In the corridors of the AI infrastructure industry, a company that most people outside of technology had never heard of two years ago is now making investment decisions on a scale that rivals the world’s largest corporations. CoreWeave plans $30 to $35 billion in capital expenditure in 2026 for AI data centres, more than doubling its 2025 spend of $14.9 billion. The company, which began as a cryptocurrency mining operation before pivoting to GPU cloud computing, has positioned itself as the specialist infrastructure provider for the most compute-intensive workloads in artificial intelligence, and its growth trajectory tells a story about the explosive demand for AI compute that the established hyperscalers alone cannot satisfy.
CoreWeave’s rise is a case study in how quickly the AI infrastructure market has evolved. In less than three years, the company has gone from a niche GPU hosting provider to one of the largest customers of Nvidia hardware outside of the Big Four hyperscalers. With $3.13 billion in available cash and multi-year contracts with some of the most important AI companies in the world, CoreWeave is building an infrastructure empire at a pace that reflects the urgency of AI demand. For those tracking the future of marketing technology, CoreWeave’s expansion illustrates the infrastructure layer that powers every AI-driven innovation.
The Scale of CoreWeave’s 2026 Investment
| CoreWeave Metric | Figure |
|---|---|
| 2025 Capital Expenditure | $14.9 billion |
| 2026 Capex Plan | $30-35 billion |
| Available Cash | $3.13 billion |
| Key Customer Deal | Perplexity (multi-year inference) |
The leap from $14.9 billion to $30-35 billion in a single year represents more than 100% growth in infrastructure spending. This acceleration is driven by customer demand that has outpaced CoreWeave’s most optimistic projections. AI companies need dedicated GPU capacity for training and inference workloads, and the established hyperscalers, despite their own massive capex programmes, cannot meet all of this demand. CoreWeave fills this gap by offering bare-metal GPU cloud infrastructure optimised specifically for AI workloads, without the overhead of general-purpose cloud services.
The Perplexity Partnership and Customer Base
Perplexity, the AI search company that has emerged as a potential challenger to Google, signed a multi-year deal to use CoreWeave data centres for inference workloads. This partnership illustrates the type of customer that CoreWeave serves: AI-native companies that need massive, reliable GPU compute for real-time applications where latency and throughput are critical.
The inference workload market is particularly important for CoreWeave’s growth. While training workloads tend to be episodic, large and concentrated among a small number of frontier AI labs, inference workloads are continuous and grow proportionally with user adoption. Every time a ChatGPT user sends a query, every time a Perplexity search runs, every time an enterprise application calls an AI model, inference compute is consumed. As AI applications reach hundreds of millions of users, the inference compute market is growing even faster than the training market, and CoreWeave is positioning itself to capture a significant share.
The GPU Cloud Model
CoreWeave’s business model differs fundamentally from the general-purpose cloud offerings of AWS, Azure and Google Cloud. Rather than offering a broad suite of cloud services including storage, databases, networking and application hosting, CoreWeave focuses specifically on GPU compute. This specialisation allows the company to optimise every layer of its stack, from the physical data centre design to the software that manages GPU allocation and workload scheduling, for the specific requirements of AI training and inference.
The financial model is built on long-term contracts. By securing multi-year commitments from customers like Perplexity, CoreWeave can finance its infrastructure expansion with greater confidence in future revenue. This contract-based approach also provides visibility that allows the company to plan its GPU procurement, data centre construction and capacity expansion with more precision than a spot-market approach would allow. Understanding this infrastructure layer is essential for those tracking generative AI applications and the platforms that enable AI-powered products.
Competition and the Broader Infrastructure Market
CoreWeave operates in a competitive landscape that includes not only the Big Four hyperscalers but also other GPU cloud providers such as Lambda, Together AI and various regional providers. However, CoreWeave’s scale, having spent $14.9 billion in 2025 and planning $30-35 billion in 2026, places it in a different category from most competitors. At this level of investment, CoreWeave is building infrastructure at a scale that approaches the smaller hyperscalers.
| Infrastructure Provider | 2026 AI Capex | Focus |
|---|---|---|
| Amazon AWS | ~$200 billion | Full-stack cloud + AI |
| Alphabet Google Cloud | $175-185 billion | Cloud AI + Search + TPUs |
| CoreWeave | $30-35 billion | GPU-specialist cloud |
| FiberLight (connectivity) | $350 million | Fibre for AI data centres |
The Path Forward
CoreWeave’s $30 to $35 billion capex plan for 2026 signals that the GPU cloud market has matured from a niche segment into a critical layer of the AI infrastructure stack. The company’s growth from a crypto mining operation to a $30 billion-per-year infrastructure investor in less than five years mirrors the explosive trajectory of the AI industry itself. As AI models continue to scale, as enterprise adoption accelerates, and as inference demand grows with every new application and user, the demand for specialised GPU compute will continue to expand. CoreWeave’s bet is that specialisation, rather than generalisation, will define the winners in the AI infrastructure market of the late 2020s. For those following global advertising technology and the infrastructure enabling the AI economy, CoreWeave’s trajectory offers one of the clearest signals of where the industry is heading.
