someone at Intel did comment shows both discord in the ranks (Intel is off its message, which is bad for brand value) and might even be panicked.
There’s a continual debate at Intel about the residual value of the Intel Architecture (IA). Intel management sees that value as significant, some people see it as zero (the cloud and mobile have made processors effectively interchangeable) while I actually see the value of the Intel Architecture as a negative number. That’s because it costs money to maintain IA and I see that expense as no longer directly generating revenue for the company.
So we’ll see in this cloud fabric business the same thing we are seeing from Intel in mobile, where they are spending huge amounts of money to come from behind. It’s the same thing Microsoft is doing with Windows Mobile and it won’t work for either company. That’s why I predicted Intel would buy Qualcomm this year and Microsoft will buy RIM, because it’s too late to strictly build: at least some buying is in order.
And SeaMicro/AMD is going to make that even more imperative for Intel in the cloud space because things are going to get very exciting from here — much more exciting than had Intel bought SeaMicro.
Had Intel bought SeaMicro it would have become a new division or part of an old division at Intel with an old-line Intel manager in charge. Little AMD can’t do that: this is a big purchase to AMD. They’ll make SeaMicro its own division managed by the people who founded the company and have done a great job so far. So if Intel bought SeaMicro things would have immediately slowed down. But with AMD buying and throwing some real money into the SeaMicro business, things will actually get moving faster.
I don’t imagine SeaMicro will dump Intel processors. That would be stupid. But they’ll immediately add Opteron support. And since AMD gets the best Opteron pricing in the world, Opteron-based SeaMicro products will have a cost advantage.
But wait, there’s more! Since Intel has little presence yet in cloud fabric components and AMD has to know they are coming, they’ll do even more to be disruptive. Here’s where I’ll take a risk and predict that AMD will become an ARM licensee to extend SeaMicro’s fabric chipsets in that direction, supported, of course, by an ARM version of Windows and extensive Linux support.
ARM is the mobile AntiChrist to Intel.
And that’s why we’re seeing signs of panic in Santa Clara.