Jane Street goes to Silicon Valley
Making bets on CoreWeave, Anthropic and the AI future
Hedge funds and trading firms investing in private companies is not a new phenomenon. Many large tech focused hedge funds like the Tiger Cubs invest around half their client assets in private tech companies, often start-ups.
Hedge funds have also been gotten into the digital assets wave. Marshall Wace’s XO Digital Fund made profits of hundreds of millions of dollars from participating in a private fund-raising round for the stablecoin issuer Circle. Brevan Howard’s multi-billion-dollar digital assets fund has been one of the most active at investing in stakes in crypto infrastructure firms.
Some of the largest investments have come from proprietary trading firms. DRW and to a lesser extent Jump Trading are the forefront of digital assets investing and market structure.
Citadel Securities is a firm that has traditionally shied away from this space despite having taken on crypto VC Paradigm as a shareholder in 2022. But late last year it led a $500m fund raising round valuing the digital assets infrastructure firm Ripple at $40bn. It also invested $200m in crypto exchange Kraken (and committed to a strategic partnership) as part of a $800m raise at a $20bn valuation.
The mighty Jane Street is also deep into both financial market structure and crypto trading. It was also part of the Kraken fund-raising round. But what has fascinated me in recent years has been Jane Street’s private investments further afield in technology and AI. Many of these investments are into businesses where Jane Street is an existing customer, but they are not your bread-and-butter trading infrastructure businesses.
In today’s piece, I dig into these investments in more detail.
Jane Street flexes its balance sheet
There is perhaps no firm in the proprietary trading space that has grown their capital base in recent years as much as Jane Street. The firm likely ended 2025 with an equity base of $45bn (versus $29bn at the end of 2024) and debt of $11.2bn giving a trading capital base of more than $55bn.
I wrote here on this Substack early last year about Jane Street’s enormous growth over the last decade…
I also wrote about the business model differences between Jane Street and Citadel Securities… https://www.ifre.com/people-and-markets/2308284/trading-titans-diverge-as-jane-streets-prop-push-pays-off
Bloomberg recently reported that around 7% of Jane Street’s trading capital was invested in private companies or funds. Looking at Jane Street’s April 2025 bond prospectus gives no details at all about this burgeoning investment portfolio, especially how the recently publicly listed CoreWeave is treated.
According to Bloomberg, 12% of Jane Street’s Q3 2025 net trading revenue of $6.83bn was from well not trading... “The market-making firm told investors that its investments in private companies and funds lifted its third-quarter trading revenue by about $830 million, said one of the people, who asked not to be identified citing private details. Anthropic PBC, which has seen its valuation surge this year, was responsible for a vast majority of those gains, the people said.”
Looking at Jane Street’s quarterly revenue trajectory, and not just the overall growth but the huge volatility in it (even relative to peers), a plausible explanation is that this is caused not only by large proprietary trading bets, but also these other venture capital type investments.
Playing the AI boom
Jane Street’s investment in Anthropic was well known. But what was a little surprising is that this was being booked in trading revenues and not elsewhere.
It started in March 2024 with a $100m investment as part of the FTX bankruptcy. In a late 2023/early 2024 fund raising round Anthropic had been valued at around $18bn. CNBC also reported at the time that “Jane Street’s head of quantitative research, Craig Falls, has also proposed to personally buy around $20 million worth of shares.” as part of the stake sale. FTX founder Sam Bankman-Fried is of course the most famous of Jane Street’s alumni.
Jane Street subsequently participated in a Series E round at a $61.5bn valuation in Q1 2025: https://www.anthropic.com/news/anthropic-raises-series-e-at-usd61-5b-post-money-valuation and a Series F round at a 183bn valuation in Q3 2025: https://www.anthropic.com/news/anthropic-raises-series-f-at-usd183b-post-money-valuation.
By November 2025 that valuation had risen again to $350bn: https://www.cnbc.com/2025/11/18/anthropic-ai-azure-microsoft-nvidia.html
We don’t know exactly what percentage of Anthropic that Jane Street owns currently and did the valuation increase between the Series E and Series F all get booked in Q3 2025 trading revenue or was there a small contribution in the second quarter. But it is likely that Jane Street’s trading revenue in Q1 2025 benefited by the tune of a few hundred million dollars from the value uplift between the Series D and Series E rounds.
By the same logic Jane Street’s trading revenue in the final quarter of 2025 could benefit from around a billion dollars owing to the huge value uplift seen for Anthropic in its November valuation round.
With a potential Anthropic IPO in 2026, there could be further volatility in this valuation running through Jane Street’s 2026 P&L.
The Anthropic investment is fascinating as it is far from Jane Street’s core competency in trading market infrastructure and the new disclosure highlights the challenges of analyzing the P&L of private companies, which do not have the same disclosure requirements as publicly listed ones.
Jane Street gets into the GPU infrastructure game
The other very large investment that Jane Street has made - beyond Anthropic – is AI data centre provider CoreWeave. Jane Street is a leader in data, quantitative research, and using machine learning within its processes. The firm is also well known as one of the largest buyers of GPUs on Wall Street. It mentions these on its website: https://www.janestreet.com/join-jane-street/machine-learning/
Given this and the requirement for very large and variable usage it makes complete sense for Jane Street to rely on a firm like CoreWeave that provides GPU compute. This is what Jane Street says about CoreWeave:
CoreWeave was a private investment and then became a public company. It is unclear where the gains or losses on this are booked. In many ways it seems odd to capture this in net trading revenue, but then again it isn’t any different to a change in the valuation of Anthropic. In fact, it is arguably a more transparent valuation than that of Anthropic.
Publicly listed banks have seen gains in their reported revenues from the revaluation of stakes in the past. For instance, in 2024 JP Morgan’s revenues were bolstered by a $7.9bn net gain from the revaluation of its Visa shares. Smaller gains of $0.5-1bn were reported by the likes of Bank of America, Citigroup and PNC Financial Services for their Visa stakes. But being publicly listed, these companies flag such revaluations clearly.
In December 2023 Jane Street was a major participant in a funding raising round that valued CoreWeave at $7bn Fidelity and Jane Street Back CoreWeave at $7 Billion Valuation - Bloomberg. In November 2024 Jane Street led a $650m secondary fund-raising round in CoreWeave, that valued the firm at a $23bn valuation.
Jane Street’ trading revenues rose from $4-4.5bn per quarter in the first half of 2024 to almost $6bn per quarter in the second half of 2024. Did CoreWeave add $0.5-1bn of gains to second half revenue for Jane Street?
CoreWeave had its IPO in late March 2025, and the share price trebled in the second quarter of 2025. The increase in the valuation of Jane Street’s circa 5-7% CoreWeave stake was around $2.5-3.5bn in Q2 2025.
In the second half of last year, CoreWeave stock started to decline by 16% in Q3 2025 and 39% in Q4 2025. 5-7% of CoreWeave’s market capitalization decline equated to a loss of around $500m in Q3 2025 and $1.5-2bn in Q4 2025.
New bets
CoreWeave and Anthropic are by far the most significant venture capital type investments that Jane Street has made. But in recent quarters it has become much more active in this space. When you are throwing off the kind of cash flow that Jane Street’s trading franchise is and are seeing those two large investments move positively, it is hardly surprising.
Much of these other investments also have an AI flavour to them. The highest profile of these are Thinking Machines led by former OpenAI CTO Mira Murati. In mid 2025, Jane Street was part of a $2bn fund raise valuing the firm - which still had an undecided business plan - at $12bn.
In late 2025 Jane Street and their strategic partner CoreWeave led a seed round investment in AI software developer Numerata. Jane Street and CoreWeave Announce Seed Investment in Numerata | CoreWeave
Reflecting on its VC type tech investments Jane Street said at the time “Jane Street’s partnership with Numerata underlies our expanding commitment to back transformative technology companies as both a strategic partner and investor throughout their whole lifecycle, from seed stage to public markets...Our approach combines a technical-first diligence with patient, flexible capital, backed by deep infrastructure expertise from building and operating global technology systems.”
At around the same time, Jane Street was leading two other tech scale-up funding rounds. It anchored a $700m raise by Amazon backed X-energy, a firm developing a fleet of small modular nuclear reactors. There was also a $105m raise by startup Antithesis, an AI coding testing and security firm. Antithesis co-founder Will Wilson told the FT “It’s not a coincidence that Jane Street is a massive user of AI coding.”
Is Jane Street a two-headed beast: part trading firm and part AI play?
The investments I have outlined in this piece are to date a rounding error in comparison to the huge scale and profitability of Jane Street’s rockstar trading franchise.
But could they and similar investments made by Jane Street be more important in the future?
When Citadel Securities sold a stake to Sequoia Capital and Paradigm in early 2022 at a valuation of $22bn, the implied multiple was around 3x revenues or 7.5x net income. The market wasn’t hugely surprised. After all, this is volatile market making revenues that can go up or down depending on the level of market volatility and macro conditions.
Recently the hedge fund Millennium sold a stake, valuing itself at $14bn. Because of its pass-through model pushing huge amounts of cost (sometimes around 10% of AUM) through its investors, it didn’t have a classic recurring fee model. It did introduce a management fee but only for those rare years when returns are very low. Given the firm’s consistent returns, performance fees are much more recurring and less volatile than for other hedge funds hence the implied valuation multiple of 8-10x post-tax performance fees didn’t seem crazy.
Neither Citadel Securities nor Millennium need the cash, and the transactions were more likely for strategic reasons, but the valuations came to my mind again when I saw the lofty prices being paid for stakes in digital assets market infrastructure and even more so AI.
Jane Street’s deployment of significant sums of capital in AI and crypto and other related fields is a fascinating development. So far, their track record is fantastic, and they have the patient capital and technical know-how (often as a client) to generate significant returns on their capital here.
On an upside scenario it could mean that the franchise value and IP of firms like Jane Street is worth more than just a simple multiple of revenues or earnings associated with a trading or market making business model. To be clear given its diversification and quality, I would argue that even the trading franchise of Jane Street is worth more than the multiples I outlined above.
But could AI provide a further embedded option value. It’s still too early for this today but given what they have achieved over the last 25 years I wouldn’t bet against Jane Street!
On the flip side, the idea that the blockbuster 2025 revenue and earnings growth at Jane Street could be to some degree owing to one-off lumpy stakes in AI and tech companies was largely unknown and suggests that perhaps their true underlying trading performance is not as off-the-charts as we had all assumed. The investment banks and likes of Citadel Securities can take a sigh of relief.
There is also the not so trivial matter of whether expectations and valuations in AI have got ahead of themselves.
Really strong breakdown of the accounting grey zone here. The part about booking Anthropic revaluations directy into trading revenue is wild - I remember when our team was analyzing prop trading P&L volatility a few years back and we kept hitting these weird spikes that didnt match market activity. Turns out private investments were getting blended in similarly. Its fascinating that Jane Street is transparent enough to disclose the 12% contribution, but also shows how murky comparisons across firms can get when everyone's got their own treatment for illiquid stakes.
I believe CoreWeave has a high probability of filing for Bankruptcy in 2026. The business is in far worse shape today than the public knows. Here are 5 key issues it is facing that cumulatively will lead to its insolvency next year.
1. The Cashflow Problem
A major problem CoreWeave has is its cashflow. CoreWeave sold its future to fund its growth. Customers have already pre-paid, and the money they paid is already spent. For Example, Open AI has prepaid over $2 Billion to CoreWeave for credits that it will slowly burn through over 2 years. While CoreWeave will recognize that $2 billion as revenue they will get no cash from Open AI as Open AI will just be burning through credits. The $5.3 Billion in deferred revenue on CoreWeaves balance sheet will go down, and revenue will be recognized, but they will get no cash until those pre-paid credits are used up. This leaves no cash for CoreWeave to pay its operating expenses, or pay its amortized DDTL Loans, much less continue the $28-30 Billion in Capex they have forecasted. They are almost certainly going to have to lower the Capex and renegotiate the contracts they have with customers to survive this. While reported revenue numbers will look good the company will be in great financial stress.
2. The Credit Problem
They already have junk status with all the major rating agencies, and CDS spreads are projecting a 42% chance of default in the next 5 years (I think it is closer to 45% this year). They have $3.7 billion in current debts due 12 months from September 30th, 2025 (per the most recent 10q). Based on table 10. “Debt” in the 10q about $2.6 billion of that is amortized payments on its DDLT loans and the other is a large balloon payment of OEM Financing Arrangements from Nvidia and Dell. They have the maturities of these financing arrangements listed as due March 2026-August 2028 but given the amount of current debt they show this is extremely misleading. Almost 100% of the OEM financing debt is due in the first half of 2026. Keep in mind that as of September 30th the DDTL loans are not fully drawn, and that amortization payments on those loans start Q1 of 2026. When those loans are fully drawn current debt will rise dramatically.
Having said all of that, they do not have near enough cash to cover these payments along with operating expenses, which is why they had to have their minimum liquidity levels lowered to $100 million to avoid technical default. Furthermore, they have cross collateralized the DDTL loans with the entire company. This will likely prevent them from being able to issue more bonds or convertible bonds. This cross collateralization gives the DDTL lenders first position on chips and either first position or equal position to bondholders on the rest of the company.
This is why they need unlimited equity cures to get through the next few months. They need to be able to print stock to not default on DDLT Loans. This is all noted in the 8k reported in early January discussing the modification of CoreWeave’s DDTL loans.
3. The Equity Problem
I can send you a PDF that goes into greater detail on this, but basically the founders are converting founders shares into sellable class b shares at a clip of about 9 million shares a quarter. On top of this Magnetar, a big early investor, is selling about 27 million shares a quarter, and Magnetar along with the 3 founders already have presold shares in a VPF contract that will hit the market on June 19, 2026. In addition to this they have a 6% evergreen provision that automatically adds 6% of the total share count on Jan 1 that goes to help employees. (5% Equity and 1% Employee Stock Purchase Plans) Another 29.8 million shares. They will also need to print at least 62 million shares (assuming the stock prices stays around $78) just to survive the year. In reality they probably need much more but let’s be conservative. That means with insider selling and new shares issued, the new supply of shares will be close to 215 tradable shares. Just newly printed shares will be around 91 million. With current outstanding shares at 497 million existing shareholders will be diluted by at least 18% and the tradable share float will go up 43%. Not great for existing shareholders.
4. The Contract Execution and Construction Timing Problem
The fourth problem they have is their inability to meet deadlines and deliver on contracts. They have strict contract realization deadlines that they have to meet to secure DDTL Financing. They have to show billings by February 28, 2026, to be able to collect the remaining money on the DDTL Loans. The big one is the DDTL 3.0 loan for the Denton Open AI Facility. They cannot actually get the money from the DDTL Loan to pay suppliers until they show revenue, but they are 60-90 days behind schedule. They have cited “Weather Delay” which I believe to be totally bogus. The real reason for the delay is the redesign of the facility to meet the liquid cooling requirements of the new Blackwell chips.
Having said that, the Shell for the Denton facility was delivered to CoreWeave per The Texas Department of Licensing and Regulation permit issued on December 29th, 2025. (See the link to the permit below.) This was the final step for Core Scientific, the landlord to deliver the shell and begin collecting lease payments. It will take 5-6 months potentially to build out the facility, and to get all the chips up and going. This means that they will not be able to bill Open AI for at least 5 months. If they cannot get the facility up and running by March they will have to start paying Open AI penalty payments in the form of compute credits, if they do not have it up and running by May 1, 2026 Open AI can cancel the Denton Contract, and if it is delayed until July 1, 2026 they can terminate all deals with Coreweave. I do not believe Open AI will cancel the deal because they have already pre-paid for so much compute at such low rates. The penalties could add up to a significant number though for Coreweave further pushing out its cashflow.
The timing is a big problem for CoreWeave’s cashflow. This is where the unlimited equity cures come in. They cannot bill open AI because they are not finished installing the chips, and they cannot get the money from the DDTL loans to pay their suppliers until they bill Open AI. So, what the banks have allow them to do is substitute these “Billings” with stock sales to the open market to generate cash that will give CoreWeave liquidity to pay the banks. This is not an option. They have to do this to get their money even if they have cash in the bank to pay the loan. If they do not issue equity cures to replace billings the bank will not fund the remaining portion of the DDTL Loans. The Denton facilities billings alone are estimated to be around $100 Million /month conservatively. That means at a minimum $100 million in stock that must be issued per month of delays just to satisfy the banks and get funding. This is just for the open AI Denton, TX project. CoreWeave has 3 of these going at the same time in similar stages. Their sites at Ellendale, ND and Lancaster, PA are also experiencing similar delays.
In my opinion construction delays and contract timing risk are potentially the biggest of all the issues CoreWeave has in the short term. They could blow everything up very quickly.
5. Their Margins Suck
For this I lean on others’ research. The Kerrisdale Capital Report posted on September 15, 2025, lays everything out pretty well. Micheal Burry also has a similar view that the depreciation schedule companies are using for these chips is far to long. Basically, what they say is that reported margins are roughly 20% based on a 6 year chip life cycle, but margins are essentially 0% because the chips only actually last 4 years. Also, CoreWeave will not realize any free cashflow from its chips for 3 years due to harsh amortization schedules and high interest rates on its DDTL loans. By the time they get the DDLT loans paid off they will have made 0 money on their chips, and they will basically be worthless.
Kerrisdale Report
https://www.kerrisdalecap.com/wp-content/uploads/2025/09/Kerrisdale-CoreWeave.pdf
Employee Benefit Plan Filing
https://www.sec.gov/Archives/edgar/data/1769628/000119312525058309/d899798dex103.htm
The Texas Department of Licensing and Regulation Permit
https://www.tdlr.texas.gov/TABS/Search/Project/TABS2025005706