Why Cisco should merge with Dell
Disco – A real data center powerhouse
Comment Cisco Systems has unleashed a world of hurt on its shareholders, masquerading as a consumer company that sells Flip phones, Umi telepresence, and Linksys routers. There's nothing wrong with selling gadgets to consumers. Apple does it brilliantly, and profitably. But Cisco isn't Apple.
In a sense, Apple might be a place where Cisco can find comfort in its time of trouble. People forget how screwed up Apple was in the 90s. The fact that Apple turned itself around can provide a ray of hope to Cisco shareholders and the company's beleaguered CEO, John Chambers. If Chambers can keep his job.
Server and PC marker Dell has also had its share of problems in recent years – accounting scandals, poor sales execution, exploding products – but of late, it has been on the upswing. It has had some success building custom servers for hyperscale data centers. It bought its way into the storage racket proper instead of just reselling other companies' kit (LSI and EMC, mainly). And it moved into services with its Perot Systems acquisition.
The rub is that Dell's networking business is no more real than its storage business used to be. It is merely rebranding kit from Brocade and others. Until Dell fixes that, it will not be a credible infrastructure stack player.
Meanwhile, Cisco has done a remarkable job building a server business out of nothing with its "California" Unified Computing System blade and rack servers – and with a fair amount of actual engineering – but it is not a profitable volume manufacturer any more than IBM was. Big Blue couldn't cut it in printers, memory, disk drives, and PCs, and Cisco is having trouble with anything other than core switches and routers.
One could argue that Cisco should not have entered the server market at all, but that is an over-simplified view. No doubt, the move into servers has not only annoyed Cisco's former partners in the server business, but compelled them to start selling their own networking products, either rebadged (from Brocade, Juniper, or Blade Network Technologies) or built with the help of ODMs (Sun and now Oracle does this, and so does Dell).
Hewlett-Packard has had a switching business for decades, and the company, like Cisco, predicted that over time, server and storage networking would converge and virtualize. This is the consensus view now among all server and network equipment makers these days, but it was probably a radical view years ago when Cisco hatched its plans for Project California. Once Cisco figured networks would converge and flatten, it had no choice but to enter the server space.
That flattening has been more of a problem than server-storage convergence. Cisco bought its way into switching, and convinced customers that the best way to build networks was with three tiers: an access layer, a distribution layer, and a core layer. But as companies go to build clouds of all types and styles, they want to get rid of that middle distribution layer and use leaf-spine architectures that are cheaper and, quite frankly, scale further and with less latency on fewer hops from server to server or server to storage.
This, more than anything else, is what is hurting Cisco's switching business. Mellanox, Juniper, Brocade, Extreme Networks, Arista Networks, and IBM can make faster and lower-cost top-of-rack switches, and there are cheaper alternatives to Cisco's end-of-row plain-vanilla Catalyst 6000 and converged Nexus 7000 switches.
Chambers may believe that the price/performance of the Nexus 7000 compared to earlier Catalyst products is what is driving Cisco's switching biz down. But it is probably more accurate to say that Cisco's prices are too high relative to the competition, which is like a swarm of piranhas around the juicy Cisco carcass that has fallen into the Amazon.
Maybe Cisco should have not bragged about its router and switching hardware margins quite so loudly? You sow the seeds of your own destruction. Or, perhaps, resurrection.
Cisco's entry into the server market in March 2009 caused the server makers to get their networking acts together, or at least make an attempt at it. And this will presumably be good for customers, although I would say that tightly integrated stacks usually tempt vendors to milk their customers.
Think IBM's AS/400 midrange customers, who paid for every point of market share that Big Blue gained in the Unix racket in the past fifteen years with exorbitant hardware and software pricing, or IBM's mainframe customers, who pay felonious prices for their hardware and software because they really don't have another choice.