Emerging Computing Power Providers Quietly Profit Amid GPU Shortage

The AI boom has driven a tremendous demand for GPUs, with global demand expected to increase by 10 to 100 times. Over the past five years, GPUs have been in short supply due to the rise of cryptocurrencies and the 2020 supply chain crisis. The current hype in areas such as AIGC and autonomous driving has further exacerbated this situation. Therefore, tech giants like Microsoft are also facing GPU supply shortages.

Due to Nvidia's dominant position in the GPU market, it is difficult for other manufacturers to compete with them in AI model training. This may lead to a period of monopolistic situation, making it challenging to address the short-term limitations in GPU supply.

Nvidia GPUs are facing the challenge of supply-demand imbalance, which has also driven significant revenue growth for emerging GPU computing power providers like CoreWeave and Lambda Labs. It is expected that the demand for AI inference computing power will increase substantially, and the number of available GPUs will also increase. In the long run, the supply constraints are not expected to last too long. However, from the perspective of model training, shortages may persist.

CoreWeave and Lambda Labs are emerging private cloud GPU computing providers. Users can access NVIDIA's AI products and services in the cloud or in local data centers.

Both CoreWeave and Lambda Labs have close ties to NVIDIA. In April 2023, NVIDIA made a $100 million equity investment in CoreWeave, while Lambda received a $300 million investment from NVIDIA in another larger financing round.

CoreWeave, formerly known as Atlantic Crypto, initially focused on Ethereum mining operations. They provided cryptocurrency mining services by purchasing a large number of NVIDIA GPUs. In 2019, they shifted towards providing AI computing power services due to the increasing demand from AI startups like Stability/Stable Diffusion. The introduction of OpenAI's ChatGPT and the rise of generative AI further boosted their business revenue.

CoreWeave's revenue was around $25 million in 2022 and is expected to reach approximately $600 million by 2023. CoreWeave has adjusted its preliminary outlook for 2023, lowering its expected revenue and capital expenditure but increasing its multi-year commitments. This adjustment may be due to the GPU supply constraints that CoreWeave is also facing.

Lambda Labs was established in 2012 and transitioned to an AI computing provider in 2019/2020, offering cloud-based GPU capacity for rent. In terms of scale, Lambda is smaller than CoreWeave, but it is projected to generate revenue well over $100 million in 2023.

Emerging computational power providers are leveraging architectural advantages and pricing differentials. In addition to the supply advantage of NVIDIA, the rapid growth of CoreWeave and Lambda can be attributed to their differentiated competition compared to the three major players, including architectural and pricing differentiations.

CoreWeave claims to have designed clusters specifically for NVIDIA GPUs, citing their auto-scaling capabilities (partly based on their serverless orchestration technology and NVIDIA's InfiniBand network). This allows customers to access NVIDIA GPU clusters faster than Microsoft and Amazon.

CoreWeave offers 1-3 year Reserved Instance (RI) contracts with discounts for customers. Their pricing range for A100 is $1.10-$1.40 per hour, while for H100 it is $2.00-$2.20 per hour. In comparison, Lambda has lower hourly rates for GPUs, but their software stack is less mature.

Lambda differs from CoreWeave in terms of infrastructure. Lambda focuses on providing key services such as fast file system storage and GPUs to meet market demands more quickly. This is one of the reasons for its rapid growth.

Lambda offers the cheapest GPU cloud services in the world, including an H100 at $1.89 per hour and on-premises AI compute clusters. If users achieve 100% utilization, the prices can be further reduced. Lambda consistently maintains high utilization rates, enabling them to provide affordable prices. In contrast, many larger cloud providers tend to overprovision, resulting in higher peak capacity but also incurring costs for unused capacity to support price fluctuations and low utilization of resources.

The current popularity of artificial intelligence has driven a significant demand for GPUs. However, due to supply shortages, Nvidia is currently the only supplier in the market, which may lead to a temporary monopoly situation.

Emerging private cloud GPU infrastructure providers such as CoreWeave and Lambda Labs are rapidly gaining traction. They have attracted the attention of technology investors and experienced fast business growth due to their architectural advantages and pricing differentials. At present, the emerging computing power providers and the three cloud giants are cooperative relationships, but it is also possible to become competitive relationships in the future.

While GPU supply constraints may persist for some time, in the long run, the imbalance between supply and demand may be alleviated with increasing supply and technological advancements.