What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
High-frequency trading firms are treating AI compute as a strategic resource, not an ordinary cloud expense. Jane Street illustrates the pattern: it committed about $6 billion to CoreWeave’s AI cloud, bought $1 billion of CoreWeave Class A stock, is reported to be considering a 100–200 megawatt data center, and led a $700 million funding round for inference-chip maker Etched after testing its hardware. Together, those moves secure near-term GPU access while building influence over future capacity, facilities and chip design.
The aim is an advantage in availability and deployment speed. Training and inference workloads need accelerators, high-bandwidth networking, storage and reliable power; a firm that can obtain and schedule those resources faster can iterate models while competitors wait. The facility and spending figures below are attributed to company announcements or reporting, not presented as audited Jane Street disclosures.
Why are high-frequency trading firms investing in AI compute?
Machine learning has been part of Jane Street’s quantitative research for years. Its official overview says the firm trades across more than 200 electronic exchanges, where model quality and the speed of research can affect pricing, hedging and execution decisions. As models grow and datasets become noisier, access to accelerators and the surrounding infrastructure becomes a recurring operating constraint.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchFor a market maker, a delayed experiment can be a competitive cost. Compute scarcity can stretch training schedules, postpone model releases and force researchers to share hardware with unrelated workloads. Long-term capacity agreements, equity stakes and owned facilities attack different parts of that bottleneck.
Popular on Amazon
- NVIDIA Volta GV100 Architecture — 4,608 CUDA Cores, 640 1st-Gen Tensor Cores delivering 14 TFLOPS FP32 and 112 TFLOPS deep learning performance for AI training, inference, HPC, and scientific computing workloads
- 32GB HBM2 ECC Memory — 900 GB/s Bandwidth — High-bandwidth memory on a 4096-bit bus with ECC error correction provides the memory capacity and throughput required for the largest AI models, simulations, and datasets
- PCIe 3.0 x16 Interface — 250W TDP — Standard PCIe Gen3 connectivity with passive cooling designed for enterprise rack server deployment in HPE ProLiant, Dell PowerEdge, and Supermicro platforms with adequate chassis airflow
- NVLink — Scale to 96GB Unified Memory — Connect two V100 GPUs via NVLink at 300 GB/s bi-directional bandwidth to scale GPU memory from 32GB to 96GB for larger AI training and HPC workloads
- Multi-Precision Computing — Supports FP64 (7 TFLOPS), FP32 (14 TFLOPS), FP16 (112 TFLOPS) and INT8 precision modes for flexible deployment across training, inference, and scientific simulation workloads
- Neocloud contracts provide immediate access to large GPU clusters, dedicated storage, networking and managed deployment.
- Equity investments align the trading firm with a provider or chip company whose expansion it needs.
- Owned or sponsored facilities provide control over power, cooling, network topology, hardware scheduling and deployment dates.
- Custom-chip investments can target better inference efficiency and an additional source of supply beyond general-purpose accelerators.
What did Jane Street agree to with CoreWeave?
CoreWeave announced on April 15, 2026 that Jane Street committed approximately $6 billion to its AI cloud and invested $1 billion in CoreWeave Class A stock. CoreWeave said the commitment covers access to next-generation compute across multiple facilities plus supporting software and services.
| Element | Publicly reported figure or detail | What it does |
|---|---|---|
| AI-cloud commitment | Approximately $6 billion, announced April 15, 2026 | Reserves access to next-generation compute and related services across multiple facilities |
| CoreWeave equity investment | $1 billion in Class A stock | Creates a financial relationship alongside the service contract |
| Earlier financing | $650 million round in 2024, with Jane Street among the investors | Shows that the relationship predates the 2026 agreement |
A CoreWeave Securities and Exchange Commission filing says Jane Street had relied on the provider since 2024 for trading-model training. The filing says Jane Street selected CoreWeave after seeing its Hopper hardware come online quickly. Craig Falls, Jane Street’s head of quantitative research, said the firm needed the latest hardware quickly and trusted CoreWeave to deliver it faster than other providers.
CoreWeave’s case study describes direct connections, dedicated storage, upgrade options, managed deployment and scalable training and inference workloads. That combination is different from buying occasional on-demand GPU hours: it is a planned supply relationship intended to keep researchers productive as clusters expand.
Recommended Free Tools
Is Jane Street building its own AI data centers?
Jane Street is already reported to operate a Texas facility containing more than 4,000 Nvidia GPUs in 56 server racks, according to Reuters in 2026. Reuters also reported that the firm was seeking roughly ten times more compute capacity. Those are reported figures rather than audited company disclosures.
Also popular on Amazon
- Delivers 25 TFLOPS of AI inference performance using four MemryX MX3 accelerators in a compact M.2 2280 M-key module.
- Connects through PCI Express Gen 3 and above for straightforward integration into embedded and edge systems with an M.2 M-key slot.
- Includes 42 MB of distributed on-chip weight storage to support low-latency, batch-1 pipelined inference.
- Designed for low-power deployment with a 1 W–12 W operating range and host-managed airflow requirements.
- Supports ONNX, TensorFlow, Keras, and TFLite models through the MemryX software stack on Linux and Windows hosts.
Data Center Dynamics, citing Bloomberg, reported in June 2026 that Jane Street was discussing a possible 100–200 MW data center. The talks were described as early-stage; no final site or capacity had been selected. Ron Minsky, Jane Street’s co-head of technology, said, “You need to get the data centers built all over the place.”
A facility in that range would be a strategic-control project, not simply another leased cluster. Multi-megawatt sites can be designed around GPU density, cooling, power-delivery constraints, storage paths and low-latency networking. They also reduce dependence on whichever third-party capacity happens to be available when a research team is ready to train.
Why did Jane Street back an AI-chip startup?
Jane Street led Etched’s announcement of a $700 million financing round at a $21 billion valuation in 2026. Etched said Jane Street tested its inference hardware before leading the round and that its first rack had shipped to the firm. Jane Street described the early results as positive.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsInference chips are designed to run trained models efficiently, rather than perform every kind of workload needed during model development. Testing hardware in an actual trading-research environment lets Jane Street evaluate throughput, latency, software compatibility and operational reliability before making a larger commitment. The investment therefore has a potential operating rationale as well as a financial one, although public information does not establish the chip’s production scale or Jane Street’s eventual deployment volume.
Also popular on Amazon
How do other market makers approach AI infrastructure?
Reuters reported several different mixes in 2026. The firms are not following one template; they are choosing combinations of leased capacity, owned sites, hyperscaler services and startup investments.
| Firm | Reported approach |
|---|---|
| Hudson River Trading | Uses multiple data centers and signed a multiyear CoreWeave agreement. |
| XTX Markets | Built a €1 billion facility in Kajaani, Finland. |
| Citadel Securities | Uses hyperscalers and Google’s TPU infrastructure. |
| Jump Trading | Backed Positron and tested Etched hardware. |
| Susquehanna | Was seeking additional capacity in Texas. |
| Tower Research | Operates its own undisclosed facility. |
| Optiver | Has a multiyear data-center capital-spending plan. |
These strategies can be compared on four practical dimensions: whether capacity is leased or owned, which hardware is used, how much geographic redundancy is available, and whether spending is an operating contract, equity investment or construction capital.
How large is the wider AI-infrastructure build-out?
A 2026 Federal Reserve note describes data centers, processors, networking and cloud services as the physical and digital foundation of AI adoption. It estimated 4,049 data centers in the United States in 2024 and 5.8 gigawatts of additional data-center capacity that year.
The note cautions that aggregate hyperscaler capital expenditure includes non-AI projects. It also says leased capacity can make headline infrastructure figures understate the total investment supporting AI. That distinction matters for trading firms: a company can commit billions through a cloud contract without appearing to own the corresponding buildings or servers.
Also popular on Amazon
- Memory Size: 16 GB GDDR6 ECC.
- Memory Bus Width: 128-bit.
- Memory Bandwidth: 200 GB/s.
- CUDA Cores: 1280.
- Peak Single Precision floating point performance: 18 Tflops (GPU Boost Clocks).
What are the risks of the compute race?
- Power and permitting: A large site needs grid interconnection, land, cooling systems and permits; a reported target is not evidence that construction has begun.
- Hardware concentration: Dependence on one accelerator family or provider can expose a firm to delivery delays, software constraints or rapid product transitions.
- Utilization risk: A reserved cluster must stay busy enough to justify its cost, even when research priorities change.
- Chip execution: A promising inference processor still needs mature software, reliable manufacturing and support for the models a firm actually runs.
- Disclosure limits: Public estimates of GPU counts, annual AI spending and future capacity may come from company statements or news reports and should not be treated as audited totals.
What should observers watch next?
The clearest evidence of execution will be concrete: a selected site and utility interconnection for any large Jane Street facility, delivery of additional racks, documented expansion of the CoreWeave relationship, and measurable deployment of tested custom chips. Until those milestones appear, the announced contracts and investments demonstrate strategic intent and financial commitment, not a completed build-out.
Frequently Asked Questions
What does Jane Street’s reported 4,000-GPU figure actually cover?
Reuters described more than 4,000 Nvidia GPUs in 56 server racks at a Texas data center in 2026. It is a reported snapshot of that facility, not a verified count of all Jane Street hardware worldwide or a forecast of final capacity.
Do equity investments guarantee a trading firm priority access to chips?
No. An investment can align incentives and provide insight into a supplier’s expansion, but delivery still depends on manufacturing, power, networking and software readiness. The CoreWeave announcement couples the equity purchase with a separate cloud commitment.
Why might a market maker use both GPUs and custom inference chips?
GPUs remain flexible for varied training and inference workloads, while a specialized inference processor may improve efficiency for a narrower, well-understood model class. Running both lets researchers preserve flexibility while testing whether specialized hardware delivers a practical advantage.
Could this spending make it harder for smaller trading firms to compete?
It can raise the cost of obtaining scarce accelerators, power and engineering talent. Smaller firms may still compete by using public clouds, efficient models or specialized workloads, but they are less able to absorb long-term capacity commitments and construction risk.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
