Waters Wrap: CME, Google and the pursuit of ultra-low-latency trading in the cloud

In November 2021, news broke that Google had invested $1 billion into CME Group and the two companies signed a 10-year partnership agreement. At the time, we wrote that this was likely to be the first of many such pairings between major exchanges and the Big Tech cloud providers. 

Sure enough, soon thereafter, Nasdaq announced a similar deal with Amazon Web Services, LSEG teamed up with Microsoft Azure, and, like CME, Deutsche Börse landed on Google. While not every major exchange has signed these types of exclusive partnership deals—most notably, the Intercontinental Exchange—every exchange understands that cloud is the future of exchange technology.

Still, while cloud may be inevitable, major challenges remain regarding latency, multicasting, and cost to the trading community.

To that last fear, a little over a year ago, we published a story in which three different Chicago-based proprietary trading shops expressed their displeasure about potentially having to move into a new datacenter as a result of the CME–Google tie-up. At the time, there were even rumors that CME might move its datacenter from Aurora, Illinois, which is about 35 miles from downtown Chicago, to Google’s facilities in Virginia.

“The answer is, prop traders don’t want to move,” said one executive.

“The industry is furious. It’s causing a lot of frustration,” said a second executive.

A third executive said smaller firms would struggle to cover the cost of setting up new network infrastructures to continue their trading activities.

“We just wish they’d come out and say” if they’re going to relocate, said the first source.

Well, that individual and the rest of the Chicago trading community now have more clarity on the exchange’s transformation plans, and it’s mostly good news. Last week, CME announced plans to build a new co-location facility in Aurora and that it would be housed in what’s called “a private Google Cloud region.” (More on the definition of that in a little bit.)

There are a lot of interesting things at play here, and I’ll do my best to add some additional context to the press release because this is much more than version 2.0 of the originally announced partnership. Actually, I believe that this marks a major milestone in how markets of the future will operate.

From the release: “The private Google Cloud region will host Google Cloud’s new, industry-first specialized platform designed to support global trading of CME Group’s futures and options markets, and offer derivatives traders cloud-based, ultra-low-latency networking and high-performance computing.”

The emphasis on that last part is my own because that’s what I first want to focus on.

Last week, I spoke with Ken Vroman, chief transformation officer at CME Group, who is overseeing this project from the exchange’s end. He said that this week will mark his 23rd anniversary at the company. So, the man has seen tectonic shifts in technology and how these evolutions affect markets firsthand.

“It took us two years to make the announcement we just made,” he said. “As I would describe it, this is a slow, careful walk with our customers and regulators because we absolutely have to get this right.”

It took us two years to make the announcement we just made, As I would describe it, this is a slow, careful walk with our customers and regulators because we absolutely have to get this right

Ken Vroman, CME Group

Or, as I would describe it, it’s a slow, careful walk to go really fucking fast.

Vroman said that over the last two years, they’ve “done a lot of engineering work to prove that this both works and how to do it.”

CME is not alone in this; Nasdaq has been discussing low-latency trading and multicasting in the cloud for several years now. But as you can probably derive from the statements made by the execs at those prop-trading shops, HFT—and the hardware they’ve built in the Aurora facility to trade fast—is very important in Chicago. The work Nasdaq is conducting in the cloud is important, but the HFT world is taking the most interest in what’s going on in Aurora (and, thankfully, not Virginia). And Google—and the tech behemoth’s reputation—will play a big part in this experiment.

“It’s a good proving ground for [Google] to say, ‘We have multicast in the cloud; we’re going to have high-precision timestamps and clocks that are synced to the requirements that are expected from regulators,” a source with knowledge of Google’s thinking told me. “These are all new components that are not available in cloud today that [Google] expects to have and bring to market.”

Or, as Vroman noted: “What you don’t see anywhere are ultra-low-latency markets operating in the cloud. … That’s where we are trying to get.”

Rohit Bhat is Google Cloud’s managing director for capital markets, digital assets, and exchanges. Via email, he told me that this project is unique, even for a company like Google.

“Google Cloud’s private cloud region mirrors many public cloud features but with a key distinction: it houses a purpose-built capital markets platform, optimized for ultra-low-latency [trading] with determinism to enable trading and high-performance computing. This region prioritizes CME Group client access to meet their performance demands,” Bhat says.

Now, if you’re like me, you’re probably asking yourself, “What the hell is a ‘Google Cloud region?’” Or, maybe you’re smarter than me—kudos to you. For those who don’t know, though, a Google Cloud region is a place that provides core services, such as storage, networking, and compute, as well as advanced analytics and AI services—staples of Google as a company—to support customers in “the region.”

According to Bhat, each cloud region is made up of multiple zones, which are collections of physical resources (servers, networking infrastructure, etc.) in one or more datacenters. Currently, Google Cloud has 9 regions in the US, 12 in North America, and a total of 40 regions with 121 zones globally.

This new private cloud region (and, thus, co-lo facility) is being built, according to Vroman, “really close” to CME’s current Aurora facility. Construction on the region will begin later this year. A timetable for completion hasn’t been released.

At stake

Beyond serving as an interesting proving ground, CME and Google both stand to benefit from this new facility once it is live.

According to Vroman, this move will allow CME to spin up markets more easily and introduce new products to those markets. Users will also be able to leverage real-time data and analytics in the cloud that are coupled with Google’s AI and ML engineering capabilities, and those will sit “right next to the matching engine in the cloud,” he said.

A big part of the future is this allows us to position ourselves to have our markets in the cloud—not next to it, but actually in it

Ken Vroman, CME Group

Furthermore, the exchange will be able to spin up new products and services “during the week,” as opposed to the current setup, where projects and testing often happen at night and on weekends and thus take more time to roll out. Finally, CME will look to leverage Google’s cybersecurity and resiliency in the cloud expertise.

“A big part of the future is this allows us to position ourselves to have our markets in the cloud—not next to it, but actually in it,” Vroman said.

CME Group has already moved 26 petabytes of the exchange’s data into Google Cloud. Vroman said that for non-ultra-low-latency workflows (i.e., data and clearing), about two-thirds of that has been moved to the cloud. 

Vroman also said that before the new co-lo facility goes live, clients will likely be live on a new disaster recovery (DR) site located in Dallas. Once the new co-location facility in Aurora is built and goes through testing, CME will give customers 18 months’ notice before the exchange moves its markets to the new platform.

“We think we’ll be up and running with disaster recovery well before the building is built,” Vroman said. “I think we did everything within our capability to minimize that potential disruption to our client base and smooth that transition down the road as much as possible.”

Again, right now there isn’t a predicted date for when the Google Cloud region and co-location facility will be live. Another question is the cost to the trading community (once more, think back to the comments of those aforementioned execs at the prop-trading shops). Vroman said that at this point, the exchange is not ready to talk about cost, offsets, or discounts.

“It’s too early to talk about the economics associated with the various solutions, so I think it would be premature to speculate on that,” he said. “I think it’s fair to say that CME [views this as being] important—our customers are everything to us, so facilitating that transition in whatever way that may be is something that is important to us, but I wouldn’t speculate on cost or subsidies at this point.”

As for Google, if they crack the ultra-low-latency trading and multicasting in the cloud conundrums, that will be quite the feather in their cap and it will open up a world of opportunities for exchanges around the globe—for Google customers and for customers of other cloud providers who will learn from Google. (The same is true for AWS and Microsoft Azure if they solve these problems first.)

Specific to Google, though, this is an important project for its expansion plans. I’ve written about this before, but Big Tech companies view the capital markets as offering many potential areas of opportunity for growth. They do not simply want to be data storage companies; they want to be providers of critical infrastructure components of the financial markets. 

Well, another key piece of this announcement was this: Once the facility is live, whether a client chooses to run their trading workloads on Google Cloud’s specialized platform or on their own hardware in the new co-location facility, they will experience the same network latency to the exchange’s matching engines hosted on Google Cloud.

Once customers are in the new facility and see all the cool new tools being made available, will they want to spend quite so much money building their own proprietary tools and services?

That means that customers do not necessarily need to set up their trading operations on Google’s platform, at least initially. But, once customers are in the new facility and see all the cool new tools being made available, will they want to spend quite so much money building their own proprietary tools and services? Or, will it start to make sense to explore opportunities working with Google directly? These are, after all, sophisticated HFT firms with deep pockets.

There will also be current and future CME customers who either don’t have the budget or engineering know-how to take full advantage of the new facility. Once again, I’m sure that Google sales reps will be more than happy to assist them in bridging those gaps.

Finally, while the ultra-low-latency trading component of this project is fascinating, the source with knowledge of the project said there will be plenty of customers who aren’t as latency-sensitive. “They don’t need the hyper-optimized deployment of co-lo and could benefit from a more effective operating model of existing on the cloud,” the source says. “Or, they can look at it as, what trading strategies have we not been able to launch because we didn’t have the infrastructure in co-lo or we had supply chain hurdles, couldn’t get it set up in time, and had to wait and missed a market opportunity? Well, now they can go to the cloud and create a bare metal instance and start trading from that once this is live.”

Now, this column has been a little bit cheerleader-y. Being a technology reporter, there are times when you need to be critical, but there are also times when a particular tech project is cool and has the potential to impact a particular market massively. For example, I wish that I had been a reporter when the Bloomberg Terminal was first rolled out (and hopefully, I would’ve seen its potential rather than shitting on it).

Cloud computing is inevitable, but how those cloud infrastructures, trading platforms, and applications are developed and released is anything but inevitable. If you think I’m missing something from this column, please do let me know at anthony.malakian@infopro-digital.com.