An interesting story a few days ago from our pals Cade and TPM put forward some interesting theories about how Google’s activities and acquisitions of companies and talent might add up to the searcher building its own server chip. Plausible? Yeah, I think it might be.
We’re not talking about a chip designed to compete with the highly sophisticated Xeon or Power processors. But doing its own customized ARM implementation? It could make a lot of sense, given Google’s scale and internal needs.
As noted in the story, it’s estimated that Google has more than 1.8 million servers – which is far more than almost any other commercial company – and its box count is only going to expand from there. With that scale, even minor increases in efficiency can add up to huge dollars. Tim and Cade go through the reasoning in their well thought-out article.
For my part, I’m wondering about implications. The business of designing and producing processors (and any other computing component) is a volume game: the more volume, the lower the cost of production and the larger base over which to amortize development costs. Google would farm out production, of course; you won’t see it building its own fab facilities. It minimizes its upfront investment and cuts production time. There’s enough slack capacity out there to give Google both primary and secondary supplier options.
As for the costs involved, ARM chips – even customized ones – can be turned out in huge quantities at very low cost. Google could consume quite a bit of volume just satisfying its own demand – and even build itself out a bit more in order to bulk up their cloud computing capacity as well.
Going a bit further out on this limb, if Google adds some significant IP to its ARM implementation, why wouldn’t they try to sell it to others? If they get more bang for the buck on their own stuff, odds are that the chip will work for others too. Google could license the IP on the chip to all comers – along the lines of its other initiatives and never have to hassle with actually making or selling hardware.
Rank speculation? Maybe. But it could pencil out, business-wise, assuming that the Google version of ARM can churn out a reasonable amount of work per watt and still run cool enough to be tightly packed together. Any ARM gurus out there want to comment? ®