Product1 publisher2 min readPublished
CoreWeave's current customers all plan to expand their use, a 13-buyer study finds
All seven CoreWeave customers running workloads in a 13-buyer SiliconANGLE study expect to grow their spending or workloads. Three of the ten buyers past evaluation are not running on it, and hyperscalers still hold the rest of the application stack.
The Product Desk · Product desk
What happened
- SiliconANGLE and research firm Qualitate ran 13 in-depth interviews, totalling more than seven hours, with CoreWeave customers and prospects who had evaluated the platform.
- The panel was sourced independently of CoreWeave and includes evaluators and buyers who looked at the platform and passed.
- Seven respondents report significant current CoreWeave use, three are still evaluating, one finished a successful pilot without rolling out, and two chose not to proceed.
- All seven current users describe plans to expand, some into a higher spending band and others within their band or by adding production usage.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- decision Evaluating CoreWeave turns into a per-workload decision, since the hyperscaler running the rest of the application stays in the architecture even where CoreWeave wins the GPU work.
- constraint A passed pilot settles less than it appears to: with three of ten buyers past evaluation off the platform, the rollout case has to be judged separately from the pilot result.
- exposure CoreWeave's answer to investor worry about concentration in AI labs and hyperscalers now rests partly on enterprise accounts moving from pilot to production and growing from small starts.
One financial-services account in the panel is running some production load on CoreWeave while its pilot and evaluation are still under way [10]. Someone at that firm is answering for live traffic on a vendor the company has not finished choosing.
Here's what teams tell themselves about a neocloud: the GPUs were the reason to buy. Here's what the buyers in this study describe, according to SiliconANGLE and its research partner Qualitate. GPU scarcity opens the door, and performance, cost and the operating experience give them reasons to stay [3]. The GPU demand covers both workload types. Inference is growing on a steep curve, SiliconANGLE says, and training is still growing alongside it [4].
Count only the respondents who are past evaluation, and seven of ten run workloads on CoreWeave [1]. The two who declined went to Azure and to on-prem infrastructure [11]. SiliconANGLE writes that hyperscalers remain deeply embedded in the application estate and that not every successful proof of concept turns into a signed customer [5].
The growth plans are intentions. They come from senior people with budget approval or final say over the purchase [7]. The study does not report renewal rates or how deep usage runs inside each account. Accounts also start small: the current-use group's reported spending begins below $100,000 [13].
SiliconANGLE checked its interviews against CoreWeave's financial disclosures and public statements by Chief Executive Michael Intrator and Chief Financial Officer Nitin Agrawal [2]. Intrator's statement, as SiliconANGLE describes it, answers investor concern about high customer concentration among AI labs and hyperscalers [14]. Nine respondents here work at large enterprises, in the US, Canada, the Netherlands, the UK and Germany [6]. SiliconANGLE says the panel is not a statistically representative market survey [8].
For the person deciding where the next workload goes, I'd sort each candidate on two questions. Is GPU capacity what is blocking it today? And how much does it need to call data and services that already live in a hyperscaler?
Blocked and loosely coupled, such as a training run or a batch inference job, is the door SiliconANGLE describes. Pilot it, and judge the result on cost and throughput you measured yourself. Blocked and tightly coupled is where I'd expect a successful pilot to sit unrolled, because the rest of the application stays put [5]. Unblocked and loosely coupled comes down to price and staffing, and owning hardware is a real option; one buyer in this panel chose on-prem [11]. Unblocked and tightly coupled is the case for staying with the incumbent cloud. One buyer here stayed with Azure [11].
What to watch
- Whether CoreWeave discloses retention or enterprise revenue figures at its Fully Connected conference that test the seven current users' stated growth plans.
- Whether the three respondents still evaluating, and the one with a completed pilot, move to production or stay where they are.
- Whether the financial-services account running production load during its evaluation signs a full contract.