Gigaom’s Stacey Higginbotham reports that Amazon is “trying to design its own server chips,” having hired a number of former chip engineers as well as the former CTO at failed ARM server startup Calxeda. Calxeda ran out of cash in December after having raised over a $100 million from a number of key VCs and strategic partners.
Speculation continues to swirl that Google and Facebook are looking at building their own server chips as a means to cut costs and take Intel’s margin. There are other benefits to designing one’s own chips, namely to maximize a server/chip design for specific cloud computing tasks as well as the resulting energy efficiency/power savings.
The barrier had always been cost. Calxeda burned through over a hundred million and most have assumed the cost is in the hundreds of millions to design your own chip. But others, like AMD/SeaMicro head Andrew Feldman have suggested it could be done for tens of millions.
The reality is that if anyone’s going to take a stab at it, it’s going to be well capitalized tech giants that have granular knowledge of the ROI of matching custom servers and chips with compute tasks. Only the likes of Facebook and Amazon will have precise metrics on the benefits of building their own chip. My question is, if Amazon were to build their own chips, would they only use them in house or would they be so bold as to enter the server market? Or perhaps use them as part of AWS so that the cloud provider could further differentiate itself from competitors.