①a16z may have spent hundreds of millions of dollars to purchase thousands of Nvidia H100 GPUs; ②a16z ultimately plans to expand the GPU cluster to more than 0.02 million blocks, a number similar to the number of GPUs used by Musk's xAI; ③Sequoia Capital did not rush to buy GPUs, but instead posted a message on its blog saying that the peak of GPU shortage has passed.
"Star Daily" July 10th News (Editor Zhu Ling) With the rapid development of artificial intelligence technology, the importance of GPUs as core computing resources is becoming more and more prominent. According to the US technology website The Information, venture capital firm Andreessen Horowitz (a16z) may have spent hundreds of millions of dollars to purchase thousands of Nvidia H100 GPUs.
a16z ultimately plans to expand the GPU cluster to more than 0.02 million blocks, a number almost equivalent to the number of GPUs used by Elon Musk's xAI for training Grok large models, costing up to $5 billion. These GPU resources will be used to support start-ups invested by a16z, helping them reduce the cost of artificial intelligence development, attract top talents, and stand out in competition.
a16z's GPU resources are undoubtedly a huge temptation for start-ups that urgently need GPU resources for AI large-scale model development. These start-ups can obtain the right to use Nvidia H100 GPUs by selling equity to a16z. This not only solves the problem of computing resources for start-ups, but also brings more investment opportunities and benefits to a16z.
a16z's plan is called the "oxygen plan", which means that GPU resources are as important to start-ups as oxygen.
Although the problem of GPU shortage has been alleviated for a period of time, it is still very difficult for small start-ups to obtain enough GPUs to meet their needs. Their budgets are limited and they have to compete with large companies, which have much greater demand for GPUs and are more likely to sign contracts with cloud computing vendors.
a16z's GPU resources are undoubtedly a huge temptation for start-ups that urgently need GPU resources for AI large-scale model development. These start-ups can obtain the right to use Nvidia H100 GPUs by selling equity to a16z. This not only solves the problem of computing resources for start-ups, but also brings more investment opportunities and benefits to a16z.
Luma AI is one of the first companies to use a16z GPUs to train models. In January of this year, a16z led a $43 million Series B investment in the California-based start-up.
Luma's co-founder and CEO, Amit Jaan, said that other venture capital firms offered higher valuations to Luma, but he valued a16z's GPU resources more than the higher valuations.
"For AI-based model companies like us, computing power is almost the entire core competitiveness," Jaan said.
It is worth noting that a16z's approach may further exacerbate the resource gap between start-ups, making it even more difficult for start-ups that do not receive a16z support to survive.
In fact, a16z's layout in the GPU field is not limited to purchasing and renting, but also involves in-depth cooperation with chip suppliers. According to Forbes, a16z is negotiating with chip suppliers to build a computing project to further enhance its competitiveness in the field of artificial intelligence.
In addition, a16z's move coincides with its internal restructuring and the critical period of raising new funds. Earlier this year, a16z successfully raised a total of $7.2 billion in new funds, including a fund specifically for supporting start-ups in AI infrastructure, whose technology is the foundation of generative AI.
It is worth mentioning that a16z's bet in the GPU field is not unique. Other venture capital firms, such as Nat Friedman and Daniel Gross, bought Nvidia H100 server chips worth about $100 million last year. Index Ventures also reached an agreement with Oracle to rent GPU servers for some start-ups.
However, as the GPU shortage situation eases, some investors have begun to adjust their strategies. Sequoia Capital did not rush to buy GPUs, but instead posted a message on its blog saying that the peak of GPU shortage has passed.
Sarah Guo's early-stage venture capital firm, Conviction, leased GPUs from cloud service providers last year and resold them to start-ups at cost, but has reduced its order volume this year and put some servers on the market for sale.
