Andreessen Horowitz has a large cluster of Nvidia H100 GPUs to help its portfolio of AI startups meet their computing needs, the venture capital firm confirmed for the first time on Wednesday. The program, called “Oxygen,” allows portfolio companies to train or operate AI models without negotiating market prices.
A16Z's Oxygen cluster gives startups breathing room to compete with big tech companies like Google, Meta, and Microsoft in building AI models at scale. Over the past few years, these companies have been engaged in bidding wars for GPUs, particularly the AI industry standard H100, often winning deals by promising larger, longer-term contracts. So while many AI startups may not be able to take on the large contracts needed to secure GPUs, the OpenAIs and Anthropics of the world can.
“This started with the realization that the many AI founders we serve every day have a common problem: We are in the midst of a supply shortage, and we are running low on Nvidia H100 capacity. There was a shortage,” said Anjney Midha, an A16Z partner who helped. Create an Oxygen program with a podcast transcript shared with TechCrunch. “As a startup, large cloud was being deprioritized in favor of larger customers. This was very difficult for them.”
This infrastructure is also a good way for A16Z to attract new startups. Companies building AI models often require unfettered access to large computing clusters for short-term training runs, but once the models are trained, they It will disappear. Unless, of course, AI model usage explodes, inference for AI models typically requires less compute than training. The A16Z's Oxygen cluster provides startups with flexibility, allowing them to access GPUs when they need them without long-term contracts with cloud providers or large cash outlays. In return, they are giving A16Z a stake in their business in exchange for (among other things) low-cost GPU rentals.
But A16Z isn't the only investor pledging to provide startup founders with GPUs as equity in their companies, in addition to financial capital and guidance. According to Forbes, investment partners Nat Friedman and Daniel Gross are offering their startups access to a 4,000 GPU cluster called Andromeda Cluster. Y Combinator also provides startups with GPU clusters for training execution through partnerships with various cloud providers, most recently Google Cloud.
A spokesperson for Andreessen Horowitz declined to comment on the size of the Oxygen cluster, but the A16Z may have the venture's largest GPU power. The Information reported in July that Oxygen could have more than 20,000 GPUs.
At one point in the podcast, Midha says Oxygen clusters can ease the pressure on AI startups to raise money at premium valuations to pay for computing. This is another way Oxygen benefits A16Z. Rather than investing more in startups to be able to buy GPUs from Microsoft, for example, A16Z can provide the GPUs in-house and at a lower cost.
A16Z sees Oxygen as a core value proposition for startups going forward, as long as the AI boom continues as expected. Perhaps that's why the startup made the huge investments necessary to secure these chips.