Baseten Raises $1.5B at a $13B Valuation Amid the Inference Explosion

Inference startup Baseten is nearing a $1.5 billion round valued at $13 billion, reports TechCrunch. The round is done on a “split price” basis, meaning that some investors will come in at a $11 billion valuation. The round is led by Spark Capital, Sands Capital, Altimeter Capital and Wellington Management. This is unconfirmed, but the trend is notable.

The key here is the speed. Baseten raised a $300 million Series E at a $5 billion valuation five months ago, and a $150 million round nine months prior. Assuming these figures are accurate, Baseten’s valuation increased by around 160% in less than six months. Baseten was founded in 2019 and has moved from an obscure infrastructure company to one of the most heavily funded players in the AI plumbing stack.

1What Baseten actually does

Baseten does not build large models themselves. Instead, they run other people’s models, aka inference, which occurs after the prompt is sent. Baseten’s value proposition is that they can route requests to the right model for the job, and more specifically direct traffic away from expensive cutting edge models and towards cheaper, but still competent, open source models. The benefits are time and money saved, as they can provide the same results at a reduced cost and latency.

This is what makes them attractive to investors. While most of the AI hype has been focused on training models, which requires building larger and larger models, serving those models to millions of users is an ongoing expense. The “inference gold rush” refers to the race to fund the plumbing of AI, and Baseten’s recent rounds are a perfect illustration of this.

However, we should temper our excitement with a little bit of skepticism. A 160% increase in valuation over a few months, especially when done on a split price round designed to make the top line look better, is the sort of maneuver that works great when the market is going up, but could prove fragile if the enthusiasm for AI spending wanes. There is certainly a need for cheap, fast inference, but rapidly increasing valuations remind us that the AI infrastructure market is betting on a future that hasn’t arrived yet.