Having bootstrapped and built a successful ultra low latency trading operation from start to finish, I can assure you that in this space the star is the technology. In other forms of HFT, such as statistical arbitrage, the traders are the stars. But my understanding is that the question specifically regarded low latency trading.
To give an example, a whole cohort of traders used a simple lead-lag signal between futures and equities. Up until mid 2009, it was possible to consistently make money merely by waiting for futures to move, sending the information down from chicago (where the CME sits) to new jersey (most of the equities exchanges are housed in new jersey) and trading the relevant stocks. At the time, a popular trade regarded ES (the on-the-run emini s&p 500 futures contract) and SPY (the s&p 500 depositary receipts, designed to replicate the daily returns of the S&P 500). Here, everyone and their mothers knew that this relationship existed, and this relationship was published in various papers, both academic and professional. In this context, the "trader" who brings the idea isn't really adding any value to the process. The success of the trade was wholly determined by the performance of the platform.
In a large HFT, the management cleverly structure the relationships so that the technology staff is lopped together with support and told that the traders bring value. This is intentional: there are numerous traders out there now who still believe they have value-add, when by now all of the tricks have been spilled.
The quoted estimate is averaged over a year of actual expenses, whereby relationships with clearing brokers and exchanges were set up (bypassing external vendors)
"We had access to all kinds of advantages in this regard that normal ccompanies wouldn't get." <-- that makes it sound like the tools to actually do ultra low latency trading are out of the reach of most people, and in reality they aren't. A competent person can build the business from 100K of personal savings :)