#JaneStreet70亿押注CoreWeave A quant giant earning 20.5 billion USD a year, turning around to spend 7 billion on GPUs—what are they building?

On April 15th, a piece of news caused a stir in both the finance and AI circles: Jane Street signed a $7 billion deal with CoreWeave—$6 billion in compute leasing, plus $1 billion in direct equity investment. You might not be familiar with the name Jane Street. No worries, just remember one number: this company’s net trading revenue in 2024 is $20.5 billion, accounting for over 10% of the entire U.S. stock market trading volume. Goldman Sachs’ trading division doesn’t hold a candle to it. It’s not a hedge fund, but one of the world’s largest market makers—you might be trading stocks against it.

A trading firm spending $7 billion on GPU time. This move surpasses many AI labs. The $21 billion deal Meta signed with CoreWeave makes sense—Meta needs compute power to train LLaMA.

But what about Jane Street? What are they building?

Three big deals in three weeks, CoreWeave has become a weapons supplier in the compute power industry.

First, let’s clarify who CoreWeave is.

This company went public last year (NASDAQ: CRWV), and its core business is: buying大量 NVIDIA GPUs, building data centers, and renting them out as cloud services. You can think of it as a “GPU arms dealer.”

In the past three weeks, it signed three clients:

• April 9: Meta, $21 billion (until 2032)

• April 10: Anthropic, multi-year agreement (amount undisclosed)

• April 15: Jane Street, $6 billion in cloud services + $1 billion in equity

These three clients come from completely different industries: social media/AI research, AI model companies, and quantitative trading. GPU compute power is becoming a cross-industry infrastructure, like electricity—everyone needs it, but not everyone builds their own power plant.

A detail worth noting in Jane Street’s contract: they gained priority access to NVIDIA’s next-generation Vera Rubin computing platform. Vera Rubin was just announced at GTC in March, designed for reasoning, planning, and agent-level AI computing, with large-scale deployment expected by 2027. Jane Street is willing to start investing before this platform even hits mass production—showing that their demand for AI compute isn’t “future-oriented,” but “urgent—more is better now.”

The hidden story of quantitative trading: from mathematical formulas to large-model-based trading, the technological evolution is subtle but profound.

Early quantitative strategies (2000s) focused on statistical arbitrage and mean reversion, running on CPU clusters.

By the high-frequency trading era (2010s), latency was king—who’s faster makes money, with FPGA and microwave communication becoming standard.

In the 2020s, machine learning started to infiltrate: deep learning for prediction, NLP for extracting signals from news and social media, reinforcement learning for automatic strategy generation. But these were still “auxiliary tools.”

The real change has happened in the past two years: quantitative trading is shifting from “AI-assisted strategies” to “AI-native trading paradigms.” What does that mean? Previously, the process was: humans design strategies → optimize parameters with AI → execute. The new approach is: large models directly discover patterns from vast market data that humans could never conceive, then make autonomous decisions. It’s not AI helping you trade; it’s AI doing the trading. This paradigm requires compute power comparable to training a large language model. Jane Street’s $6 billion is probably aimed at this. An interesting detail: Jane Street’s traders don’t use C++, but a programming language called Mojo—designed to be “Pythonic in style but C-level in speed.” The company is preparing its toolchain for the AI era.

S&P’s warning: nobody wants to talk about systemic risks

Jane Street spending $7 billion on compute power also has a rarely discussed dark side. S&P Global Ratings released a report in April: U.S. banks’ total exposure to hedge funds and high-frequency trading firms has reached trillions of dollars, with leverage at an all-time high. The report used a heavy term—“endogenous fragility”—meaning the system is becoming internally unstable.

Jane Street’s 2024 revenue of $20.5 billion surpasses many large banks. When such giants entrust core trading logic to AI models, several risks are amplified:

First, the “black swan” of AI models. Traditional quantitative strategies are written by humans, so failures can be traced. Large models are black boxes, potentially producing unpredictable behaviors in extreme market conditions. When crude oil futures went negative in 2020, traditional models failed—what if AI was trading autonomously then?

Second, concentration of compute power. CoreWeave services Meta, Anthropic, Jane Street. If its data center experiences a failure—cyberattack, natural disaster, or technical glitch—the impact isn’t just on AI services but on global financial markets.

Third, “herding” resonance. If multiple quant firms use similar AI architectures (like Transformer-based models), they might make convergent decisions during market volatility. This isn’t just theory—people have already pointed out that quant funds’ herding behavior amplified the flash crash triggered by yen carry trades in August 2024.

Questions worth pondering:

First, $7 billion isn’t a big number for Jane Street. They earn $20.5 billion annually, and spending a third of that on compute power indicates they believe AI-native trading’s future revenue will far exceed this investment. This isn’t a gamble; it’s an all-in move.

Second, GPU compute power is becoming a financial infrastructure. Previously, quant firms competed on latency and algorithms; now, compute power is added to the mix. Without enough GPUs, quant firms are like internet companies without enough servers—they simply can’t get on the table.

Third, regulation is far behind reality. Last year, Jane Street was investigated by Indian securities regulators over derivatives trading—still a traditional strategy issue. When autonomous AI trading becomes mainstream, who will audit AI models executing billions of trades daily? The current regulatory framework isn’t prepared.

Retail and subjective investors face not a fairer market, but an increasingly unfair one. The advantages of AI quant trading aren’t just speed—its data scope, pattern discovery, and execution precision surpass human capacity. Jane Street’s $7 billion isn’t just for compute; it’s for market dominance.

This isn’t just Jane Street’s story. On the same day, Hong Kong AI quant platform AlphaNet completed a $10 million seed round. Bezos’s AI lab is poaching talent from Citadel. OpenAI just acquired a personal finance company. The intersection of AI and finance is becoming the most crowded track of this decade.

The difference is, Jane Street isn’t at the starting line anymore—it's already running.