Addressing the widespread demand for Nvidia GPUs, which dominated Silicon Valley conversations last summer, has evolved into a significant business opportunity within the AI sector.
This development has led to the emergence of new industry giants. For instance, Lambda, a company specializing in GPU cloud services powered by Nvidia GPUs, recently announced it has secured $320 million in funding, reaching a valuation of $1.5 billion. The company plans to use this investment to grow its AI cloud services.
This announcement followed a report from The Information that Salesforce had made a substantial investment in Together AI, valuing the company at over $1 billion. Furthermore, in December 2023, CoreWeave, another GPU cloud service provider, reached an impressive valuation of $7 billion after a $642 million investment from Fidelity Management and Research Co.
Nvidia’s stock has seen significant growth, and AI startups are eagerly seeking access to Nvidia’s high-performance H100 GPUs for large language model training. This desperation led Nat Friedman, a former GitHub CEO, to create a marketplace for GPU clusters, offering access to resources like “32 H100s available from 02/14/2024 to 03/31/2024.”
Moreover, Forbes reported that Friedman and his investment partner, Daniel Gross, have built a supercomputer known as the Andromeda Cluster, featuring over 4,000 GPUs. This resource is offered to portfolio companies at a rate below the market price.
Friedman shared with Forbes his role in assisting AI startups with acquiring GPUs, emphasizing the high demand for these resources.
The conversation about Nvidia GPU access continues against the backdrop of a report from The Wall Street Journal. OpenAI’s CEO, Sam Altman, has proposed reshaping the AI chip market, a venture with significant cost and geopolitical implications.
However, not everyone agrees with this approach. Databricks CEO Ali Ghodsi expressed skepticism about the ongoing “GPU hunger games,” predicting a decrease in AI chip prices and a rebalance of supply and demand within the next year. He compared the situation to the early 2000s concerns about internet bandwidth, suggesting a similar resolution could occur for GPUs, potentially alleviating the current scarcity affecting AI startups.