CoreWeave
Last reviewed
May 17, 2026
Sources
28 citations
Review status
Source-backed
Revision
v3 ยท 6,984 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 17, 2026
Sources
28 citations
Review status
Source-backed
Revision
v3 ยท 6,984 words
Add missing citations, update stale details, or suggest a clearer explanation.
CoreWeave, Inc. is an American artificial intelligence cloud computing company headquartered in Livingston, New Jersey. The company specializes in providing cloud-based GPU infrastructure to AI developers and enterprises, and also develops its own chip management software. Originally founded in 2017 as a cryptocurrency mining operation, CoreWeave pivoted to GPU cloud services in 2019 and has since become one of the largest specialized cloud providers for AI workloads. The company went public on the Nasdaq stock exchange on March 28, 2025, under the ticker symbol CRWV. By May 2026, CoreWeave was widely considered the bellwether of the so-called neocloud category, with a market capitalization of roughly $59 billion, a contracted revenue backlog of $99.4 billion, and customer commitments from nine of the world's ten largest AI model developers.
CoreWeave was founded in 2017 in New Jersey by Michael Intrator, Brian Venturo, and Brannin McBee, along with Peter Salanki. All three primary founders came from backgrounds in commodities trading. Michael Intrator had been working as a portfolio manager and founder of Hudson Ridge Asset Management, an energy industry hedge fund. The founders had built a machine learning model to help them pick investments in the data-heavy energy industry, and Brannin McBee ran the data firm they used.
The company was originally incorporated as Atlantic Crypto Corp. and operated as an Ethereum mining company. The founders used NVIDIA GPUs to verify blockchain transactions, running operations out of a New Jersey data center. At the time, Bitcoin and other cryptocurrencies were rising sharply in price, making GPU-based mining a lucrative business.
Following the cryptocurrency market crash of 2018, Ethereum mining became significantly less profitable. In 2019, the company was renamed CoreWeave, and the founders decided to leverage their large existing inventory of GPUs to provide cloud computing infrastructure to companies. This pivot proved prescient, as the demand for GPU compute was about to explode with the rise of generative AI.
The company initially targeted clients in visual effects (VFX) rendering, scientific computing, and other GPU-intensive workloads. CoreWeave differentiated itself from traditional hyperscale cloud providers like Amazon Web Services, Microsoft Azure, and Google Cloud by offering GPU-specialized infrastructure at lower costs and with faster provisioning times.
As the AI industry accelerated, particularly following the release of ChatGPT in late 2022, demand for GPU compute surged. CoreWeave was well-positioned to capitalize on this trend. In April 2023, the company raised $221 million in a Series B funding round led by Magnetar Capital, with participation from NVIDIA, former GitHub CEO Nat Friedman, and former Apple executive Daniel Gross. This round valued the company at approximately $2 billion.
In January 2023, CoreWeave acquired Conductor Technologies, a cloud-based rendering startup that simplified access to cloud resources at scale. This acquisition enhanced CoreWeave's offerings to VFX studios and brought its headcount to 95 employees. Mac Moore, former CEO of Conductor Technologies, began leading CoreWeave's Media and Entertainment division.
In September 2023, CoreWeave and NVIDIA announced that they had built what they described as the fastest AI supercomputer in the world, located in Plano, Texas. The system, built using over 3,500 NVIDIA H100 GPUs, trained a GPT-3-scale large language model with 175 billion parameters in under 11 minutes, over 29 times faster than competitors in a benchmark test.
In May 2024, CoreWeave raised $1.1 billion in Series C funding led by Coatue Management, with participation from Magnetar Capital, Altimeter Capital, Fidelity Management & Research Co., and Lykos Global Management. This round valued the company at $19 billion.
Also in May 2024, the company secured a $7.5 billion debt financing facility led by Blackstone and Magnetar Capital, one of the largest private debt deals in the technology sector. A further $650 million credit facility was announced in October 2024. The Blackstone facility, internally referred to as DDTL 2.0, was unusual in being secured by GPU contracts rather than traditional real estate or hard assets.
In November 2024, a secondary round was completed with lead investors including Fidelity, Jane Street Capital, Macquarie Capital, and Magnetar Capital, at a valuation of $23 billion. During 2024, CoreWeave hired key executives to prepare for its public listing. Nitin Agrawal, formerly of Google, joined as Chief Financial Officer. Sachin Jain, previously from Oracle's AI department, was hired as Chief Operating Officer in August 2024.
CoreWeave went public on March 28, 2025, listing on the Nasdaq under the ticker CRWV. The IPO priced at $40.00 per share, with 37.5 million shares offered (36.59 million sold by CoreWeave and 910,000 by selling stockholders). The offering raised approximately $1.5 billion, making it the largest AI-related IPO by amount raised, according to Dealogic.
The IPO launched during a period of broader market volatility related to U.S. tariff announcements, and CoreWeave had to cut the offering size and price in the final hours before listing. NVIDIA emerged as a critical anchor on the deal, agreeing to take a portion of the offering itself, which several reports described as effectively rescuing the IPO from collapse. CEO Michael Intrator described the listing as "incredibly successful."
By May 2025, the company's market capitalization had doubled to roughly $49 billion. The stock reached a 52-week high of $187.00 in late 2025 before pulling back into the $60-90 range during a broader sector rotation that affected AI infrastructure names. In February 2025, Glenn Hutchins was appointed lead independent director of the board.
CoreWeave used its newly public equity aggressively to expand both vertically and into the software layer above raw compute.
On May 5, 2025, CoreWeave completed the $1.7 billion acquisition of Weights & Biases, the leading developer tools and observability platform for AI model training, evaluation, and monitoring. The Weights & Biases platform was used by more than 1,400 organizations, and CoreWeave framed the deal as creating an end-to-end stack that ran from bare-metal GPU clusters through model lifecycle tooling. Following the close, CoreWeave launched three new AI cloud software products in June 2025: Mission Control Integration, Weights & Biases Inference, and Weave Online Evaluations.
In July 2025, CoreWeave announced an all-stock agreement to acquire Core Scientific, a publicly traded Bitcoin miner that had been hosting a significant portion of CoreWeave's compute capacity. The implied equity value of approximately $9.0 billion would have given CoreWeave roughly 1.3 gigawatts of gross power across Core Scientific's national data center footprint, with an additional one gigawatt or more of expansion potential. Core Scientific stockholders were offered 0.1235 newly issued CoreWeave Class A shares per Core Scientific share, valuing the target at about $20.40 per share, a 66% premium to its pre-announcement price. On October 30, 2025, Core Scientific terminated the merger agreement after preliminary results showed the proposal did not receive the necessary stockholder approval, with shareholders concluding the offer undervalued the company in light of CoreWeave's surging share price during the summer. CoreWeave continued to lease capacity from Core Scientific as a customer following the termination, and Intrator publicly suggested the company would pursue verticalization through self-builds and other acquisitions instead.
The first half of 2026 was defined by an unusual concentration of multi-billion-dollar customer deals that pushed CoreWeave's contracted backlog to nearly $100 billion. Industry analysts referred to this stretch as the company's "contract supercycle."
On March 4, 2026, CoreWeave announced a multi-year strategic partnership with Perplexity to power Perplexity's inference workloads on CoreWeave Cloud, with Perplexity running on dedicated NVIDIA GB200 NVL72 clusters. Perplexity began running production inference using CoreWeave Kubernetes Service and Weights & Biases model management tools. In a reciprocal arrangement, CoreWeave deployed Perplexity Enterprise Max for use across its own organization. The deal sent CoreWeave shares up 5.7% in pre-market trading on the day it was announced.
On April 9, 2026, CoreWeave and Meta announced an expanded long-term agreement worth approximately $21 billion that runs from 2027 through December 2032. The expansion extended a prior $14.2 billion contract signed in September 2025, bringing total committed value between the two companies to roughly $35 billion. Meta agreed to deploy CoreWeave capacity across multiple locations with some of the initial deployments of the NVIDIA Vera Rubin platform, with the workload mix oriented toward AI inference rather than training. CoreWeave executives told analysts on subsequent calls that the expanded Meta relationship would ensure no single customer would represent more than approximately 35% of total sales going forward, a meaningful improvement on the Microsoft concentration risk that had defined the pre-IPO and immediate post-IPO period.
On April 10, 2026, the day after the Meta announcement, CoreWeave disclosed a multi-year agreement with Anthropic to support the development and deployment of the Claude family of AI models. Capacity under the Anthropic agreement was scheduled to come online beginning in the second half of 2026, with a phased rollout that analysts at Wells Fargo, Citi, and Morgan Stanley modeled in the $4 billion to $7 billion range over five years. Intrator told Bloomberg the contract was "a multibillion-dollar contract" but declined to disclose specific terms. The signing of Anthropic meant that nine of the ten largest AI model developers in the world were now CoreWeave customers, with the lone exception being a closed Chinese lab. The stock jumped roughly 12% to $103 on the announcement.
On April 15, 2026, CoreWeave disclosed a $6 billion multi-year cloud agreement with Jane Street Group, the quantitative trading firm, accompanied by a separate $1 billion equity investment from Jane Street at $109.00 per share. Jane Street agreed to use CoreWeave capacity, including early Vera Rubin systems, for the trading firm's machine learning training and inference workloads. The combined commitment made Jane Street one of CoreWeave's five largest shareholders and was widely interpreted as a sign that demand for high-end GPU capacity was extending beyond model developers and into financial services. Together with the Anthropic, Meta, OpenAI, and Perplexity deals, CoreWeave reported it had secured at least $63.9 billion of new contracts in the first quarter of 2026 alone.
| Name | Role | Background |
|---|---|---|
| Michael Intrator | Co-founder, CEO | Former portfolio manager and founder of Hudson Ridge Asset Management, an energy industry hedge fund |
| Brian Venturo | Co-founder, Chief Strategy Officer | Former commodities trader specializing in energy futures |
| Brannin McBee | Co-founder, Chief Development Officer | Ran a data analytics firm used by the other founders |
| Peter Salanki | Co-founder, CTO | Technical co-founder responsible for engineering and architecture |
| Nitin Agrawal | CFO | Formerly at Google; joined in 2024 |
| Sachin Jain | COO | Formerly at Oracle's AI department; joined August 2024 |
| Michelle O'Rourke | Chief People Officer | Appointed in early 2026 to lead talent strategy as headcount scaled past 2,000 |
| Glenn Hutchins | Lead Independent Director | Appointed February 2025; co-founder of Silver Lake |
As of mid-2025, CEO Michael Intrator's personal wealth had grown to an estimated $10 billion, making him one of the wealthiest individuals in the AI infrastructure industry. Co-founders Brian Venturo and Brannin McBee held similarly outsized stakes that became liquid for the first time at the IPO.
CoreWeave operates as a specialized cloud computing provider focused exclusively on GPU-accelerated workloads. Unlike general-purpose hyperscale cloud providers that offer a broad range of computing services, CoreWeave concentrates on delivering high-performance GPU infrastructure optimized for AI training, inference, rendering, and high-performance computing (HPC).
CoreWeave primarily generates revenue through long-term contracts with enterprise customers who reserve GPU capacity. These contracts typically span multiple years, providing the company with predictable, recurring revenue. As of the end of Q1 2026, CoreWeave reported a revenue backlog of $99.4 billion, up sharply from $66.8 billion at the end of 2025. The backlog mostly consists of take-or-pay capacity reservations, which obligate the customer to pay for committed GPU hours regardless of usage.
The company's customer concentration has been a notable feature of its business and a recurring topic in equity research. In 2024, Microsoft accounted for 62% of CoreWeave's total revenue. By the June 2025 quarter, Microsoft still represented 71% of revenue. The 2025-2026 wave of deals with OpenAI, Meta, Anthropic, Perplexity, and Jane Street was explicitly designed to dilute that concentration. CoreWeave management told analysts that, on a forward-looking basis, no single customer would exceed roughly 35% of revenue once the new contracts ramped, with Meta expected to become the single largest customer in the second half of the decade.
CoreWeave claims to be up to 80% more cost-effective than major hyperscale cloud providers for GPU workloads, thanks to its purpose-built infrastructure and usage-based billing model. The company's instances can launch up to 35 times faster than traditional virtual machines on legacy cloud platforms. In Q1 2026 earnings commentary, executives also emphasized speed-to-power and speed-to-GPU as competitive moats, arguing that the time required to bring large clusters online has become as important as the price per GPU-hour.
The acquisition of Weights & Biases gave CoreWeave a software business that includes a model training platform (W&B Models), an inference serving product (W&B Inference, launched mid-2025), and an LLM evaluation and observability suite (Weave). The strategy is to layer high-margin software on top of the underlying compute, allowing CoreWeave to capture a larger share of the AI developer's total spend and to make the platform stickier than raw GPU rental would otherwise be.
CoreWeave's cloud platform is built on a Kubernetes-native architecture designed specifically for large-scale, GPU-intensive tasks. The CoreWeave Kubernetes Service (CKS) is a purpose-built orchestration layer that allows AI researchers to deploy thousands of GPUs in seconds.
CKS clusters use bare-metal nodes with NVIDIA BlueField Data Processing Units (DPUs) for offloading node and resource management processes. The platform natively integrates with workload orchestration tools like Slurm, KubeFlow, and KServe.
CoreWeave developed SUNK (Slurm on Kubernetes), an implementation of Slurm deployed on Kubernetes using Helm. SUNK is designed to be a unified training system for demanding AI workloads, delivering production-grade reliability and operational visibility for large, long-running training jobs.
SUNK packages everything needed to run a Slurm cluster in a configurable Helm chart that deploys Kubernetes resources, with base images that pre-install CUDA, NCCL, InfiniBand, environment modules, and Conda. Each Slurm node runs as a Kubernetes Pod with a slurmd container, and SUNK uses declarative node definitions to define and scale the cluster.
CoreWeave Mission Control monitors cluster health end-to-end, detects silent hardware issues and GPU stragglers, and mitigates failures before they stall synchronous training. Mission Control integration with Weights & Biases, announced in mid-2025, surfaces hardware-level events directly inside the W&B experiment tracking interface.
CoreWeave's network infrastructure is built on NVIDIA InfiniBand with SHARP (Scalable Hierarchical Aggregation and Reduction Protocol), providing high-bandwidth, low-latency interconnects between GPUs. The company also uses NVIDIA Quantum-2 InfiniBand networking for its latest deployments. Quantum-X800 networking, suitable for the Vera Rubin generation, was being qualified in CoreWeave labs in early 2026 for production rollout alongside the Rubin platform.
CoreWeave offers a range of NVIDIA GPU accelerators across multiple generations:
| GPU family | Models available | Key use cases |
|---|---|---|
| NVIDIA Vera Rubin | Initial GB and CPX deployments (2H 2026) | Next-generation training and inference, including early Meta and Jane Street workloads |
| NVIDIA Blackwell Ultra | GB300 NVL72 | AI reasoning, agentic workloads, large-scale inference |
| NVIDIA Blackwell | GB200 NVL72 | Large-scale AI training, inference at scale |
| NVIDIA Hopper | H100, H200 | Large language model training, HPC |
| NVIDIA Ada Lovelace | Various | Inference, rendering, VFX |
In April 2025, CoreWeave became the first AI cloud provider to make NVIDIA GB200 NVL72 systems generally available. Cohere, IBM, and Mistral AI were among the first customers to gain access, with Cohere using GB200 capacity for its North enterprise agent platform and reporting up to three times more training throughput on 100-billion-parameter models compared with the previous Hopper generation. IBM used CoreWeave's GB200 systems to train its next-generation Granite models for the watsonx platform, and Mistral AI used its initial thousand-GPU allocation to train new open-weight models.
In July 2025, CoreWeave became the first hyperscaler to deploy the NVIDIA GB300 NVL72 platform, which delivers up to a tenfold improvement in user responsiveness and a fivefold improvement in throughput per watt compared to the previous-generation Hopper architecture. In 2026, CoreWeave was named by NVIDIA as one of the first cloud providers expected to deploy the Vera Rubin platform in the second half of the year. Some of the initial Rubin deployments were earmarked for Meta and Jane Street under the contracts signed in April 2026.
In June 2025, CoreWeave, in collaboration with NVIDIA and IBM, submitted the largest-ever MLPerf Training v5.0 benchmark using nearly 2,500 NVIDIA GB200 Grace Blackwell Superchips. The system trained the Llama 3.1 405B model in 27.3 minutes, achieving a record result on the most complex model in the benchmark suite.
CoreWeave has reported up to 20% higher GPU cluster performance compared to alternative cloud solutions, attributing this to its purpose-built infrastructure and network optimization. In MLPerf Inference benchmarks released in 2026, GB300-based systems run on CoreWeave's platform delivered some of the leading published throughput numbers on Llama 3 405B and Mixtral 8x22B inference workloads.
By May 2026, CoreWeave operated across nearly 50 data center locations and had surpassed one gigawatt of operational data center capacity, with a stated target of exceeding 1.7 gigawatts by the end of 2026. The company also disclosed that it had approximately 3.5 gigawatts of total contracted power available across its existing leased and self-built footprint, with a stated goal of bringing the bulk of that capacity online by the end of 2027 and reaching as much as 8 gigawatts by 2030.
Following the termination of the Core Scientific acquisition in October 2025, CoreWeave shifted toward a hybrid model that combines third-party colocation leases with self-built greenfield facilities. The self-build push allows the company to capture the margin that would otherwise accrue to a landlord and to control the design of the buildings themselves around liquid-cooled, high-density GPU racks.
| Location | Details |
|---|---|
| Plano, Texas | Approximately 450,000 sq ft facility housing thousands of NVIDIA GPUs; original $1.6 billion supercomputer built in partnership with NVIDIA in 2023 |
| Kenilworth, New Jersey | 392,600 sq ft AI data center; up to $1.8 billion investment; first award under New Jersey's Economic Development Authority data center tax credit program; targeted to be operational in early 2027 |
| Ellendale, North Dakota | Polaris Forge 1 campus leased from Applied Digital; 400 MW committed across multiple buildings; first 50 MW achieved Ready for Service in October 2025, 100 MW phase live in Q4 2025, 150 MW second building targeted for mid-2026, third 150 MW building optioned for 2027 |
| Lancaster, Pennsylvania | Up to $6 billion planned investment in a purpose-built AI data center supporting AI workloads exclusively |
| Douglasville, Georgia | High-density colocation facility operated in collaboration with Flexential |
| Hillsboro, Oregon | Facility operated by Flexential |
| Volo, Illinois | Facility powered by Bloom Energy solid oxide fuel cells |
| Multiple states (joint venture) | $5 billion joint venture with Blue Owl Capital and PowerHouse Data Centers covering New Jersey, Pennsylvania, Texas, Nevada, and Kentucky |
| Location | Details |
|---|---|
| Crawley, United Kingdom | First European data center, operational October 2024; hosts NVIDIA H200 GPUs with Quantum-2 InfiniBand |
| London Docklands, United Kingdom | Second UK facility, operational December 2024 |
| Norway | Planned facility as part of $2.2 billion continental expansion; powered by 100% renewable energy |
| Sweden | Planned facility as part of $2.2 billion continental expansion; powered by 100% renewable energy |
| Spain | Planned facility as part of $2.2 billion continental expansion; powered by 100% renewable energy |
CoreWeave's European expansion represents a total commitment of $3.5 billion, combining the $1.3 billion UK investment with the $2.2 billion continental expansion. The European data centers are designed to meet regional regulatory and operational requirements, providing data-sovereign cloud services with low-latency performance.
| Date | Round | Amount | Lead investors | Valuation |
|---|---|---|---|---|
| October 2019 | Series A | $2.5 million | Hudson Bay Capital | - |
| November 2021 | Convertible Note | $50 million | Magnetar Capital | - |
| December 2022 | Debt Financing | $100 million | Magnetar Capital | - |
| April 2023 | Series B | $221 million | Magnetar Capital | ~$2 billion |
| 2023 | DDTL 1.0 | Up to $2.3 billion | Magnetar Capital | - |
| May 2024 | Series C | $1.1 billion | Coatue Management | $19 billion |
| May 2024 | DDTL 2.0 | $7.5 billion | Blackstone, Magnetar Capital | - |
| October 2024 | Credit Facility | $650 million | - | - |
| November 2024 | Secondary Round | - | Fidelity, Jane Street Capital, Macquarie Capital, Magnetar Capital | $23 billion |
| March 2025 | IPO | $1.5 billion | Public offering on Nasdaq | ~$23 billion (at IPO price) |
| July 2025 | DDTL 3.0 | $2.6 billion | - | - |
| July 2025 | Senior Notes | $1.75 billion | - | - |
| January 2026 | Private Placement | $2.0 billion | NVIDIA | At $87.20/share |
| Q1 2026 | New debt raised | $8.5 billion | - | - |
| April 2026 | Private Placement | $1.0 billion | Jane Street | At $109.00/share |
By the end of Q1 2026, CoreWeave had raised more than $20 billion of debt and equity year-to-date and closed the quarter with approximately $25 billion in total debt outstanding. The capital stack was a mix of senior secured term loans (DDTL 1.0, 2.0, and 3.0), senior unsecured notes, GPU-collateralized vendor financing, and equipment leases. The DDTL 1.0 facility carries a floating interest rate in the mid-teens, while the more recent DDTL 2.0 and 3.0 facilities priced at lower spreads as CoreWeave's credit profile improved. NVIDIA's January 2026 $2 billion equity placement and a separate $6.3 billion commitment by NVIDIA to purchase any unsold compute capacity through April 2032 jointly served as a backstop that materially lowered CoreWeave's cost of debt by mid-2026.
CoreWeave has more than 30 institutional investors. Major holders include:
| Year | Annual revenue | Year-over-year growth |
|---|---|---|
| 2022 | $15.8 million | - |
| 2023 | $229 million | ~1,349% |
| 2024 | $1.92 billion | ~737% |
| 2025 | $5.13 billion | ~168% |
| 2026 (guidance) | $12 billion to $13 billion | ~134% to ~153% |
CoreWeave reiterated on its Q1 2026 earnings call that annualized revenue should exceed $30 billion by the end of 2027, a trajectory implied by the contracted backlog rather than by future bookings.
Despite rapid revenue growth, CoreWeave has not yet achieved profitability on a net income basis due to heavy capital expenditures on GPU hardware, data center buildouts, and debt servicing costs. The company's profitability picture is, however, sharply positive on an adjusted EBITDA basis.
| Metric | FY 2024 | FY 2025 | Q1 2026 |
|---|---|---|---|
| Revenue | $1.92 billion | $5.13 billion | $2.08 billion |
| Net loss | $863 million | $1.167 billion | $740 million |
| Adjusted EBITDA | $1.22 billion | $3.09 billion | $1.16 billion |
| Adjusted EBITDA margin | ~64% | ~60% | ~56% |
| Revenue backlog (period end) | ~$15.1 billion | $66.8 billion | $99.4 billion |
The company's Q4 2025 results showed revenue of $1.57 billion (up 110% year-over-year), adjusted EBITDA of $898 million (a 57% margin), and a net loss of $452 million. In Q1 2026 CoreWeave reported revenue of $2.08 billion, up 111.6% year-over-year, beating Wall Street's consensus of $1.97 billion. Adjusted EBITDA rose 90.9% to $1.16 billion. The diluted net loss of $1.40 per share reflected continued heavy depreciation of GPU assets and rising interest expense on the company's growing debt stack. Q2 2026 guidance came in modestly below analyst expectations, which the company attributed to the timing of new data center commissioning rather than weakening demand.
Microsoft has been CoreWeave's largest customer since the company's early growth phase, with much of the capacity reportedly subleased to OpenAI for training and inference of frontier models. In 2024, Microsoft accounted for 62% of CoreWeave's total revenue, and this concentration remained high at 71% of revenue in Q2 2025. Microsoft has publicly stated plans to spend roughly $10 billion on CoreWeave's services by the end of the decade. CoreWeave has been actively working to reduce its revenue dependence on Microsoft through the diversification campaign of 2025-2026.
CoreWeave has built a substantial relationship with OpenAI through a series of escalating contracts:
| Date | Contract value | Details |
|---|---|---|
| March 2025 | Up to $11.9 billion | Initial infrastructure agreement; OpenAI received $350 million in CoreWeave equity |
| May 2025 | Up to $4 billion | Expanded agreement for additional cloud computing capacity |
| September 2025 | Up to $6.5 billion | Further expansion of the partnership |
| 2025-2026 (cumulative) | Up to $22.4 billion | Combined value across all direct agreements |
The OpenAI partnership complements OpenAI's existing commercial relationships with Microsoft and Oracle, as well as its joint venture with SoftBank on the Stargate project. CoreWeave was also a major beneficiary of the May 2025 multi-party arrangement in which Google Cloud agreed to sell GPU capacity to OpenAI. Under that structure, CoreWeave supplies a portion of the underlying compute to Google Cloud, which then resells the capacity to OpenAI, with all parties commercially involved. The arrangement gave CoreWeave a third significant indirect channel to OpenAI demand, on top of its direct contracts and its Microsoft-mediated capacity.
In September 2025, CoreWeave secured a $14.2 billion contract with Meta to provide AI cloud infrastructure. Under the original agreement, Meta committed to pay CoreWeave up to approximately $14.2 billion through December 2031, with the option to expand its commitment through 2032. On April 9, 2026, CoreWeave and Meta announced an expanded agreement that adds approximately $21 billion in committed value running from 2027 through December 2032, bringing the cumulative total between the two companies to roughly $35 billion. Capacity under the expanded deal will use some of the initial NVIDIA Vera Rubin platform deployments and is distributed across multiple CoreWeave locations to optimize performance, resilience, and scalability for Meta's AI inference workloads. The Meta relationship is expected to become CoreWeave's single largest customer relationship over the life of the contract.
On April 10, 2026, CoreWeave announced a multi-year agreement with Anthropic to support the development and deployment of Anthropic's Claude family of AI models. Anthropic will run production-scale workloads on CoreWeave's cloud platform, with deployments coming online starting in the second half of 2026 and expanding through a phased rollout. Wells Fargo, Citi, and Morgan Stanley analysts modeled the contract value in the $4 billion to $7 billion range over five years. Intrator described the agreement as a multibillion-dollar contract without disclosing specific terms. With Anthropic onboard, nine of the ten leading AI model developers (Microsoft via its OpenAI relationship, OpenAI directly, Meta, Mistral, IBM, Cohere, NVIDIA's research arm, Google's research arm, and Anthropic) were CoreWeave customers.
On March 4, 2026, CoreWeave entered into a multi-year strategic partnership with Perplexity to support Perplexity's inference workloads. Perplexity runs production inference on dedicated NVIDIA GB200 NVL72-powered clusters via CoreWeave Kubernetes Service, and uses W&B Models to manage models from experimentation through production. The infrastructure supports Perplexity's growing consumer search product and its Sonar and Search API ecosystem. In a reciprocal commercial arrangement, CoreWeave deployed Perplexity Enterprise Max for its own employees to use across internal search and research workflows.
On April 15, 2026, Jane Street Group, a global quantitative trading firm, agreed to a $6 billion multi-year cloud capacity contract with CoreWeave and made a separate $1 billion equity investment at $109.00 per share. Jane Street is using CoreWeave compute, including early Vera Rubin systems, for the firm's machine learning training and large-scale inference workloads across its trading business. The equity stake made Jane Street one of CoreWeave's five largest shareholders and was widely cited as the first major capital-markets endorsement of CoreWeave's business by a non-traditional cloud customer.
In addition to the marquee contracts, CoreWeave provides infrastructure to:
The relationship between CoreWeave and NVIDIA is one of the most important dynamics in the AI infrastructure industry. NVIDIA is simultaneously a strategic investor, technology partner, key supplier, and customer of CoreWeave.
NVIDIA first invested in CoreWeave during the 2023 Series B funding round. In March 2025, during the IPO, NVIDIA anchored the offering with an additional purchase that was widely reported to have helped close the book. In January 2026, NVIDIA made a $2 billion private placement investment at $87.20 per share, lifting its total ownership to approximately 9%. Alongside that placement, NVIDIA committed to purchase up to $6.3 billion of CoreWeave compute capacity through April 2032 that would otherwise be unsold, effectively underwriting CoreWeave's buildout with a take-or-pay demand guarantee.
CoreWeave has consistently been among the first cloud providers to deploy new NVIDIA GPU architectures. The company was the first to make NVIDIA GB200 NVL72 systems generally available (April 2025) and the first hyperscaler to deploy NVIDIA GB300 NVL72 systems (July 2025). CoreWeave was named in 2026 among the first cloud providers expected to deploy the NVIDIA Vera Rubin platform in the second half of the year, with initial allocations earmarked for Meta and Jane Street. CoreWeave is also a participant in NVIDIA's DGX Cloud Lepton marketplace.
CoreWeave uses NVIDIA BlueField DPUs in its data centers for network and resource management, and its network infrastructure relies on NVIDIA InfiniBand and Quantum-2 networking technology. Quantum-X800 networking is expected with the Rubin generation.
CoreWeave operates in the rapidly growing GPU cloud infrastructure market. The category as a whole, sometimes called "neoclouds," was estimated at roughly $35 billion of revenue in 2026.
The traditional "Big Three" cloud providers (Amazon Web Services, Microsoft Azure, and Google Cloud) all offer GPU instances for AI workloads. CoreWeave differentiates itself through GPU specialization, faster provisioning, and competitive pricing. The hyperscalers offer broader service portfolios, while CoreWeave focuses exclusively on GPU-intensive compute. Microsoft has been both a customer and a competitor: it uses CoreWeave capacity at scale for OpenAI workloads while also building out its own GPU clusters under the Azure AI Foundry brand.
CoreWeave competes with a growing category of GPU-focused cloud providers:
| Company | Focus | Differentiation |
|---|---|---|
| Nebius | GPU cloud for AI | Public-market neocloud spun out of Yandex assets; signed $17.4 billion Microsoft deal in 2025 and $3 billion Meta deal in late 2025; smaller than CoreWeave but growing faster on a percentage basis |
| Lambda Labs | GPU cloud for AI | Developer-friendly experience; strong focus on researchers and ML engineers |
| Together AI | Training and inference | Pay-per-token model; focused on open-source model hosting |
| RunPod | On-demand GPU cloud | Lower price point; popular with individual developers |
| Vast.ai | GPU marketplace | Peer-to-peer model; lowest cost option |
| SambaNova | AI accelerator platform | Custom chips (RDU) rather than NVIDIA GPUs |
| Fireworks AI | Inference optimization | Focused on fast, cost-efficient inference |
| Crusoe Energy | Modular AI factories | Vertically integrated with power generation; deep Stargate involvement |
CoreWeave is positioned at the enterprise end of the GPU cloud market, requiring multi-year contract commitments for exclusive access to large GPU clusters. Lambda Labs targets more flexible, growth-stage companies, while Together AI focuses on startups with pay-per-token pricing. Nebius, in particular, emerged in 2025-2026 as the closest direct comparable to CoreWeave, with Wall Street analysts often modeling the two companies side by side.
CoreWeave has experienced rapid headcount growth. At the time of the Conductor Technologies acquisition in January 2023, the combined company had 95 employees. By the end of 2024, headcount had grown to 881. As of 2025, the company employed approximately 1,450 people. By early 2026, headcount had reached approximately 2,189 employees, augmented further by the integration of the Weights & Biases workforce. Michelle O'Rourke joined as Chief People Officer in early 2026 to lead talent strategy, culture, and HR as the workforce continued to expand alongside the data center buildout. Employee discussion forums periodically referenced narrow workforce reorganizations within specific teams during 2025-2026, but CoreWeave did not announce any broad-based layoffs and continued to hire net positive throughout the period.
CoreWeave trades on the Nasdaq under the ticker CRWV. Since its IPO at $40.00 per share in March 2025, the stock has been volatile but has trended sharply higher on a longer time frame:
| Date | Approximate share price | Market capitalization | Context |
|---|---|---|---|
| March 28, 2025 (IPO) | $40.00 | ~$23 billion | Initial public offering on Nasdaq |
| May 2025 | ~$80 | ~$49 billion | Stock doubles within two months of listing |
| Late 2025 | $187.00 (high) | ~$95 billion | 52-week high reached during contract-flow euphoria |
| March 2026 | ~$83.67 | ~$43 billion | Pullback during AI sector rotation |
| April 2026 | ~$103 | ~$54 billion | Rally on Meta expansion and Anthropic deal announcements |
| May 16, 2026 | ~$106.92 | ~$58.6 billion | Post-Q1 2026 earnings, Jane Street equity reset baseline |
Analyst sentiment turned more constructive over the spring of 2026. Wells Fargo's Michael Turrin raised his price target to $155 from $135 on May 8, 2026, maintaining an Overweight rating. Barclays raised its price target to $120 from $106 on May 11, 2026. Citi raised its price target to $158 from $155 on May 14, 2026. As of mid-May 2026, approximately 33 sell-side analysts covered the stock with a consensus Buy rating and a price-target midpoint near $132. Coverage was notable both for the breadth of bullish ratings and for the unusually wide range of bull and bear cases, reflecting genuine disagreement about how to value an AI infrastructure pure-play with rapid revenue growth, large losses, and a heavy debt load.
CoreWeave's aggressive expansion has been funded primarily through debt. By the end of Q1 2026, total debt had reached approximately $25 billion, with an additional several billion in lease liabilities. The DDTL 1.0 facility carries a floating interest rate in the mid-teens, while the more recent DDTL 2.0 and 3.0 facilities priced at lower rates. The NVIDIA $6.3 billion capacity backstop and Meta and OpenAI take-or-pay contracts together cover a substantial portion of debt service, but the absolute size of the debt stack remains historically large for a company at CoreWeave's revenue scale. Servicing this debt requires sustained revenue growth and presents a significant financial risk if demand for GPU compute were to slow.
CoreWeave's historical reliance on a small number of large customers, particularly Microsoft (which accounted for 71% of Q2 2025 revenue), represented a concentration risk. The wave of 2025-2026 deals with OpenAI, Meta, Anthropic, Perplexity, and Jane Street is designed to bring the largest single customer share down to around 35% over time, with Meta expected to overtake Microsoft as the largest single customer late in the decade. However, the customer base is still heavily skewed toward the handful of frontier AI labs and the largest internet platforms.
The GPU cloud business is extremely capital-intensive, requiring continuous investment in the latest GPU hardware and data center infrastructure. GPU technology evolves rapidly, and older hardware can become less competitive quickly. CoreWeave must continually invest billions to maintain its technological edge. Depreciation schedules on GPU assets have become a point of disagreement between management and short sellers, with bears arguing that the useful life of high-end GPUs is shorter than CoreWeave's depreciation assumptions imply, which would reduce reported earnings power over time.
The GPU cloud market is attracting significant investment from both hyperscale providers expanding their GPU offerings and new entrants building specialized infrastructure. CoreWeave's ability to maintain its competitive advantages in pricing, performance, and provisioning speed will be tested as the market matures. Nebius, in particular, has emerged as a fast-growing direct competitor.
CoreWeave's business depends almost entirely on NVIDIA GPUs, creating a single-supplier dependency. Any disruption in NVIDIA's supply chain, changes in NVIDIA's allocation priorities, or the emergence of competitive chip alternatives such as those from AMD or hyperscaler-built accelerators could affect CoreWeave's operations. The deep equity, commercial, and capital alignment with NVIDIA partially mitigates this risk but also concentrates strategic exposure on a single partner.
The October 2025 termination of the Core Scientific acquisition removed an opportunity to vertically integrate roughly 1.3 gigawatts of power into CoreWeave's footprint and forced the company to continue paying lease rates to a hosting provider whose stockholders rejected the offer. While CoreWeave responded by accelerating its self-build program, the failed deal underscored execution risk in large strategic M&A.
The industry-wide shift in 2025-2026 from a "GPU shortage" to a "power shortage" has direct implications for CoreWeave's growth model. Bringing the committed contract backlog online by end of 2027 depends on securing power, transmission, and site rights at unprecedented scale, in some cases competing with hyperscalers and other neoclouds for the same substations and utility interconnections.