Coreweave Co
Coreweave was founded in 2017 and was initially known as Atlantic Crypto, initially known for mining Ethereum. When the crypto market crashed in 2018 and GPUs flooded the market, the company pivoted and became Coreweave. The team took surplus GPU compute and began monetizing it as a niche cloud provider. The company lurked in the shadows until 2022/2023, until the generative AI boom skyrocketed demand for GPUs. With their access to Nvidia’s supply chain and rapidly spinning up data ranches, Coreweave emerged as a king. The firm backs OpenAI and other AI players who need fuck loads of fast, cheap, and scalable compute. Unlike traditional players who shoehorn GPUs into general-purpose clouds, Coreweave built the stack from silicon up. Coreweave isn’t an AWS clone but is the Ghost Recon of GPU cloud game offering fast, flexible, and optimized for speed to market in AI.
Coreweave's mission is to become the AI infrastructure layer for the post-cloud era. In the company’s own words, they are building “the most powerful AI cloud in the world.” Coreweave isn’t chasing AWS, nor are they trying to build a knockoff Azure; its vision includes weaponizing their control over GPU clusters and renting that control to AI companies. Think of Coreweave as an arms dealer, built for the AI water with deep, embedded contracts, customer infrastructure, and long-term recurring cash flow. Coreweave is building a closed-loop capital flywheel that includes acquiring GPUs before other mf’ers, building low-latency and high-density data ranchs to maximize utilization, and selling it to high-intensity clients. The company’s roadmap is to deepen vertical integration and own a larger share of the power and cooling stack by acquiring additional physical infrastructure.
CoreWeaves' biggest catalysts revolve around its ability to weaponize long-term contracts with industry titans. The company has already signed yuge deals with OpenAI, Microsoft, and Meta, and its revenue backlog now stands at $30 billion. These contracts are forward commitments for compute capacity that will provide Coreweave with visibility that few in the infrastructure world can claim. Coreweave has become the de facto compute backbone of the most important AI labs on the planet. Although customer concentration is a major risk for Coreweave, a departure or slowdown in any client could hurt its growth. Coreweave is an arms race that requires perfect execution, with guidance for 2025 capex sitting around $20 billion. Coreweave is betting the farm on Nvidia, but should the Blackwell underperform or if competitors leap ahead, billions in assets could be stranded.
CoreWeave is offering a GPU-first cloud infrastructure platform built for AI and high-performance workloads, unlike AWS, Azure, and Google Cloud, which are made for general use. Corewave stack is purpose-built; its services range from virtualized GPU compute pools that can be rented on demand to dedicated, custom-configured infrastructure clusters for enterprise clients who need entire racks to themselves. CoreWeave's revenue is highly concentrated, with a majority share of its consumption coming from Microsoft, OpenAI, and Meta. This is both a blessing and a curse; they have secured revenue growth but are left vulnerable to corporate giants. Other revenue streams exist but remain irrelevant to the multibillion-dollar contracts that currently dominate the books.
Coreweave’s footprint is primarily based in the United States, with 33 data ranchs as of their most recent report. This represents about 470MW of active power and about 2.2GW. With plants to continue their buildout in regions with cheap, reliable, and favorable regulatory frameworks. The company recently announced a $6 billion data ranch in Pennsylvania, with plans to scale from 100MW to 300MW at that site. CoreWeave has begun to push into the European markets, with offices in London and plans to deploy data ranches within the continent. If Coreweave wants to compete on a global scale, it must meet regulatory demands. Although a majority of short-term growth will remain US-driven, given the density of AI labs and corporate clients in North America.
2025 has already been an excellent year for Coreweave; the company went public, and shortly after, it landed a $6.5 billion deal with OpenAI. They also landed a $14.2 billion infrastructure contract with Meta and announced a $9 billion acquisition of Core Scientific. CoreWeave also became the first cloud provider to commercially deploy Nvidia’s Blackwell Ultra NVL72 Systems commercially, enabling clients with next-gen performance well before AWS and Azure. Each of these moves represents major growth opportunities for Coreweave as it transitions from a GPU lessor to an entrenched AI infrastructure utility. Coreweave has a strong pipeline of commitments and physical scale that few competitors can match outside of the big hyperscalers.
AI infrastructure is no longer a cloud feature; it's the product. Public cloud spend is expected to reach almost $1 trillion in 2025, with IaaS and PaaS the fastest-growing slices as enterprises race to start up AI workloads. Nvidia’s data ranch revenue alone surged to $40 billion in a single quarter and continued to rip higher in Q2. Power is a major bottleneck, with global data ranch electricity demand on track to more than double by 2030. In this environment, any provider that can lock GPUs, energized shells, and interconnect at scale will earn the lion's share of the market. Research nerds estimate that the AI infrastructure market sits around $200 billion in 2025, and will compound toward $400 billion by 2032. Citi projects that AI Capex will jump up to $3 trillion by 2029, reflecting a multiyear retooling of compute, data pipelines, and facilities. With AI as the driver, utilities and public power districts across the U.S are already flagging requests that rival their entire loads.
Several forces will support continued growth, compute intensity, electrification & grid policy, and capital availability and policy momentum. With model complexity and token throughput continuing to rise, Nvidia’s data ranch growth through 2025 and the early Blackwell ramp will capture the industry’s upgrade cycle and the stickiness of accelerator roadmaps. The experts expect data ranch electricity consumption to be over 2x by 2030, and the US Department of Energy warns of a resource adequacy with imports, covering almost 80% of the U.S. power supply in 2025. With sovereigns and enterprises treating AI as strategic infrastructure, debt markets are flowing with compute projects.
On the technology and innovation side of things, we are in a constant platform shift with firms consistently releasing new and updated gear. The Blackwell systems from Nvidia are engineered for longer context, higher batch throughput, and improved interconnect, which lowers cost per token and reduces time to train. At the systems level, innovation pressure is migrating from single-node FLOPs to cluster orchestration observability, and data-adjoining storage, where inference gateways become differentiators as much as silicon. While on the demand side, the mix is shifting away from sporadic training to always-on inference, particularly as video native models scale. Frontier training is also breaking limits, with research projecting individual training runs using 4-16GW by 2030. Winners will be those who can bind cutting-edge accelerators with software control planes and secure power into a product that enterprises can actually buy and use at scale.
Fuel The Keyboard
This blog runs on caffeine and 10-k’s. Send a cup of joe my way to keep the words flowing! Support me here.
Coreweave does not battle AWS, Azure, and Google Cloud head-on for generalized compute, but it lives in a specialized space. The big dawgs still control a bulk of the global market, AWS with 31%, Azure at 25%, and Google Cloud sitting around 11% as of 2025. While these giants are still dominant in the market, their GPU infrastructure is heavily oversubscribed, bureaucratically provisioned, and often optimized for long-tail enterprise workloads rather than AI use cases. Coreweave can take advantage of this within the AI GPU segment; it punches far above its weight. A more relevant comparison is Lambda Labs, Crusoe Cloud, and other niche providers that are chasing the same AI native clients. Yet none of Coreweave's competitors have signed multi-billion-dollar deals with OpenAI or Meta.
Coreweave’s moat is built on several layers that include its access to hardware, its orchestration and utilization software, and its being known as the AI-first cloud. Coreweave has built a strong relationship with Nvidia, enabling them to secure early allocations of the Blackwell Ultra systems while other clouds are still praying to get some H100s. CoreWeave’s proprietary scheduling layer maximizes GPU utilization and makes large-scale cloisters consumable in more ways than hyperscaler SKUs catalogs cannot. Coreweaves' business model emphasizes pre-paid, multi-year contracts, often with upfront cash flows that fund infrastructure expansion projects. Coreweave is becoming known for being an AI-first cloud, often being enterprises' first call. Coreweave has built a credible moat that is similar to what AWS once had with startups in the 2010s
The barriers to entry for the AI cloud game are big. It starts with capital intensity. CoreWeave guided $20-23 billion in capex for 2025. This reflects GPU purchases, data ranch buildouts, and power contracts. Something that few venture-backed challengers can deploy. Another barrier includes supply chain access, with allocation to Nvidia products as their strategy ensures that only trusted, proven operators with scale and client credibility get first dibs on next-gen GPU shipments. Clients like OpenAI and Meta cannot easily migrate to another frontier model for training without significant downtime. Multi-year contracts with prepayments further lock customers in, making clients more incentivized to expand rather than exit. Lastly, regulatory and operational bottlenecks in data ranch permitting, power availability, and grid interconnection act as natural barriers.
CoreWeave reported $1.21 billion in revenue in Q2 of 2025, reflecting a 207% increase YoY. This was driven by explosive demand for its products from mega contracts with hyperscalers. Coreweave is EBITDA profitable but GAAP negative, a class profile for a hyperscaler where prepayments and adjusted metrics can mask cash burn. Coreweave is a capital-intensive, leveraged, and ready-to-build company. As of their most recent report, their balance sheets show nearly $3 billion in quarterly capex additions, marking the most aggressive spending in company history. On the liability side, interest expenses of $267 million in Q2 highlight a large debt load, much of it tied to GPU financing and long-term leases. Coreweave asset-heavy positioning means depreciation and amortization will continue to climb.
Coreweave is currently running a growth-first monetize-later model. Operating cash flow is buoyed by prepayments from long-term contracts, but this is more working capital float than sustainable cash flow. Capex is devouring every dollar of inflow, with almost $3 billion in Q2 capex alone and FY25 guidance of $20-30 billion. Free cash flow is deeply negative and will be for a while. Basically, Coreweave is running on advanced customer financing plus debt markets; true organic cash generation is years away.
Fuel The Keyboard
This blog runs on caffeine and 10-k’s. Send a cup of joe my way to keep the words flowing! Support me here.
Coreweave has a revenue backlog of over $30 billion, and its growth trajectory is built on capacity velocity, with a core problem being insufficient energy capacity and difficulty in finding customers. The backlog is effectively deferred growth by precommitted contracts. On the product side, Coreweave is expanding its AI Cloud Platform beyond compute with its ‘Mission Control’ product. Mission Control is a GPU native scheduling and workload manager. This will give Coreweave stronger pricing power by bundling orchestration, monitoring, and optimization as value-added layers. This will lead to higher ARPU and improve stickiness, especially for large clients. Coreweave operates 33 data ranches across North America and plans new deployments in Europe, particularly in the UK, Ireland, and the Netherlands. Coreweave can charge clients a premium, as customers can work with them or wait for a 6-9 month GPU waitlist.
CoreWeave’s inorganic expansion includes strategic partnerships and a recent acquisition of Core Scientific. Just like CoreWeave, Core Scientific started as a crypto mining company. This deal will convert Coreweave from a lease-heavy, asset-light GPU renter into an infrastructure owner-operator. Integrating assets will enable Coreweave to have direct control over power and thermal management, as it has inherited gigawatts of power-ready space that are ready to deploy. Coreweave maintains strong partnerships with Nvidia, Dell, and Equinix to accelerate GPU provisioning and global connectivity. These moves unlocked contracts with hyperscalers and extended its lead time over Azure’s rollout.
Global AI infrastructure spending is projected to reach $3 trillion by 2030, driven by exponential model scaling, AI video slop, and enterprise adoption of LLM-based systems. As AI sovereignty laws emerge in Europe, the Middle East, and Asia, nations and corporations will need local AI clouds. Coreweaves data ranches can be replicated at scale faster than AWS-style campuses. This enables Coreweave to act as a white label infrastructure partners, selling capacity to sovereign clients who need compute without building it themselves. AI video generation, genomics, drug discovery, and simulation workloads are ramping up aggressively, and all depend on low-latency, large memory GPUs.
Coreweave is scaling from 470MW active power to over 202GW contracted capacity across its footprint. The company has plans to build a new $6 billion data ranch in Pennsylvania, which will anchor its next ramping and establish Coreweave’s largest AI cluster. CoreWeave is also cooking up expansions in Texas and North Carolina to further distribute geographic risk and latency profiles. Unlike traditional cloud providers that rely on third-party colocation partners, Coreweave owns the metal and the ground. Especially with the Core Scientific deal, which gives them access to facilities, power substation infrastructure, and relationships with local utility players. Coreweave’s R&D is about GPU maxxing and minimizing energy, turning each watt into more FLOPs per dollar.
CoreWeave’s supply chain is the backbone of its business, and unlike traditional cloud providers, it is designed around scarcity, not abundance. The company’s early access to Nvidia’s latest hardware is the foundation of its competitive position. Coreweave has built a hyper-efficient deployment infrastructure that enables it to compress the time between hardware arrival and when it can monetize the GPU. Coreweave partners with Dell Technologies and Super Micro to preassemble racks offsite, wiring and testing each pod before shipment. So when the racks reach CoreWeaves ranches, they are ready to roll. This method reduces downtime and eliminates class bottlenecks that hurt hyperscalers, whose scale often slows due to physical installation. Additionally, Coreweave’s 2.2 GW of contracted power is spread across Pennsylvania, Texas, and other low-cost states, ensuring both scale and resilience. Coreweave's ability to source, price, and control electricity is one of its most important differentiators in the AI compute market. Coreweaves network behaves more like a distributed HPC grid and less like a traditional enterprise cloud, and is optimized for moving multi-petabyte model checkpoints, training data, and inference traffic seamlessly across facilities.
Coreweave’s operation engine is running at full speed, but the faster they scale, the thinner the margin for errors becomes. Its entire model rests on access to Nvidia GPUs and complex logistics that span rack assemblers, energy providers, and data ranch contractors. Any delay in shipments, transformer shortage, or supplier insolvency could paralyze CoreWeave’s network. The Core Scientific acquisition faces integration complexity across employees, systems, and physical assets. Execution risk also lurks as CoreWeave’s leadership team is mostly founders and early-stage executives. Coreweave has shown finesse in its hardware procurement and contract capture, but scaling compliance and enterprise-level operations is a different ball game.
The AI infrastructure boom looms over how governments treat compute capacity as strategic infrastructure, subject to national security oversight, environmental regulation, and AI-specific policy frameworks. The United States has already imposed export controls on GPUs destined for China and other restricted markets. As per usual, in Europe, AI sovereignty and data localization laws will raise compliance costs for any expansion. Coreweaves' current pricing advantages may not always be around, if the supply-constrained GPU market is already attracting new entrants and cost-based competition. GPU scarcity could ease by late 2026, and hyperscalers could respond by cutting their prices on dedicated AI instances to protect market share. AI training is at an all-time high, but inference and application monetization remain uncertain. If AI adoption slows or if AI models become more compute-efficient, the demand curve for GPUs could flatten
Coreweave’s valuation hinges on continued hyper growth and strong contract conversion. The market has priced Coreweave more like a software platform than a utility. EV/EBITDA multiples hang around the 15-20x range when hyperscalers like Amazon and Microsoft trade at roughly 10-13x on mature cash flows. This gap is fueled by hype and illusions. Coreweave is burning cash with negative ROE and ROIC; they are funded by debt and customer prepayments. In short, its valuation risk is a narrative risk; if the AI music stops, Coreweaves' multiples will collapse with it. Coreweave also faces reputational risks from environmental impacts, or any allegations of unsafe labor conditions, wage disputes, or policy violations could quickly bend CW over. CoreWeave could also be drawn into lawsuits through clients whose AI systems generated harmful results or through disputes over power allocations, construction permits, or data breaches.
In a bull case, CoreWeave continues to strengthen its position and becomes a core utility provider for high-performance computing in the gen AI era. Revenue will scale to $22 billion by 2030 from its current $5.3 billion. In this bull case, the AI demand continues to rip, and Nvidia continues to innovate. In the bullverse, Coreweave becomes a digital utility similar to how energy companies rent out electricity. In a base case, investors begin to view CoreWeave as a growth utility, not a speculative AI play. Their story will remain compelling, but the company will remain an advantaged platform with a strong moat, but misses some of the speculative upside. Though in a bear case the company overbuilds, GPU supply normalizes faster than demand matures. Hyperscalers like Amazon and Microsoft begin to deploy comparable clusters at a lower cost, and the scarcity premium CW has evaporated. This would lead to margin compression, continued negative FCF, and its balance sheet would start to strain under leverage. Coreweave can survive, though its aura is gone, and it becomes another capital-intensive player.
The next twelve months for Coreweave are crucial, with the largest short-term inflection point expected from the activation of its Meta and OpenAI contracts, which are anticipated to boost its revenue in Q4 2025 and continue through 2026. These deals are not just revenue drivers but reputation drivers, showing that Coreweave can handle billion-dollar, multi-tenant workloads reliably. A successful integration of Core Scientific, a process that will determine whether the company can achieve the cost savings and power synergies it has promised. Continued expansion into Europe, particularly in London or Frankfurt, would mark the company’s transition from a U.S dog to a global operator.
Coreweave’s long-term upside is big, as they can become the backbone of the global AI economy, leasing out the physical substrate of intelligence itself. The combo meal of GPU priority, power control, and proprietary orchestration software gives it an advantage that few can replicate. As AI demand increases, CoreWeave could generate stable, recurring revenue streams that resemble a strong, regulated monopoly. In the best-case scenario, CoreWeave evolves into a hybrid of AWS economics and NextEra’s energy model, capital-intensive but tied directly to the growth of AI as a general-purpose technology. Though if Nvidia's supply advantage fades and GPU supply saturates, CoreWeaves' pricing model could take a hit.
With a revenue backlog of $30 billion or 2,037 miles high worth of $1 dollar bills, demand is visible and unmatched. The combo of power infra control, exclusive GPU access, and vertical integration places CoreWeave in a rare tier of operators. For these reasons, the bad boys over at Azar Capital Group will be giving CoreWeave a ‘BUY’ rating. The CWeave offers investors exposure to not only software hype, but tangible infrastructure bottlenecks that power AI. As long as AI continues to expand, as all the evidence shows, CoreWeave remains ready to roll as it can monetize every watt, every token, and every inference.
Fuel The Keyboard
This blog runs on caffeine and 10-k’s. Send a cup of joe my way to keep the words flowing! Support me here.
Disclosure
Buckle up—this analysis is strictly for informational and entertainment purposes only and is absolutely, positively NOT financial, investment, legal, or professional advice of any kind. It’s not a golden ticket, a sure bet, or a substitute for your own brainpower. Markets are a rollercoaster, and losses can hit harder than a freight train—consider yourself warned. Investors must do their own hardcore due diligence, dig into the details, and/or consult a licensed financial advisor, accountant, lawyer, or whoever else you trust before even thinking about making investment decisions. Past performance? It’s not a fortune teller’s promise for future gains—things can and will go sideways. The author, this platform, and anyone remotely connected to this content take zero responsibility for your financial moves, wins, or wipeouts. Proceed at your own risk, and don’t come crying to us if the market bites back!