Invest in the Next AWS: Cloud Infrastructure Opportunity in Private Markets

The Next AWS? Investing in Cloud Infrastructure Before It Hits Wall Street

If you pause for a moment to look around, you’ll notice something so ordinary, it’s almost invisible: everyone is clothed. It’s so ingrained in our daily lives that we often overlook the transformative journey to affordable, accessible clothing.

Centuries ago, creating garments was an extraordinarily laborious process. Fibers were spun into thread by hand, woven into fabric on manually operated looms, and sewn together stitch by stitch. This was the norm until human ingenuity stepped in with a wave of mechanical innovations. The spinning wheel streamlined thread-making, while devices like the flying shuttle significantly sped up weaving. Mechanical looms followed, further accelerating production but they now faced a critical bottleneck: power. For most of history, the only way to drive these machines was through muscle—either human or animal.

That all changed with the arrival of the steam engine - it unleashed a new, abundant source of power that could run looms around the clock. The result? Clothing went from being a luxury for the wealthy, to a necessity everyone could afford. Steam power wove its way into the fabric of society, democratizing access to one of life’s most basic needs.

Fast forward to today, and another revolution is upon us. AI is taking the world by storm, promising to transform industries from healthcare to logistics. But like those early spinning wheels and looms, AI faces its own bottleneck: compute power. Training advanced AI models requires immense computational resources, and most companies rely on cloud infrastructure to meet these demands.

Most of this compute power comes from the dominant cloud providers—AWS(Amazon Web Services), Google Cloud, and Microsoft Azure—built primarily on CPUs (Central Processing Units). While reliable, CPUs weren’t designed for the high-intensity workloads that AI demands. That’s where CoreWeave comes in. Unlike traditional providers, CoreWeave focuses on GPUs(Graphic Processing Units), which are designed for high speed, parallel processing - making them ideal for AI. By leveraging GPUs, CoreWeave has managed to cut costs by up to 80% compared to traditional cloud services, offering a faster and more efficient solution for AI-driven businesses.

CoreWeave is doing for AI what the steam engine did for textiles—fueling a dramatic increase in productivity and enabling a new generation of innovators to emerge. Just as steam power once wove a more connected, clothed world, CoreWeave is weaving the infrastructure for an AI-driven future.

CoreWeave’s Traction

CoreWeave’s growth story is a series of remarkable milestones:

• 2017: CoreWeave was founded as Atlantic Crypto, initially renting GPUs for cryptocurrency mining.

• 2019: Raised $1.2 million in seed funding and pivoted to cloud infrastructure, growing revenue by 271% in three months.

• 2021: Rebranded as CoreWeave, signaling its shift to GPU-focused cloud computing.

• 2022: Reported $25 million in revenue, laying the groundwork for exponential growth.

• 2023:

  • Signed a multi-year agreement with Microsoft, valued in the billions, to supply GPU compute for AI workloads.

  • Became a preferred cloud services provider for Nvidia, gaining access to elite pricing and cutting-edge GPUs.

  • Expanded its physical infrastructure from 3 data centers to 14–18 facilities, each housing approximately 20,000 GPUs.

  • Projected to close 2023 with $500 million in revenue, a 25x increase over the previous year.

• 2024 (Projected): Revenue is estimated to reach $2.3 billion, continuing CoreWeave’s trajectory as a dominant GPU cloud provider.

Business Model: How CoreWeave Makes Money

CoreWeave has positioned itself as a next-generation cloud provider, specializing in GPU-based computing power to support industries that require immense processing capability. These include artificial intelligence (AI), machine learning (ML), visual effects, and other computationally intensive fields. By focusing on GPUs—originally designed for video games but now critical for cutting-edge technologies—CoreWeave enables businesses to access high-performance infrastructure without the need for costly hardware investments.

Instead of purchasing and managing their own equipment, customers rent GPU power from CoreWeave on a pay-as-you-go model, allowing them to scale up or down based on their needs. This flexibility saves businesses both time and money, making CoreWeave particularly attractive to startups and enterprises scaling AI/ML applications.

CoreWeave’s business revolves around three primary sources of revenue, all designed to give customers flexible and scalable solutions for their compute needs:

1. GPU Rentals:

   • At the heart of CoreWeave’s offerings are its GPU rentals, allowing businesses to access cutting-edge processing power without upfront hardware costs.

   • Prices range from $0.24/hour for basic GPUs to $4.25/hour for the latest high-performance models like the Nvidia H100. These GPUs are ideal for tasks like AI model training, rendering, and simulations.

2. Add-On Services:

   • CoreWeave expands its value by offering complementary services such as:

     • CPU Rentals: For workloads that don’t require GPU acceleration.

     • Data Storage: Flexible, scalable storage options priced by the gigabyte.

     • Networking: Includes public IP addresses ($4/month) and private virtual networks ($20/month per VPC) for secure and seamless deployments.

3. Specialized Solutions:

   • CoreWeave caters to customers with complex needs by offering features like Kubernetes-based infrastructure management and high-performance networking. These solutions ensure that businesses can fine-tune their setups for maximum efficiency.

The company’s financial model boasts ~85% gross margins, derived from the difference between operational costs and customer revenue. CoreWeave's major expenses include the upfront purchase of GPUs, electricity, cooling, and staffing for data center operations and customer support. Despite these costs, GPUs have a multi-year lifespan, during which CoreWeave can rent them out repeatedly, maximizing their profitability.

Examples of GPU Economics:

• High-End GPUs (Nvidia H100):

  • Cost: ~$30,000 per unit.

  • Break-Even: At ~80% utilization, these GPUs recover their cost over time, despite narrower margins and higher operational expenses.

• Older GPUs (Nvidia A40):

  • Cost: ~$4,500 per unit (purchased pre-AI boom).

  • Annual Revenue: These GPUs generate ~$8,800 annually at 80% utilization, offering significantly higher margins than newer, more expensive models.

CoreWeave employs strategies to improve operational efficiency and further enhance margins, such as reducing electricity costs and optimizing cooling systems in its data centers. Additionally, its expansion into adjacent services mirrors the growth strategy of AWS, offering specialized solutions like CPU compute, networking, and storage to diversify revenue streams and meet varied customer needs.

CoreWeave’s Future Opportunity

CoreWeave is uniquely positioned at the intersection of two transformative trends: its strategic partnership with Nvidia and the surging demand for GPU compute driven by AI. Unlike other cloud providers, CoreWeave’s specialization in GPU infrastructure avoids conflicts of interest with Nvidia’s chip design business. This alignment mirrors the relationship between TSMC and its customers, enabling CoreWeave to secure access to cutting-edge GPUs at favorable pricing. The result is a competitive advantage that CoreWeave executives estimate places them two years ahead of their closest rivals. Over time, this partnership has the potential to evolve into a formidable moat, cementing CoreWeave’s dominance in the GPU cloud market.

Meanwhile, the growth of AI is reshaping the computing landscape. Between 2021 and 2023, over $173 billion was invested into approximately 58,000 AI startups worldwide. These companies are creating entirely new categories of compute-intensive workloads that traditional cloud providers weren’t built to handle. CoreWeave’s GPU-first infrastructure is purpose-built to meet these demands, offering superior performance and cost efficiencies that AI companies need to scale. While legacy providers attempt to retrofit their services for this new reality, CoreWeave has positioned itself as the provider of choice for the AI-driven future, poised to capture a significant share of this rapidly expanding market.

In combining the strategic advantages of its Nvidia partnership with the rising wave of AI innovation, CoreWeave is uniquely equipped to redefine the possibilities of cloud computing and secure its place as a market leader.

CoreWeave’s Competition

CoreWeave operates in a highly competitive cloud computing market, facing pressure from both general-purpose cloud giants and niche GPU-focused providers. Each competitor offers distinct advantages, targeting various segments of the AI and machine learning ecosystem.

Legacy Cloud Providers

The most formidable long-term competition comes from the major cloud providers: Amazon Web Services (AWS) ($80 billion in revenue in 2023), Google Cloud ($75 billion in revenue in 2023), and Microsoft Azure ($26 billion in revenue in 2023). With their enormous revenue scales, these giants have the resources to invest heavily in acquiring GPUs and developing their own AI-focused silicon to compete with Nvidia’s GPUs.

However, CoreWeave has maintained an edge by leveraging its preferential treatment from Nvidia. Unlike AWS, Google Cloud, and Azure, which are all developing proprietary AI chips, CoreWeave remains solely focused on providing GPU compute, making it an ideal partner for Nvidia. This alignment has allowed CoreWeave to secure better access to GPUs, even as global demand continues to surge. Additionally, legacy providers' infrastructures are not fully optimized for the compute-intensive demands of AI/ML, allowing CoreWeave to offer more efficient and cost-effective solutions tailored specifically to these workloads.

GPU-Specific Competitors

Among GPU-focused providers, Lambda Labs presents notable competition. Like CoreWeave, Lambda Labs rents GPUs from Nvidia and provides them to companies building AI applications. Lambda positions itself as a cost-effective option for smaller companies and developers, offering H100 PCIe GPUs at $2.49 per hour (compared to CoreWeave’s $4.25 per hour). However, Lambda lacks the capability to serve larger-scale workloads, as it does not offer the more powerful HGX H100 configurations that CoreWeave provides at $27.92 per hour for groups of eight GPUs, which are optimized for massive AI training tasks. Lambda Labs projected $250 million in revenue for 2023, with ambitions for $600 million in 2024, fueled by backing from investors like Thomas Tull’s US Innovative Technology fund and Gradient Ventures.

Together operates more as a GPU reseller than a direct competitor. The company aggregates GPUs from sources like CoreWeave, Google Cloud, and even crypto miners, then bundles them with software for training and fine-tuning open-source AI models such as Meta's Llama 2 and its own RedPajama. Together’s Forge product, launched in 2023, combines compute and training capabilities, offering Nvidia A100 and H100 clusters at 20% of AWS's cost. While Together achieved a $10 million annual revenue run rate by the end of 2023, its reliance on reselling GPUs from providers like CoreWeave limits its ability to directly compete in high-performance computing.

What Sets CoreWeave Apart: While legacy providers cater to general compute needs and GPU-focused competitors target specific niches, CoreWeave’s early market leadership, Nvidia partnership, and access to cutting-edge GPU configurations position it as the go-to choice for enterprise-scale AI applications. By specializing in high-performance GPU workloads and maintaining a strong relationship with Nvidia, CoreWeave offers tailored infrastructure and unmatched efficiency for a rapidly expanding market.