There haven't been a ton of initial public offerings (IPOs) in recent years, let alone interesting, prominent companies with the potential to take on the top tech stocks in the market. However, one company right in the middle of the artificial intelligence (AI) revolution may be about to make its public debut. Not only is this stock growing at eye-popping levels, but it also counts Nvidia (NVDA -0.29%) as a big investor and Microsoft (MSFT -0.39%) as a huge customer.
Here's what you need to know about CoreWeave ahead of its IPO.
NASDAQ: NVDA
Key Data Points
A "Neocloud" backed by Nvidia
CoreWeave, which plans to list on the Nasdaq under the symbol CRWV, finds itself at the center of the AI revolution. However, the company didn't actually start out that way. In 2017, CoreWeave was founded by three executives at Hudson Ridge Asset Management, a natural-gas-focused hedge fund, with the original mission of mining cryptocurrencies.
The experience with cryptocurrencies honed CoreWeave's skills in deploying Nvidia graphics processing units (GPUs) -- which were used to mine crypto -- and managing energy-intensive computing clusters. Those skill sets also turned out to be incredibly important in AI computing.
In 2020, the company pivoted to build the CoreWeave Cloud Platform, and in April 2023, Nvidia invested in the company. Today, the AI chip giant owns just over 5% of the stock. In addition, Nvidia is a customer and likely uses CoreWeave to run its software offerings and perhaps test AI applications.
The Nvidia investment came at a very interesting time. As one may recall, May 2023 was the first blowout Nvidia earnings report, ushering in the hypergrowth stage of the AI buildout.
CoreWeave's competitive advantages
Some may wonder what exactly CoreWeave delivers that sets it apart from other major cloud infrastructure platforms. Reading through the Form S-1, it appears the company does a few things very well.
One key point CoreWeave emphasizes is that its clusters are built from the ground up as AI-optimized GPU clusters. This is in contrast to generalized cloud platforms that have to build both AI and traditional cloud infrastructures across their footprints.
CoreWeave believes it has a key advantage running these clusters through its proprietary orchestration and observability software, which enable more efficient utilization of its GPUs. Since AI GPU workloads are massive, complex, and computationally intense, it is difficult to orchestrate an entire data center for multiple clients efficiently.
To put some numbers to it, CoreWeave describes a metric called the model FLOPS (MFLOPS) utilization rate, which basically measures the utilization rate of an AI cluster relative to its total theoretical compute capacity. The industry average may surprise you. Due to the complexities of AI, CoreWeave says a typical MFLOPS rate is just 35% to 45% industrywide. That gap between actual and theoretical performance is the opportunity CoreWeave believes it can narrow, largely through some novel software innovations.
One such software innovation is SUNK, a software platform that combines Kubernetes and Slurm open-source software platforms. Kubernetes is a platform for containerized workloads in traditional cloud environments, which AI customers also use to serve their models. Meanwhile, Slurm is a popular open-source software that orchestrates massive parallel computing workloads for AI training.
Customers have traditionally had to choose between one or the other for each compute cluster. However, CoreWeave's SUNK platform allows Slurm to run inside Kubernetes, enabling developers to use the best of both. That boosts the efficiency of compute utilization.
A second software innovation is CoreWeave's Tensorizer, which is an optimization software for inference and training. For inference, Tensorizer can route a model in storage to the optimally close GPU node for the client. According to data cited in the S-1, this results in a faster load time than rivals HuggingFace and SafeTensors. For training, the Tensorizer can reduce training times through similar efficiency optimizations.
In addition to software, CoreWeave likely has a time-to-market advantage over other clouds. Because of Nvidia's investment, CoreWeave is likely at the front of the line, or very close to it, for the latest and greatest Nvidia GPUs. In its S-1, CoreWeave noted it was among the first to market with Nvidia H100 and H200 systems and was actually the first cloud to have Nvidia GB200 NVL72-based instances generally available more recently.
Image source: Getty Images.
Financials show hypergrowth
Of course, nothing speaks more to the positive aspects of CoreWeave than its financials. As one can see, the company has seen explosive growth over the past two years:
|
CoreWeave |
2022 |
2023 |
2024 |
|---|---|---|---|
|
Revenue |
$15.8 million |
$228.9 million | $1,915.4 million |
|
Operating income |
($22.9 million) |
($14.5 million) |
$324.4 million |
|
Operating margin |
(145%) |
(6%) |
17% |
Data source: CoreWeave S-1.
The big 2024 jump amounts to 737% growth, an incredibly high rate even for a so-called start-up. The flip to operating profitability is certainly promising as well. Of note, the past year's results were based on 32 deployed data centers hosting about 250,000 GPUs.
While it's unclear at which valuation CoreWeave decides to go public, certain analysts estimate the company will seek to raise $3.5 billion to $4 billion at a $32 billion market cap. So, the stock will trade at roughly 16 times trailing revenue and 100 times operating income. But before the IPO, CoreWeave has $7.9 billion in debt and $1.4 billion in cash, making it a bit more expensive on an enterprise value basis.
Risks may be high enough to keep some away
At first, CoreWeave may seem like the next big AI juggernaut. While its valuation is high on the surface, the stock valuation doesn't look so expensive, given the company's current growth rates and the long-term growth potential of generative AI.
However, the composition of that growth may raise questions. In 2024, 62% of CoreWeave's revenue came from just one company: Microsoft. Microsoft has been renting spare GPU capacity from CoreWeave to supplement its Azure cloud, despite spending tens of billions of dollars annually on its own cloud infrastructure as well.
One might ask why Microsoft is such a big customer when the other major clouds, Amazon and Alphabet, aren't listed as major CoreWeave customers. This could be because Amazon and Alphabet have fairly mature custom ASIC programs themselves. Alphabet designed its own Tensor Processing Unit chips in 2015, and Amazon unveiled its Inferentia chip in 2019 and its Trainium AI chip in 2021.
Microsoft was late to the custom AI chip game but unveiled its Maia AI chip in November 2023, just a little over a year ago. It's unclear whether the lack of a custom ASIC is the total reason for Microsoft's high use of CoreWeave. After all, Microsoft may appreciate CoreWeave's ability to run data centers for a different reason. But that could also be a significant part of it.
Therefore, if Microsoft ups its game and Maia matures to the level of Google TPUs or Amazon's AI chips, Microsoft may have less use for CoreWeave's infrastructure.
Make no mistake, Nvidia GPUs are still in demand today and will likely be in the future. However, since Alphabet and Google can supplement certain workloads with their own chips, Microsoft's scaling of its own chips may free up a lot of dollars to buy Nvidia chips directly for its own data centers.
Remember, cloud companies can buy custom ASICs at foundry prices, but Nvidia has gross margins in the mid-70% range. That means it's basically three to five times more expensive to buy an Nvidia GPU than to design one's own chip and buy it directly from a foundry.
In addition, thanks to Nvidia's investment, part of CoreWeave's appeal is likely early access to the most advanced Nvidia chips. Therefore, CoreWeave's destiny seems tied very tightly to Nvidia's going forward.
Of course, that's a great place to be right now. But should the AI buildout slow or something happen with Nvidia's competitive position, that would affect CoreWeave significantly, too.
CoreWeave is a fascinating company
In a volatile market, CoreWeave will likely prove to be a volatile and controversial stock when it becomes public. The filings show valid reasons for investors to buy into the IPO but also several big risks that will probably keep this investor on the sidelines, at least upon the IPO's unveiling.
