More than $300 million.

http://www.telegraph.co.uk/technology/news/8753784/The-300m-...

I'd doubt whether there'd be much money in it from an HFT perspective; traditional trading might benefit from it, however.

HFT shops co-locate, so they're running their code right there next to the exchange.

Edit: In case I've missed a point in the article, it wasn't intentional, it's banned at work so wasn't able to read it.

The use case for this would be communication between different colocations, which is where the networking "race" really is for HFT. For example, to arb products on the CME vs. the equities exchanges in the New York area (really colocated in various places in NJ).

At any rate, as the author mentions this isn't really a feasible technology.

This would be true if there were only one exchange. Instead, there are many around the world, and information provided by one can influence another.

I was at the R/Finance 2012 conference in Chicago about a month ago and one of the speakers commented that about "1 millisecond was worth $10,000," but I forget whether that was an annual number as in "1 millisecond saved is $10,000 a year" or "We can make $10,000 every millisecond." Hopefully someone else can corroborate this.

it's not 10,000 a millisecond. Trading desks normally think in terms of daily PnL - he prob meant every 1ms latency reduction would make them an extra 10k a day.

Yes, now that you mention it, that sounds right.