Google Cloud is giving Y Combinator startups access to a dedicated, subsidized cluster of Nvidia graphics processing units and Google tensor processing units to build AI models. It’s part of Google Cloud’s effort to cozy up with promising early-stage AI startups, in hopes some of them will evolve into massive, compute-hungry business.
“We want to surround them with a lot of love and warmth early in their lifecycle so they get familiarized with building and working on the Google Cloud Platform,” said general manager for Google Cloud’s startups and AI business, James Lee, in an interview with TechCrunch. “As they stick around, we grow as they grow, and we become a partner of theirs throughout their lifecycle.”
Specifically, Google will provide a dedicated cluster with priority access for YC’s Summer 2024 startups, along with $350,000 in cloud credits over two years for each participating startup. The idea is that a future unicorn will fall in love with building on Google Cloud. Five percent of Y Combinator startups over the last 18 years have become unicorns with billion-dollar-plus valuations, YC group partner Diana Hu tells TechCrunch, so it’s a decent bet for Google to make on this partnership.
“There’s a lot of excitement that the next generation of startups will probably not just unicorn, but decacorn, and they’re getting built as we speak,” said Hu. “Cloud providers are still catching up in terms of how to price them, but they do know if you catch them early on, you’re going to ride the wave with them.”
On the flip side, Y Combinator says it can attract more AI startups to apply by offering generous compute resources alongside its investment and guidance. For early-stage AI startups, Hu says one of the most common issues she hears is that startups are compute-restrained. Large enterprises are able to strike multi-year, massive deals with cloud providers for GPU access, but small startups are often left out to dry. She says partnerships like these go a long way, especially when you have a dedicated cluster – that’s particularly important for training AI models.
“For GPU and AI workloads, they’re more like high performance computing workloads in batches. You don’t need the server running all the time, but you need a lot of them in a spiky workload,” said Hu. “So we have a dedicated cluster that YC companies can just use.”
As part of the partnership, Google will also offer YC startups $12,000 in Enhanced Support credits and a free year of Google Workspace Business Plus. YC startups can also connect with Google’s internal AI experts through monthly office hours.
Startup accelerators and venture capital firms are offering GPU clusters more and more these days. Andreessen Horowitz reportedly has a stash of 20,000 GPUs to attract AI startups. Google Cloud and Y Combinator would not reveal the exact number in this deal, but Hu says it’s sufficient for YC’s foundation model companies to train models on.