CoreWeave saw muted price action following the latest earnings report; yet the soft price action is rare for the AI darling. The company went public in March and has stood out as the premier IPO among AI stocks given the stock is up over 200%. The lockup expired last month, which begs the question – can CoreWeave continue to defy the odds and overcome the typical insider selling that is seen around a lockup expiration? Investors typically fare better waiting for anxious insiders to sell, yet CoreWeave has been anything but ordinary.
In terms of timing, CoreWeave has hinted the second half of the year will be stronger. We look at why CoreWeave could end the year on a high note, yet to be prudent, we also look at why there was a negative reaction to the most recent earnings report. We end with a buy plan strategy the I/O Fund is eyeing for weighing the puts and takes on what promises to be a highly volatile, fast moving stock.
CoreWeave Rivals the Big 3 as the First “AI Hyperscaler”
CoreWeave brands itself as the world’s first “AI hyperscaler” as they offer both infrastructure and a software platform for developing large language models and deploying them. Being dubbed an AI infrastructure player means CoreWeave must offer a compelling value proposition to attract business from arguably the largest competitors in the world – AWS, Microsoft Azure and Google Cloud.
In the S-1 filing, the company points out it was built for AI workloads as opposed to the legacy cloud infrastructure-as-a-service providers that were primarily optimized for the cloud software era and e-commerce era. CoreWeave also asserts that outdated cloud infrastructure leads to lower utilization rates when you factor in usage.
One of their primary value propositions is offering bare metal servers, as the company does not need to offer shared GPUs like the hyperscalers. By stripping away the virtualization layer, raw performance goes up for R&D labs, who do not need workload flexibility. Although CoreWeave offers shared infrastructure in terms of storage and networking, one of the company’s key differentiations from the Big 3 is by offering dedicated bare-metal access.
The company also offers proprietary software to help achieve higher total system performance and more favorable uptime relative to competitors. According to the S-1 filing, “by delivering more compute cycles to AI workloads and thereby reducing the time required to train models, our capabilities can significantly accelerate the time to solution for customers […].”
CoreWeave Competes with Big 3 on Higher Usage Utilization Rates (MFUs)
To further understand CoreWeave’s competitive advantage, it’s important to discuss the model FLOPs utilization gap. The “MFU gap” is a metric that describes the gap between compute capacity and usage, which today often ranges between 30% and 40%. Cloud providers are often at 100% GPU utilization, yet there is a much lower utilization rate for GPUs when factoring in maximum floating-point operations per second (FLOPs). Initially, when MFU was coined by Google’s PaLM Paper, model training was running at 20% MFUs.
According to Google’s PaLM paper, they came up with the metric to better gauge a more realistic utilization rate: “Given these problems, we recognize that HFU (hardware FLOPs utilization) is not a consistent and meaningful metric for LLM training efficiency. We propose a new metric for efficiency that is implementation-independent and permits a cleaner comparison of system efficiency, called model FLOPs utilization (MFU).”
When factoring in FLOPs, the best possible (realistic) MFU is in the range of 50% to 60%, as this translates to raw compute being the bottleneck. Lower MFUs indicate inefficiencies, which CoreWeave specializes in solving. This could involve optimizing memory bandwidth, improving communication between GPUs, clearing data input bottlenecks, and other ways in which to fix batch size, enable faster data loading, and/or better ways to balance the compute.
Popular large language models do not publicly report their MFUs, but internally, this utilization rate is a dominant factor in competitiveness and time to market. R&D labs with a higher MFU rate have an important advantage as even an incremental increase in single digits to low double digits can result in a 25% to 50% increase in training speed and cost.
Due to going public, CoreWeave has published its MFU rate of 35% to 45%, stating it is 20% higher than competitors, which means other AI data centers have MFU rates more in the 30% range. Due to FLOPs performing an astronomical number of calculations, small percentages translate to an important advantage.
To put it simply, efficiency equals money and time in large-scale AI projects — training huge models can cost millions of dollars and weeks of time, so even a few percentage points of MFU improvement can translate to a significant advantage.
We covered this in more detail on our premium site in the analysis “CoreWeave: AI Infrastructure Built for the Next Decade.”
CoreWeave is Adding Capacity Hand over Fist Supported by OpenAI, Meta Agreements Totaling $36.6B
Before looking too granular at the financials, it’s important to note that CoreWeave is adding new capacity at a rapid clip. At a high level, the company is the most competitive outside of the Big 3 for new AI workloads, evidenced by the strong partnerships it’s securing with Meta and R&D labs like Open AI.
Last week, CoreWeave announced an expansion of its agreement with OpenAI, worth an additional $6.5 billion. The extension now takes CoreWeave’s total deal value with OpenAI up to $22.4 billion, building on to its initial $11.4 billion deal in March and the first $4 billion expansion in May. The $6.5 billion deal extends through May 2031, representing average annual revenue of more than $1.2 billion.
On Tuesday, CoreWeave signed a $14.2 billion deal with Meta, also lasting through 2031, marking one of its largest single deals to date. The deal represents average annual revenue of more than $2.3 billion.
This provides CoreWeave with other major revenue anchors and additional diversification away from its largest customer Microsoft (72% of revenue in 1H), as the deals could represent nearly 30% of CoreWeave’s current 2026 revenue estimate of $12.1 billion.
When discussing its customer concentration, CoreWeave made it crystal clear that there is no better customer to have at the moment: “And that when you have a company like OpenAI or an entity like OpenAI consuming compute, they’re just doing it at an order of magnitude that these other companies have not achieved yet.”
Looking beyond Open AI, CoreWeave has a contracted backlog of $30.1 billion, up $4 billion from Q1 and has doubled year-to-date, with expectations for 50% (or ~$15 billion) to convert to revenue over the next 24 months. The most recent deals with OpenAI and Meta bring this to over $50 billion in contracted backlog.
There are many pieces that must come together to deliver this backlog in the coming years including raising capital to build the infrastructure. However, keep in mind the company has a market cap of about $60 billion or roughly 1.3X the contracted backlog. Although I’m not suggesting a valuation be based off backlog, it’s certainly convincing the stock has room to run given the sheer size of the contracts it’s securing from the largest players in AI.
In early September, CoreWeave announced that key partner and investor Nvidia had entered a new order worth up to $6.3 billion under the duo’s pre-existing 2023 master services agreement.
With the new order, whenever CoreWeave’s compute capacity is not fully utilized, Nvidia will be obligated to purchase the unsold capacity. The deal extends through April 2032 and carries a total value up to $6.3 billion. While demand is currently strong and capacity is likely sold out for the near future, having Nvidia backstop future capacity helps alleviate concerns related to high customer concentration and de-risks its future growth story by providing some degree of guaranteed revenue. CoreWeave will disclose the entire MSA in its upcoming quarterly report.
Expanding Footprint with $6B Pennsylvania Data Center, UK Investment
In mid-July, CoreWeave announced a $6 billion data center project in Pennsylvania, with an initial capacity of 100MW with potential to expand to 300MW, though the company has not been upfront about delivery timelines for the initial phase or subsequent phases. This project is expected to be the cornerstone of CoreWeave’s vision of creating a mid-Atlantic hub from New York to Virginia.
Under its first phase, it likely will become one of CoreWeave’s largest data center facilities, considering its current active power footprint averages just over 14MW per data center. At full capacity, the facility will represent nearly 14% of CoreWeave’s current contracted power. On the capex side, CoreWeave has secured $4 billion in funding for the project, including a $200 million investment into grid infrastructure to support the facility. This will limit the amount the company will have to fund out of pocket, lessening its capex needs, which will already be quite elevated come Q4.
CoreWeave is also expanding its presence in the UK with a new ~$2.0 billion investment, working with Nvidia and data center operator DataVita in Scotland to deploy Blackwell Ultra GPUs. CoreWeave’s international presence is rather limited, with just five data centers across Europe and two in the UK, and investing to expand its international presence will help the company tap into growing demand across Europe.
Capex Weighted Heavily Towards Q4
One of the larger risks for investors coming into year-end is CoreWeave’s planned capex, with the company currently expecting the majority to land in Q4 due to timing of when infrastructure will go live. CoreWeave maintained its full-year capex guide of $20 to $23 billion, though in the first half, reported capex was only $4.8 billion (total PP&E increase of $5.7 billion minus $0.9 billion related to construction progress). To put this in perspective, this is 4-5x of the company’s guided revenue for the year.
Given that CoreWeave guided for just $2.9 to $3.4 billion in capex in Q3, or YTD spend of just $7.7 to $8.2 billion, Q4 capex is implied to be in the range of $12 to $15 billion, or nearly 7-9x estimated quarterly revenue of $1.8 billion.
The first caveat here is that CoreWeave does not have nearly enough cash to fund this capex entirely by itself. CoreWeave likely will have close to $5 billion in cash and equivalents after raising $2 billion in 9.25% senior notes and upsizing another 9.0% note raise to $1.75 billion, with additional access to a third delayed draw term loan facility for $2.6 billion to fund capex and GPU acquisitions for customer contractions. Combined, this still will not fully cover projected capex for Q3 and Q4.
The second caveat is that turning to the debt market will be costly, as CoreWeave has currently been pricing senior notes at or above a 9% rate, meaning raising $10 billion in fresh debt (such as what analysts from DA Davidson expect) at a similar rate could cost nearly $1 billion annually in interest payments.
CoreWeave Only Has 20% of Contracted Power Active
The reason why CoreWeave must spend aggressively on capex is directly tied to capacity, which is needed to convert its backlog to revenue and maintain hypergrowth. As of Q2, CoreWeave had ~470MW of active power across 33 data centers, or nearly 20% of its 2.2GW of contracted power, leaving an additional ~1.73GW to be developed. At its current scale, CoreWeave has energized >250,000 GPUs, suggesting that at 2.2GW, it can easily energize well over one million leading-edge GPUs.
Bringing the rest of its power footprint online will not only be expensive but necessary to support its growth story. Management stated they are aiming to nearly double active power by year-end to 900MW, hence the substantial increase in capex in 2H:
“We are aggressively expanding our footprint on the back of intensifying demand signals from our customers, ensuring that we maintain a durable multiyear runway for growth. We are now on track to deliver over 900 megawatts of active power before the end of the year.”
Core Scientific’s acquisition has provided an extra leg for growth in power: “Upon closing, CoreWeave would own approximately 1.3 gigawatts of gross power capacity across Core Scientific’s national data center footprint with an incremental 1 gigawatt or more available for future expansion. This scale enhances our flexibility to take on new projects and meet accelerated customer demand.”
Looking ahead, and assuming power and revenue remain correlated linearly, CoreWeave’s current contracted power footprint may be able to sustain a $25 billion revenue run rate, or enough capacity to take it to 2029’s current estimate.
Unleashing CoreWeave’s Monetization Path and Stock Buy Plan
CoreWeave’s suite of software is another area the company extends its value proposition beyond the Big 3 as the company reduces the need for specialized orchestration frameworks, engineering resources component failures and the need to constantly monitor for downtime.
Sign up below to continue reading:
- How CoreWeave plans to monetize every chatbot response, API call and application
- The I/O Fund’s buy plan and risk management strategy to help participate in this stock while protecting to the downside
- Details from the recent earnings report that help clarify why the stock sold off
- Important commentary from management that hints the end of the year will be strong for CoreWeave
According to Uvation, by focusing only on GPUs and software optimizations (detailed below), CoreWeave offers bare metal servers at a cost that is up to 20% to 50% cheaper than hyperscalers. This helps to explain why the company’s rapid and aggressive growth.
