Quantitative trading companies have always been at the forefront of hardware, using programmable chips such as ASICs and FPGAs to power ultra-low latency trading systems in the era of high-frequency trading. Today, with machine learning strategies dominating the market, it’s all about who has the most computing power. The greater the computing power, the more attractive the employer.
Why do quant trading companies want GPUs? Who works with them?
GPUs provide the computing power that traders at these companies need to power their machine learning strategies. Companies like Jane Street are increasingly interested in alternative data such as satellite imagery, which can be complicated and difficult to integrate into business models without the help of an LLM to find a signal in the noise. Commercial companies use GPUs to train their proprietary LLMs. Machine learning quants use these LLMs alongside tools like PyTorch to create powerful trading models using statistical techniques like large-scale linear regression.
What does this mean for hiring? A larger GPU stack will generally result in increased machine learning headcount; Jane Street’s machine learning team has grown to the point where ML research and ML engineering have their own separate internship tracks. Several trading companies are also building AI labs to get the most out of their stacks, including Optiver and Hudson River Trading.
The infrastructure also creates new jobs for hardware engineering specialists who optimize these data centers to get the most out of GPUs. When Jane Street gave an overview of its Texas data center in May, co-chief technology officer Ron Minsky said building these centers was a “huge engineering challenge.” These centers require a delicate balance of power so that GPU racks are not overloaded, and they require complex network infrastructure to ensure data is transmitted quickly across the enterprise. Bloomberg reported yesterday that Hudson River Trading had built a data center in the caverns of a Norwegian mountain cooled by water from a nearby fjord if you’re interested in a more remote working environment.
Who has the largest GPU and compute stack?
We scoured the websites and job postings of major e-commerce companies to see how they describe their compute stack, focusing on the number of GPUs, CPU cores, and their total data storage capacity.
Key figures suggest that XTX Markets has the best on-site IT infrastructure. The British trading company led by Alex Gerko claims to have “more than 25,000 GPUs”, up from 20,000 in 2024. While other companies measure their data in petabytes, XTX claims to have “1 exabyte of storage”.
This is only one side of the coin; several large commercial companies have entered into agreements with cloud computing providers and hyperscalers to access their massive GPU clusters. Jane Street, which claims “tens of thousands of high-end GPUs,” also reportedly has deals with two major hyperscalers: CoreWeave and Crusoe. These two contracts would have a combined value of $19 billion. Citadel Securities, which reportedly does not have on-site data centers, has signed a major deal with Google Cloud in 2024 to power its quantitative research with Google’s infrastructure. Bloomberg reported yesterday that the company had “significantly increased” its use of TPUs (tensor processing units), Google’s custom chips. Citadel Securities told Bloomberg that they are four times more powerful than GPUs and 30% cheaper.
Hudson River Trading and Tower Research are each vague in descriptions of their GPU clusters, but both claim to have more than 100 petabytes of storage. Hudson River Trading says its GPU fleet is “larger than most national labs.”
Both DRW and Jump Trading have entered the computer trading space while creating their own stacks. Jump Trading’s website says it has “over 10,000 compute nodes” spread across “thousands of GPUs and TPUs.” DRW does not disclose its internal GPU footprint, but signed a deal with QumulusAI in August that provided the company with a dedicated cluster of B300 GPUs.
In-house GPU farms vs. cloud offerings: which is better?
The majority of large commercial companies operate a hybrid compute structure using both on-premises and cloud computing services. Ron Minsky, co-head of technology at Jane Street, said this week on his podcast that Jane Street found cloud offerings “really useful” despite its large internal stack. “They have abilities that we don’t have (so) sometimes they’re just better than us.” He said that while cloud infrastructure runs some workloads faster, “it often goes the other way.” Minsky said there were “real rewards” in having both at the same time, but it was difficult to achieve.
The downsides of having just one are obvious. If you build a market-leading GPU cluster today, it could lag the market in a few years. Ken Griffin said at a Citadel conference this week that he has “a growing set of concerns” about GPUs, including their lifespan: “People are stepping back and asking ‘what is the useful life of a GPU here, right now?’ newer chips like Nvidia’s Blackwell B200 can be more than 70 times more powerful depending on Nvidia’s specifications.
Griffin also warned of a point where GPU prices would “free fall” if supply exceeds demand, in which case companies that have invested billions in these fleets could lose their advantage. At Hudson River Trading’s UK entities, the company spent an additional £2.4 million per head on “property, plant and equipment”, although it is unclear whether any of this spending was specifically for AI infrastructure.
At the same time, the health of cloud computing companies is also in question. CoreWeave’s stock price is down more than 37% for the month to date, and options on its debt price a 60% probability of default over the next five years. If a compute provider were to go out of business, commercial companies using its architecture would have to find a new provider and could be subject to price gouging from other hyperscalers aware of the situation.
DRW, Citadel Securities, Jane Street and Susquehanna declined to comment. HRT and Jump Trading did not respond to a request for comment.
Know about other notable GPU fleets in quant trading companies. Contact us via any of the following methods for a strictly off-the-record conversation. Telegram: @AlexMcMurray, WhatsApp: (+1 269 237 3950). Click here to fill out our anonymous formor by email alex.mcmurray@efinancialcareers.com
Please bear with us if you leave a comment at the bottom of this article: comments are intermittently moderated by human beings. Sometimes these humans may be asleep or away from their desks. It may therefore take some time for your comment to appear. You must take full responsibility for the comments you post on this site. We will take reasonable steps to remove anything we consider offensive or inappropriate.
Gn bussni
