Wall Street traders are building AI data centres to stay ahead

High-frequency trading firms are treating artificial intelligence compute as strategic infrastructure, with Jane Street, Hudson River Trading and other market makers spending heavily on data centres, cloud capacity and specialist chips. The shift shows how the AI arms race is spreading well beyond the usual technology giants.

According to reporting summarised by The Decoder, the largest firms are investing billions in cloud contracts, start-ups and facilities of their own. Their motivation is practical: machine-learning systems increasingly help them predict market movements, but those systems need enormous amounts of computing power to train and run ¹.

One executive estimated that compute spending across the sector is now between 10 and 20 times higher than it was before ChatGPT launched. That is a dramatic change for businesses better known for low-latency trading than for operating hyperscale infrastructure.

Why traders want their own AI infrastructure

The firms are pursuing several routes at once. XTX Markets is building a data centre in Finland, while Jane Street operates a Texas facility with more than 4,000 Nvidia GPUs. Hudson River Trading is reported to spend about $1 billion a year on AI and has signed a multiyear agreement with CoreWeave, the cloud provider that specialises in accelerated computing ¹.

Jane Street has also invested $1 billion directly in CoreWeave and committed a further $6 billion to its computing services, according to the same report. That combination gives the trading firm access to capacity without requiring every workload to run in a facility it owns.

The company’s own account of its Texas site gives a sense of the physical scale. Jane Street says the centre contains 4,032 GPUs across 56 racks, uses liquid cooling and is connected by roughly 8,000 kilometres of fibre. An official tour shows the infrastructure up close, rather than presenting AI as an abstract software service ².

For trading firms, owning or reserving capacity is also a hedge against the wider shortage of advanced hardware. If competitors are buying the same cloud capacity and GPUs, waiting for prices or availability to improve could mean giving up a technological advantage before a model has even reached production.

The chip bet

Jane Street is taking a more specialised gamble as well. It led a funding round for Etched, an AI-chip start-up, and became the first customer for the company’s processor. Etched says its approach is designed to deliver more performance while using less voltage, although the commercial and technical risks of betting on a young chip company remain substantial ¹.

That risk matters because AI hardware ages quickly. The Decoder’s report describes rising costs and fast-obsolescing chips as major concerns for the industry. A data centre can be a long-lived asset, but the accelerators inside it may need replacing far sooner than the building, cooling system or network fabric.

The result is a difficult balancing act. Firms need enough capacity to keep their models competitive, yet they do not want to spend billions on hardware that becomes uneconomical before it has paid for itself. Cloud contracts, direct investment and experimental chips are different ways of spreading that risk.

Our opinion

This is a useful reminder that the AI boom is not only about chatbots and consumer apps. Trading companies have unusually strong incentives to pay for better predictions, lower latency and more efficient infrastructure, so they are willing to build capacity that would look excessive in a normal corporate IT budget.

The most revealing detail is not any single investment. It is the combination of data centres, long-term cloud commitments and chip funding. Taken together, those moves suggest that leading trading firms see compute as a core competitive asset, closer to a private power station than an ordinary software subscription.

That does not make every investment a winner. Hardware can age, start-ups can miss their targets and AI spending can outrun the value of the models it supports. But the scale of the spending makes one point hard to ignore: Wall Street’s fastest traders are now competing for the same scarce AI resources as the technology industry, and they have the money to make that competition even more intense.