Who cares about exchange consolidation when you can talk about data center consolidation? Although, the former is now in direct competition with the latter and mergers in both sectors are in part driven by the competition this all creates. Luckily this should all lead to better service and price for the end user. More locations, more connectivity options and (hopefully) more competitive pricing. There will always be the contingent of trading firms that need to be in the same room as the matching engine, but that contingent really only accounts for about 2% of trading firms in the US. So if Equinix, for example, can undercut NYSE cost per cabinet by 30%, and then Telx does the same to Equinix by another 30% the lower cost will be hard for that hedge fund in Greenwich to pass up.
McPartland says this has meant the market in New York has changed considerably over the last year. “There are not a lot of players left,” he says. And from where he sits, he can hear a lot of CTOs showing interest in smaller players that can prove they have the required latency, connectivity, proximity and power resources, suggesting that the time could now be right for smaller players that have ticked the right boxes to name their price.
The folks at DatacenterDynamics take us on a trip down memory lane trying to understand how we came to this latency-crazed trading world we live in today. Was funny for me to think about how cool those touch screens were in the late 90′s, especially with my iPad sitting on my desk over 10 years later.
“I was supporting the equities desk at JP Morgan when they brought in touch-screen technology for brokers so they could electronically transfer information to the exchange floor where a paper ticket would print out. The fact we could send down something electronically and print it out was pretty amazing back then.”
Great feature piece originally printed in the Sunday Business section of the times examining both the drivers and requirements for high speed trading. The article cites TABB Group’s estimate for high frequency trading in the US equity market (56%) and includes a few of my comments.
According to Kevin McPartland of the TABB Group, high-frequency traders now account for 56 percent of total stock market trading. A measure of their importance is that rather than charging them commissions, some exchanges now even pay high-frequency traders to bring orders to their machines.
High-frequency traders are “the reason for the massive infrastructure,” Mr. McPartland says. “Everyone realizes you have to attract the high-speed traders.”
NYSE’s new Mahwah data center is finally set to go live this coming Monday. It will only be for a few stocks, but after years of working, waiting and talking about the project its exciting to see this big investment (and risk) for NYSE get moving.
The only signage noting the property’s high-profile tenant is a plaque near a buttonwood, McPartland noted.
The plaque notes the data center “houses the most technologically advanced financial marketplace in the world.”
“It could have just as well been Area 51 or CIA headquarters,” he wrote. “Fences, cameras, guards and ID scanners are everywhere, yet the buttonwood trees and picnic tables out front made it feel like a nice place to spend an afternoon.”
This story gives an overview of the presentation I gave at the Accelerating Wall Street conference in May 2010. Much of the data was based off our our Sell Side Technology study from December 2009.
“Clearly the sell side loves its data centers,” McPartland told the audience. “There’s a lot of horsepower that has to sit behind these equity businesses. … It’s getting more and more complex to manage the infrastructure.”
And more costly. Equity firms spent $1.8 billion last year on data centers; half of that total came from sell-side shops, according to the TABB Group report, which predicts that the sell side’s use of data center space will increase slightly in 2010.
Too many dot coms brought us a bubble. Too much leverage brought us a bubble. Too many houses sold to people who couldn’t afford them – you guessed it, a bubble. In all of those cases demand was so strong that supply was generated as fast as possible to satiate those in need and to make money for those willing and able to sell. The data center space right now looks eerily similar.
Co-location, Software as a Service, Platform as a Service, Infrastructure as a Service and Anything-else-you-can-think-of as a Service has driven demand for data center space through the roof. Throughout my two days of meetings at the SIFMA Technology show in NY I had at least a half dozen meetings with people building data centers. Telco providers, data center providers, software providers and others who in previous years made their money in places other than data centers are creating raised floor space with power and cooling at an amazingly rapid pace. Creation of millions of square feet of anything in such a short time scares me a bit.
However, as much as I’d love to prophesize the next bubble I can’t see how demand for data center space will slow. Through the worst recession in decades, banks and other trading firms snatched up prime space at thousands of dollars a rack as fast as it could be built. New regulations around co-location might make pricing more transparent, but it will not stop the demand for space near top matching engines.
On the other end of the low-latency spectrum, the move to SaaS, Cloud and other methods of infrastructure outsourcing are in full swing and there is no turning back. What do these things all have in common? They need lots of data center space. The words describing the technology may be buzz-worthy but the approach itself is not. With even Microsoft beginning to provide its beloved desktop applications online it’s clear the momentum here is permanent. So where desks with room for a PC sold like crazy in the 1990’s, this decade will see all of that horsepower moved to datacenters. Case in point, Office now runs on millions of PCs; in a few years it will run on servers utilizing millions of square feet of data center space.
Smaller, faster and more power efficient servers and networking equipment could also keep data centers empty, as they can do much more with less. However, the compute and bandwidth demands of financial services firms (and the rest of the media-obsessed world) will ensure that faster and smaller technology will not shrink data center footprints – they will simply allow their users to crunch more data, trade more shares and stream more HD video with the same space they have today.
So this time around despite frantic building and quick selling, no bubble has emerged. And unless Congress bans Wall Street from using computers altogether and we all give up our iPhones, a data center bubble will remain science fiction.
The CFTC should stick to regulating markets and not try to regulate data center providers. The recently released proposal by the CFTC does a good job of preventing providers of data center space from competing on their merits and pricing according to market demand. Don’t get me wrong, I appreciate the spirit of the proposal –make pricing transparency and allow “all qualified market participants willing to pay for the services” to gain access. However, the wording leaves me wondering how some providers could stay in business.
All “that offer co-location and/or proximity hosting services must ensure that there is sufficient availability of such services for any and all willing and qualified market participants.”
This sound reasonable. If someone wants in and can afford it, they should be allowed access. The validity of the statement is killed in the next line, however: “if the availability of a service became limited, thereby leaving some market participants or third-party hosting providers without adequate access, the Commission would not view access to those services as open and fair.” So does that mean if Equinix, CME or NYSE runs out of space in their data centers they will be out of compliance until they can build out new real estate? Data centers are not virtual assets; they cannot be expanded at the push of a button. Despite a huge push to increase the inventory of prime data center space in major market centers square footage and more importantly the power to support is very much finite.
The provision relating to ‘‘Fees’’ would ensure that fees are not used as a means to deny access to some market participants by ‘‘pricing them out of the market.’’
What happened to supply and demand? Isn’t that what a capitalist society is based on? If trading firms are willing to pay $10,000 per server cabinet and the provider can sell enough of those cabinets to stay in business, then there should be no issue. Only a few paragraphs’ prior in the proposal, it states that “equal access” means providing services to those “participants willing to pay for the services.” A good portion of trading firms in the US do not use trading strategies with enough latency sensitivity to make paying the premium for co-location necessary or a worthwhile business expense. And of those, many can in fact afford it but don’t need that access – a.k.a., they are not willing.
By all means, the CFTC should encourage all co-location providers to disclose pricing and availability, but let’s not mess with the supply and demand economics that make the US capital markets some of the most efficient in the world.
I could go on, but I think you see the problem here. Time for the CFTC to go back to the drafting board.
From the Sunday edition of The Record, a interesting article focusing on the move of “Wall Street” to New Jersey. The vast majority of US equity order matching actually goes on somewhere in NJ, regardless of where the exchange itself is officially incorporated. If it wasn’t for the relatively small bit of trading that happens on the NYSE floor on Broad Street in Manhattan, it would be safe to say all US equity trading happens in NJ:
“NYSE is looking to move the epicenter of U.S. capital markets from Broad Street to Mahwah,” said Kevin McPartland, a senior analyst with the Tabb Group, a New York consulting and researching firm. McPartland estimates 8.5 billion shares are traded on U.S. equity markets every day.
All of the financial services focused data centers I’ve spoken with have expressed much more concern about running low on power than on running out of floor space. I’ve seen cages with my own eyes that are half empty because the servers used by the resident require double the power generally available per rack in that data center. Available kilowatts per rack are skyrocketing, with many in the low double digits. We will see even higher power densities in the not to distant future:
“Servers are getting faster and faster and because they are faster, they are consuming considerably more power,” said Tabb Group senior analyst and report author Kevin McPartland. “To make the best of use of the limited prime data center space around the major market centers, they need to have very high power density to be able to support as many servers as the clients in the data centers want to put in them.”
Reducing latency is critical for those in the ultra-low latency high frequency trading game; but the number of firms that fit that category isrelatively small. DirectEdge has a new offering that it hopes will appeal to the entire spectrum of latency sensitivity.
“It’s a little bit of a misconception that high-frequency firms will pay anything to save latency,” said Kevin McPartland, senior analyst at TABB Group, a financial markets research firm. “They’re still running businesses and want to save costs.”
“There’s a lot of smaller broker-dealers and hedge funds that until now just kept their servers in a closet in their office,” McPartland said. “They’re realizing that’s not a good long term strategy and need to get those servers in a data center.”