Jane Street, HRT Spend Billions on AI Compute Infrastructure

High-frequency trading giants are pouring billions into artificial intelligence infrastructure in a dramatic shift from their traditional speed-focused competition. Jane Street, Hudson River Trading, and rival market makers now view compute capacity as non-negotiable for survival, with spending on processing power surging between 10 and 20 times higher than pre-ChatGPT levels according to industry executives.

The Compute Arms Race Reshapes Trading

Market makers historically competed over nanoseconds, positioning servers closest to exchange infrastructure. Today's battle centers on raw processing power for machine learning models that predict market behavior. Firms are constructing proprietary data centers, signing multi-billion dollar cloud deals, and investing in semiconductor startups to secure advantages in an increasingly AI-driven trading landscape.

Jan Boomaars, CEO at Amsterdam-based Optiver, framed the investment as existential. "If you don't want to be in that game, it's going to be difficult to stay involved in the more systematic trading strategies which are required today to be a market maker," he said. Optiver has billions in capital expenditure planned and is replicating its US data center model in Europe and Asia.

Building From Scratch in Remote Locations

Hudson River Trading operates servers in a former Norwegian mine 3,500 miles from Wall Street, cooled by fjord water. XTX Markets constructed a €1 billion data center in Kajaani, Finland, with a second facility launching in 2027. Jane Street built a Texas facility housing over 4,000 Nvidia GPUs across 56 liquid-cooled racks, connected by thousands of kilometers of internal fiber.

The strategic shift prioritizes abundant power and cooling over geographic proximity to exchanges. Tower Research Capital executives noted resource demands from quantitative researchers escalated dramatically within the last 12 months. "There is limited supply," said John Cogman, Tower's chief investment officer, explaining the firm's decision to build and operate its own undisclosed data center.

• Hudson River spends $1 billion annually on AI, with the figure trending upward
• Jane Street generated nearly $40 billion in net trading revenue last year
• Susquehanna is investing in a startup developing wave-powered floating data centers

Strategic Cloud and Chip Partnerships

Citadel Securities took a different approach, doubling down on partnerships with hyperscalers rather than building proprietary facilities. The firm significantly scaled its use of Google's tensor processing units over the past 18 months, achieving workloads up to four times faster than GPUs with 30% lower costs according to Google Cloud.

Jane Street led an investment round valuing AI chip startup Etched at $21 billion, more than quadrupling its valuation in nine months. The market maker became Etched's first customer for custom chips designed to run at lower voltage for better performance. Hudson River and Jump Trading are also testing Etched chips and hold equity stakes.

Jump Trading backed Positron, another chip firm whose valuation jumped nearly fivefold in seven months. "We know chip design and can point out bottlenecks in certain rare cases," said Saurabh Sharma, head of investments at Jump. "That feedback is unique and appreciated by these next-generation chipmakers."

Release Date and Current Deployments

These infrastructure investments represent ongoing deployments rather than a single launch. Officially released on October 22, 2026, this reporting captures the current state of a multi-year buildout cycle. XTX Markets expects its first Finnish data center online by year-end 2026, with the second facility switching on in 2027. Hudson River signed its CoreWeave deal in August 2026 and its Lambda agreement in May 2026.

Multi-Billion Dollar Commitments

Jane Street's infrastructure spending dwarfs industry norms. The firm purchased a $1 billion equity stake in CoreWeave while pledging $6 billion for compute access, totaling $7 billion. It separately committed $13 billion to data center startup Crusoe and invested in Nscale, which faces investor pressure over valuation ahead of a planned October 2026 IPO.

Hudson River diversified across neoclouds like CoreWeave and Lambda, hyperscalers including Google and Amazon Web Services, and proprietary facilities. Iain Dunning, head of AI at Hudson River, said the firm could meet all needs with one or two sites a decade ago. "The decisions we are making now will have effects many years into the future in terms of the contracts we are signing and the life cycle of the chips as well," Dunning noted.

Risks and Competitive Pressure

The spending spree carries significant risks. Chip technology evolves rapidly, potentially rendering expensive systems obsolete. Financial markets remain notoriously difficult to crack with machine learning due to noise and increasing competition. Trading revenue levels supporting these investments may not persist indefinitely.

Cloud provider delays or lost contracts could disrupt compute access, threatening both research capacity and talent recruitment. Data scientists and mathematicians are unlikely to join firms with constrained resources. Yet most executives view inaction as more dangerous than overspending.

"It's much more expensive and you're not in charge of your own destiny" when relying solely on cloud providers, Boomaars said. "We have to be in charge of our own compute infrastructure." The comment reflects a broader industry consensus that compute capacity directly correlates with competitive survival in AI-powered trading.

What This Means

High-frequency trading firms are fundamentally restructuring their operations around AI infrastructure, treating compute access as critical as trading algorithms themselves. The 10-20x spending increase since late 2022 signals a permanent shift in capital allocation priorities. Firms backing AI chip startups and constructing proprietary data centers are betting that technological advantages will generate returns far exceeding infrastructure costs. As Positron CEO Mitesh Agrawal noted, "If you find that advantage, then it doesn't matter how expensive the infrastructure is. The return on investment becomes very, very clear." Whether these multi-billion dollar wagers pay off will define the next decade of quantitative trading.