For a business that sells artificial intelligence infrastructure, CoreWeave (CRWV) has a unique challenge: Clients desire more processing power than it can presently provide.
As demand for Nvidia-powered AI infrastructure continues to surpass what the business can bring online, CEO Mike Intrator told Bloomberg that CoreWeave is still mostly sold out, with capacity committed for many years.
However, that lack of capacity is only one aspect of the issue. CoreWeave is also working to offer clients more than just graphics processing units, or GPUs.
This week, the business unveiled Forge, a software platform that expands the company beyond a neocloud by unifying AI development tools in a single setting. The push arrives as CoreWeave’s rapid expansion produces some striking numbers.
Second-quarter revenue reached $2.58 billion, more than double a year earlier. Revenue backlog, which includes contracted business expected to become revenue later, stood at roughly $104 billion.
Those numbers have an asterisk, though. Net interest expense reached $640 million during the quarter, reflecting the strain CoreWeave is under. Its computer infrastructure remains in high demand, but meeting that need requires billions of dollars for servers and data centers, and much of that money must be borrowed.
Forge creates the opportunity for an additional revenue stream, potentially curing that headache.
CoreWeave’s Nvidia capacity is sold out for years
CoreWeave continues to evaluate chips beyond Nvidia, Intrator told Bloomberg.
Because they are built to execute the many computations needed for artificial intelligence much more quickly than traditional processors, such chips are often referred to as accelerators.
However, Intrator said CoreWeave currently lacks enough infrastructure to deploy all the Nvidia capacity its customers want. He described the company as “overwhelmed” by demand and said it is largely sold out for several years.
Regarding the discussion of whether expenditure on AI infrastructure is starting to exceed real demand, the remarks provide investors with key insight.
According to CoreWeave’s second-quarter data, contractual demand remains significant. As of June 30, the business had a sales backlog of around $104 billion. More than $25 billion in additional client commitments signed early in the third quarter were not included in that amount, according to CoreWeave.
A revenue backlog does not guarantee that all of that money will turn into revenue right away. CoreWeave still has to fulfill the conditions of those client agreements and make the promised capacity accessible.
Infrastructure deployment is particularly crucial because of this.
Intrator said CoreWeave brought about 500 megawatts of active power into its infrastructure around the time of its previous earnings call.
Power is important, since purchasing GPUs does not, by itself, increase AI capability. To operate dense clusters of processors, operators also need data centers, networking hardware, cooling, and sufficient energy.
Additionally, Intrator refuted the notion that the planned limits on data centers will eliminate the need for computing.
“Moratoriums do not alter the demand for compute,” he said, arguing that such restrictions instead influence where the capacity gets built.
However, CoreWeave must pay a high price to bring all of this infrastructure online, increasing the significance of what the business offers beyond that infrastructure.
CoreWeave’s Forge targets the margin question
CoreWeave unveiled Forge at its Fully Connected conference in San Francisco on Sept. 30.
Forge combines several pieces of AI development in one environment, including training models, running them after they have been trained, measuring their performance, and improving subsequent versions.
That second step is commonly called inference. In simple terms, training teaches an AI model, while inference is what happens when people actually use the trained model to generate an answer or perform a task.
According to CoreWeave, MasterClass and Canva are already building on Forge. Intrator said the product grew out of customers effectively pulling CoreWeave “up the stack.”
The company said more than 4,500 customers, partners, developers, and AI industry participants attended Fully Connected. And clients who came to CoreWeave for computer infrastructure are increasingly requesting that the firm provide some of the software tools they use to create AI solutions.
Financially speaking, this is significant, since CoreWeave’s underlying infrastructure is costly.
In CoreWeave’s quarterly report to the Securities and Exchange Commission, the corporation said its total debt was $35.6 billion as of June 30. Additionally, CoreWeave spent $14.1 billion on property and equipment in the first half of 2026, up from $3.9 billion in the same time last year.
That expenditure includes the infrastructure and servers needed to handle AI workloads.
Forge offers another chance. “It’s going to be accretive to margins,” Intrator said.
Intrator did not quantify Forge’s potential contribution to revenue or margins. However, the remark addresses one of the main issues with CoreWeave’s business strategy.
To clear its massive backlog, the corporation can continue to grow its infrastructure. However, if it can use that infrastructure to offer software and development tools to clients, it may be able to make more money from those connections without needing to grow physical computer power for every dollar of revenue.
Although Forge is not a solution to CoreWeave’s funding needs because of these factors, it does clarify why investors may care about the software push.
CoreWeave wants to move beyond the neocloud label
CoreWeave is attempting to alter consumers’ perceptions of the business itself.
The business has sometimes been referred to as a “neocloud,” a phrase used in the industry to designate more recent cloud providers that are primarily focused on GPUs and AI workloads rather than conventional general-purpose computing.
CoreWeave was one of the first neocloud firms, according to Intrator. However, he said the business now sees itself as an “AI cloud.”
Although the distinction may seem semantic, CoreWeave’s most recent product strategy lends additional weight to the claim.
Weights & Biases, an AI development platform, was purchased by CoreWeave in 2025. This company provides tools for developers to monitor trials, assess models, and oversee AI research.
In June, CoreWeave introduced ARIA, an AI research and iteration agent that uses Weights & Biases technology. These features are now combined with CoreWeave’s training and inference offerings in Forge.
Instead of forcing every task to reside only on CoreWeave infrastructure, CoreWeave claims that Forge stays open across models, software frameworks, and other cloud providers.
This is significant, since it expands CoreWeave’s target market.
Once GPU capacity is supplied, the corporation no longer wants its connection with consumers to expire. It encourages developers to construct, test, assess, and refine AI models using CoreWeave’s software.
More AI:
- Nvidia just made a move Wall Street wasn’t ready for
- Microsoft just took sides in AI policy fight
- OpenAI just disclosed something genuinely alarming
Additionally, the tactic provides CoreWeave with an additional means of connecting with clients that may not have access to sizable engineering teams capable of overseeing their own AI infrastructure.
Intrator said it’s one of the audiences Forge is designed to address. He cited various CoreWeave clients, including well-known industrial firms such as Caterpillar and cutting-edge AI laboratories.
There is a chance to simplify complex AI infrastructure so that conventional organizations can use it without having to handle every technological layer on their own.
If the next wave of AI investing extends beyond the small number of tech firms and AI labs that were mostly responsible for the industry’s early infrastructure buildout, that might become more important.
According to Intrator, there is still a high demand for the underlying processing capacity. He disagreed with the notion that slower advancements in AI would inevitably lead to a significant drop in computing needs, arguing that businesses should continue purchasing cloud services, software, and infrastructure while developing AI solutions.
CoreWeave now has to turn demand into better economics
Customers’ need for AI processing power is easily demonstrated by CoreWeave.
Strong contractual demand is evident from its approximately $104 billion backlog, more than $25 billion in additional early-third-quarter commitments, and Intrator’s remarks about being sold out for years.
Providing it profitably is the more difficult part. Even though revenue more than quadrupled to $2.58 billion in the second quarter, CoreWeave recorded a $626 million net loss. $640 million was spent on net interest.
These numbers contribute to the explanation of why Forge merits consideration outside of its product attributes.
If CoreWeave wants to meet the clients who are already waiting for computation, it still has to fund servers, GPUs, electricity, and data center capacity.
A hypothetical second layer of economics around that infrastructure is provided by software.
Through the purchase of Weights & Biases, ARIA, and now Forge, the approach has been progressively evolving.
There are still many things investors may be wondering. Intrator did not estimate how much the platform may increase profits, and CoreWeave has not revealed a sales goal for Forge.
Additionally, the company’s debt burden and the cash needed to increase capacity are not eliminated by the new software venture.
However, the issue with CoreWeave is now more apparent.
Finding clients who are eager to purchase AI compute is not a problem at the moment. Instead, it is having trouble producing enough of it.
Therefore, the firm’s ability to leverage Forge and its wider software platform to increase the value of each client connection, while continuing to finance one of the industry’s biggest infrastructure expansions, may determine the next chapter in the CoreWeave tale.
Related: JPMorgan backs CoreWeave’s pricing play as a bear digs in
