← Back to feed

Baseten is close to finalising a $1.5 billion funding round at a valuation the Wall Street Journal reports at up to $13 billion. The round is co-led by Spark Capital, Sands Capital, Altimeter Capital, and Wellington Management.

The numbers are striking. Five months ago, Baseten raised $300 million at a $5 billion valuation in its Series E. That round came nine months after a $150 million Series D. The new raise would represent a 160% valuation increase in under half a year.

The Deal Structure

This is a split-priced round. Some investors are entering at $13 billion; others at $11 billion. Split pricing lets lead investors lock in a headline valuation for reporting purposes while more cautious participants come in at a discount. The technique is now common in 2026 among AI startups where investor demand consistently exceeds the founders’ willingness to commit to a single price. The $1.5 billion raise itself ranks among the year’s ten largest private AI rounds.

What Baseten Does

Baseten is an inference routing company. Its core product handles what happens after a user submits a prompt: route the request to the model best suited for the task, favouring capable but cheaper open-source alternatives where they match closed frontier quality on that specific request type.

The market problem it addresses is structural. Frontier model prices have fallen 97% in three years. Open-weight alternatives now match closed models on a growing share of tasks. The practical challenge has shifted from accessing capable models to routing efficiently across a fragmented market of 400-plus options. Baseten’s position: model selection, caching, and request routing are worth building as dedicated infrastructure rather than solved case-by-case inside each application.

Why the Valuation Jump

Three factors explain the acceleration in investor conviction.

Token volumes are up. Goldman Sachs projects agent token use multiplying 24x by 2030. The models processing those tokens will increasingly be routed rather than directly called. Infrastructure sitting in that routing layer captures a margin from every transaction.

Pricing divergence has widened the opportunity. The gap between GPT-5.5 at $15/M input tokens and Qwen3.6-35B at $0.14/M input, for tasks where the cheaper model performs equivalently, is now large enough to fund an entire inference routing company from savings on a single enterprise deployment.

Inference infrastructure is consolidating. CoreWeave closed an $8.5 billion investment-grade GPU loan. Groq raised $650 million after its founding team departed. KogAI claimed 3,000 tokens per second per user on standard hardware. Each of these signals a market that is moving from experimentation to production scale, where routing decisions matter more than raw GPU capacity.

Key Numbers

• New round size: ~$1.5B (split-priced: $11B-$13B per tranche)
• Previous round: $300M at $5B, completed January 2026
• Valuation step-up: 120-160% in five months
• Co-leads: Spark Capital, Sands Capital, Altimeter Capital, Wellington Management
• Total funding to date (pre-close): ~$450M across Series D and E
• Founded: 2019