Chip giant Intel (INTC +0.33%) currently has a lock on the server chip market, a bright spot for the company, with a market share in excess of 90%. Demand for server chips is strong, with companies like Google (GOOG +1.50%) needing an increasing number of servers to power its myriad of web services, and Intel expects double-digit growth rates for its server business going forward.
A recent report from Bloomberg, however, suggests that Google may be looking into designing its own server chips based on the ARM architecture. Google accounts for 4.3% of Intel's total revenue, and that could lead other web giants to follow suit in ditching Intel. But Bloomberg's arguments are weak, and none of the evidence presented seems to point to its conclusion.
Google is designing something, all right
Bloomberg cites "a person with knowledge of the matter," along with a single job opening for a digital design engineer, as evidence of Google's plan to build server chips. The responsibilities of the digital design engineer are:
Digital design of key components in server, networking and storage systems that power Google's hyper scale data centers.
It's clear that Google is looking to design components which will go into its servers, but there's no indication that the actual server chip is the target. The job requires experience with ASIC, or application-specific integrated circuit design, but servers and network systems contain all sorts of ASICs that do all sorts of things. Network switches from big vendors like Cisco, for example, contain custom ASICs, and with Google increasingly designing its own networks and bypassing Cisco, the job opening could be related to networking. Or just about anything, for that matter.
So Google is designing something, but we have no evidence that it's a server chip.
Some things make sense, some things don't
Google and other web giants have been reducing the cost to build out their massive data centers by cutting out middle men such as server and networking vendors. HP and Dell have been feeling the squeeze as these big companies have started to design their own servers and farm out manufacturing to Asia. Generic networking hardware has been putting pressure on Cisco's proprietary switching and routing business.
It makes sense for Google to design its own servers. Servers from HP and Dell are essentially just server components thrown into a box and sold at a premium, so designing custom servers would have a high return on investment for Google. It allows the company to pack more power into a smaller space, and cutting out the middle man reduces costs significantly.
Custom networking also makes sense for Google, given the company's massive and unique networking needs. Generic hardware, along with software-defined networking technology, allows the company to avoid being locked into using Cisco's proprietary products. This saves money while allowing Google to design an optimal network for its specific needs.
Server chips, however, are a different story. Unlike servers and networking hardware, designing a server chip that is powerful and efficient enough is not an easy task, even for Google. Intel has decades of experience, and with almost the entire server market to itself, the x86 architecture is the de facto standard. Any server software would have to be rewritten for the ARM architecture if Google used its own chips.
It's not that Google couldn't design an ARM server chip. It's that the chip wouldn't be able to compete with Intel's products. Intel offers both high-performance Xeon server chips as well as high-efficiency Avoton server chips meant for micro-servers, and building an ARM-based alternative that comes close to achieving the same performance would be an extremely long and costly process. It would also require an entire team, not a single digital design engineer, to complete. The return on investment simply isn't there for Google.
The bottom line
Intel dominates the server chip market for a reason, and I severely doubt that Google will be able to create an effective alternative to Intel's products. It's more likely that Google is simply designing ASICs to further its custom server and networking initiatives, and that Intel will continue to power Google's data centers for years to come. The Bloomberg article seems to be jumping to conclusions that the evidence doesn't support.
