Baseten raises $1.5B at $13B. The 40x volume number is why.

Good morning 👋 A $13B inference infrastructure round closed Monday. The number buried inside it is more interesting than the valuation.

In today's issue:

• 🔭 Baseten closes $1.5B — what the 40x inference volume growth actually signals
• 🧠 Krea 2 goes open weights: a 12B diffusion transformer in two checkpoints
• 🔬 Three papers on eval grading, agentic commerce rails, and agent safety certification
• 🚀 One RL agent per user: MoEngage acquires Aampe
• 🔗 Agility Robotics SPAC, Menlo's $3B AI funds

🔭 THE ONE THING

⚡ Baseten raises $1.5B at $13B. The 40x volume number is why.

Baseten closed a $1.5 billion Series F at a valuation of up to $13 billion, led by Altimeter Capital, Conviction Partners, and Spark Capital. Fourth raise in 18 months. Revenue up 20x year-over-year, inference volume up 40x. The volume figure is the real signal: inference demand at the custom-model layer scaled faster than anyone modeled, and Baseten is sitting on it. The risk now is hyperscalers: AWS, Google, and Azure are all moving into inference primitives, and the moat narrows as they get closer.

🧠 MODELS & RELEASES

• 🖼️ Krea AI released Krea 2 as open weights, a 12B dense diffusion transformer that ships in two checkpoints: Raw (base, for fine-tuning and LoRAs) and Turbo (8-step distilled, built for fast inference). LoRAs trained on Raw transfer to Turbo. Train on the base, serve the fast version. Weights on Hugging Face; connectors for ComfyUI, Replicate, Together AI, and the major clouds. Krea 2

🔬 RESEARCH HIGHLIGHTS

• Grading the Grader — evaluating an agentic data analysis system is not like scoring a quiz. The key distinction the paper makes: a "diagnostic disagreement" (different explanation, same correct underlying finding) looks identical to a real error if your grader can't separate them. Anyone shipping automated evals for agents should read the taxonomy before they trust their numbers. arXiv 2606.24839

• Paying to Know: agent-native micropayments in e-commerce — the core claim is that micropayment rails (x402, AP2) break the shopping-chatbot model. When the buyer is an autonomous agent that can independently verify specs, recommendations become cheap and abundant. What gets scarce is verified product data, and that's what commands a price. Optimistic on infrastructure, but the framing of "verified data over recommendations" is the right lens for agentic commerce monetization regardless of which rail wins. arXiv 2606.24783

• World Models in Pieces: structural certification for agents — formal argument that real-world agents are inevitably domain-specialized. The problem with current safety certification: a failure on a critical bottleneck looks identical to a failure on something irrelevant. The paper proposes structural certification that maps which failures actually matter. Useful framing for "is this agent safe to deploy" versus "did it pass the benchmark." arXiv 2606.24842

🚀 AI STARTUPS

• MoEngage acquires Aampe, a per-user agentic decisioning platform running 200B+ decisions a week. One RL agent per end-user, not per segment. Agents compound learning across campaigns using Thompson Sampling and causal modeling rather than resetting each time. Terms undisclosed. link

🔗 QUICK LINKS

• Agility Robotics, maker of the Digit humanoid, is going public via SPAC at a $2.5B valuation, with Foxconn anchoring the PIPE and NVIDIA and Amazon among existing backers.
• Menlo Ventures raised $3B across two new AI funds on its 50th anniversary, pointing to its Anthropic lead bet and 60+ Anthology portfolio companies as the track record behind the ask.

Heavy on infrastructure today. Back tomorrow.

Pradeep Perugu