CoreWeave picked a bad time to IPO

March 31, 2025

Good morning. You know that $40 billion funding round that OpenAI’s been talking about? It comes with a $20 billion catch. Softbank has said that the investment could shrink by half if OpenAI doesn’t complete its conversion to a for-profit by year-end.

Interesting.

— Ian Krietzberg, Editor-in-Chief, The Deep View

In today’s newsletter:

🩺 AI for Good: Treating preemie babies
💰 Musk sells Twitter to himself
📊 CoreWeave picked a bad time to IPO

Source: Unsplash

In a new study, researchers at Stanford Medicine designed a deep learning algorithm — specifically, a variational neural network — to improve intravenous nutrition and treatment for premature babies.

The details: Trained on data from nearly 80,000 intravenous preemie prescriptions, which included information about the results of those prescriptions, the algorithm was designed to process information in a given patient’s electronic health record to predict the exact formula of nutrients needed.

The idea is that this approach would reduce medical errors while making it easier to treat the little patients, specifically in places lacking adequate resources.
Right now, these nutritional prescriptions — the result of collaborations between neonatologists, dieticians, pharmacists and nurses — are made “from scratch” for each person, every day, Dr. Nima Aghaeepour, the study’s senior author and a professor of anesthesiology, said in a statement. “Total parenteral nutrition is the single largest source of medical error in neonatal intensive care units, both in the United States and globally.”

The researchers used their algorithm to concoct 15 standard formulas, with the idea being that the algorithm would recommend a specific formula for a specific amount of time.

In a test, 10 neonatologists were shown clinical information and presented with multiple possible prescriptions; they “consistently preferred the AI-generated prescriptions to the real prescriptions,” according to the study.

Why it matters: Though far from being implemented — the researchers would next need to conduct a clinical trial — the system, if it ever does make its way to hospitals, would “make doctors better and make top-notch care more accessible,” according to one of the researchers.

Meet the best and brightest in the AI/LLM communities with Turing and Fireworks AI at our evening reception on April 10 in Las Vegas during Google Cloud Next.

Register today to:

Connect with AI/LLM leaders.
Network with Google Cloud Next attendees.
Enjoy fantastic food and creative cocktails.

Secure your spot now.

Source: Unsplash

Elon Musk’s AI company, xAI, has acquired X (which used to be Twitter) in an all-stock transaction that has valued xAI at $80 billion and X at $33 billion.

The details: The move doesn’t exactly come as a surprise. Musk — who owns both companies — has been gradually bringing them closer together (in more than just frustratingly similar names) for a while, now.

xAI’s generative AI model, Grok, is trained on data gathered from X users; X, likewise, serves as xAI’s sole means of product distribution, since access to the chatbot requires a pro subscription to X. (X trains on user data by default … here’s how to turn it off).

“xAI and X’s futures are intertwined,” Musk said, adding that “data, models, compute, distribution and talent” will now be officially combined.
The two companies share more than just data — they also share investors: Fidelity Management, Andreessen Horowitz, Sequoia Capital and Saudi Arabia’s Kingdom Holding Co. are all invested in both companies.

The Wall Street Journal reported that the valuation of each company was determined following negotiations between the two firms, which, beyond being both owned by Musk, were advised by the same people.

The details of the transaction, here, are unclear. Since both firms are privately held, the likely transaction involves a stock swap, according to CNBC, with X investors getting paid out with shares of xAI.

xAI was last valued at $50 billion in a $5 billion funding round, though any details about its revenue, expenses and roadmap to profitability remain quite unclear.

It’s not the first time Musk has used one of his companies to buy another one of his companies; in 2016, Tesla acquired SolarCity, a Musk-led solar startup, for $2.6 billion.

This acquisition, though, comes at a time when Tesla shares are in the midst of a major retreat. The stock, down 30% for the year, closed Friday’s session down 3.5%. After the market closed, Musk announced the X acquisition, and the stock fell another 1.3% in after-hours trading Friday evening.

Focus on what it looks like live, we’ll do the rest.

Publish your design as a live site in seconds. Hit publish, and your site is instantly live worldwide with Framer's state-of-the-art hosting.

Need to upgrade something? Just as fast.

Global investments: The European Commission said Friday that it plans to invest $1.4 billion in advancing AI, cybersecurity and digital skills between 2025 and 2027, a bid for up-skilling that coincides with a massive investment push by the Comission into AI infrastructure.
Transitions: Reuters reported that, if OpenAI wants the full $40 billion investment from Softbank, it must transition into a for-profit corporation by year-end. This is on top of similar terms from OpenAI’s previous $6 billion fundraise, and comes despite Elon Musk’s persistent legal action to prevent the maneuver.

23andMe bankruptcy: With America’s DNA put on sale, market panic gets a new twist (CNBC).
Scale AI is seeking a valuation as high as $25 billion (BI).
Turkey’s brain drain is taking its female coders with it (Rest of World).
We Mapped DOGE’s Silicon Valley and Corporate Connections (Wired).
I used face recognition app to hunt man behind whisky fraud (BBC).

Source: Unsplash

When I think about the market’s performance in the first quarter of this year, one word comes to mind: volatility.

The introduction of on-again, off-again tariffs has injected investors with a level of fear and caution over the state of the U.S. economy that, crossed with high-valuations, enormous investments and little near-term return on investment, has led to plenty of bleeding for all the AI-related names (I’m looking at you, The Magnificent Seven).

As of Friday evening, shares of Nvidia are down more than 20% for the year so far, Amazon is down 13%, Microsoft and Apple are both down 10%, Tesla is down 30% and Meta is down 4%.

Year to date, the tech-heavy Nasdaq is down around 10% and the S&P 500 is off by around 5%.

A lot of red.

Entering into all this volatility is CoreWeave, a company that sells access to AI cloud computing based on its operation of GPU-laden data centers.

On Friday, CoreWeave $CRWV ( ▼ 3.25% ) — in the biggest tech IPO since 2021 — debuted as a publicly traded company on the Nasdaq. It priced its IPO at $40 per share, a major decline from its initially proposed range of $47 - $55 per share.
According to CNBC, Nvidia anchored the IPO with a $250 million order.

“There’s a lot of headwinds in the macro,” CEO Mike Intrator told CNBC. “And we definitely had to scale or rightsize the transaction for where the buying interest was.”

On the day, the S&P 500, Nasdaq and Dow Jones Industrial Average were all off by around 2%, a retreat at least partly inspired by core inflation numbers for February that came in higher than anticipated.

CoreWeave itself didn’t fare too badly by the end of the day; the stock reached a high of nearly $42 and a low of $37 before closing the day right where it started, at $40 per share with a market cap just shy of $19 billion — a decidedly muted debut.

The IPO is being looked at as a barometer for the wider AI investment landscape, since it is a pure AI play.

Let’s take a look at CoreWeave’s business. When a company goes public, they are required by law to publish an S-1 document, a document that basically airs all the nasty details about the internal workings of the business, so investors can swim at their own risk.

CoreWeave’s S-1 is more than 2,000 pages long.

But a few things stick out: Let’s start with revenue and work our way down. CoreWeave has reported revenue of $16 million, $229 million and $1.9 billion for 2022, 2023 and 2024, respectively. So, revenue is growing. That’s good.

But that growth came at a very, very steep cost. That’s less good.

CoreWeave’s net loss for each of those years was $31 million, $594 million and $863 million. So, over the past three years, CoreWeave has raked in roughly $2.15 billion in topline revenue, and still lost $1.5 billion.
The “vast majority” of that revenue was derived from multi-year “committed” contracts, where a customer purchases access to CoreWeave’s platform over a given time period.

And that brings me to the next problem: the customer.

According to CoreWeave, we’re still in the “early stages” of the “AI revolution,” which means they don’t have too many customers yet. In 2022, 41% of CoreWeave’s revenue came from its top three customers. In 2023, that number was 73%.

And in 2024, 77% of CoreWeave’s revenue came from just its top two customers.

CoreWeave’s single largest customer (Microsoft) represented 62% of its revenue in 2024. The company highlighted the risk of this “limited” customer base in the “risk factors” section of its report.

On top of all this (yes, there’s more), CoreWeave is dealing with “substantial” debt. By the end of last year, the company had a total of $8 billion in debt, a massive number, especially considering its comparatively expensive revenue numbers and “history of generating net losses” due to the enormous infrastructure costs required to grow its business.

The company warned that it may “still incur substantially more indebtedness in the future,” adding that it anticipates “increases in our operating expenses in the future,” increases that may prevent CoreWeave from achieving or sustaining profitability.
Running through all these factors is an affirmation that CoreWeave has “identified material weaknesses in our internal control over financial reporting,” which affects its ability to produce “timely and accurate” financial statements.

And, the kicker — CoreWeave’s entire business is based on an assumption that AI will, as it mentions, go through a significant revolution. But, as it also mentions in the document, “the broader adoption, use and commercialization of AI technology, and the continued rapid pace of developments in the AI field, are inherently uncertain.”

CoreWeave has picked an interesting business at a tough time, at least when it comes to the public market.

Right now, there’s a race to build out computing infrastructure that is resulting in a massive boom in data center investment. Since it takes time for the hyperscalers that CoreWeave works with, like Nvidia, Microsoft and IBM, to get new data centers up and running, CoreWeave is a great auxiliary option.

But all the hyperscalers are actively investing in building out their own data centers.

At the same time as this buildout is happening, the chairman of Alibaba very recently warned of an impending data center bubble, since “people are investing ahead of the demand that they’re seeing today … they are projecting much bigger demand.”

I am reminded of what happened in 2022, when Amazon wound up with way too many warehouses on its hands, an over-buildout it conducted in response to the wildly high — and, it turns out, fleeting — surge in demand it experienced during the pandemic.

GPU demand and data center demand are very high right now. But the hyperscalers have already indicated that their investment in data centers is unlikely to persist at the same scale it’s been at for the past couple of years; now, they have the hard job of managing capacity enough to respond to demand without overdoing it.

In that, CoreWeave might be a good option, since working with CoreWeave keeps the developers from having to conduct that complex calculus. But as CoreWeave sinks deeper into debt in order to meet demand when there is no guarantee that demand will continue, I am reminded again of that word, that pesky, pervasive word.

Volatility.