The Practical AI Showwith Olga Pechnenko and Chris Pearson
Week 30

Weekly AI Funding Report, Week 30

The Money Went To The Plumbing

On paper, AI funding just got cut in half. It didn't.

Covered on Episode 47 of the Practical AI Show. Source: Crunchbase, AI companies only.

AI funding this week
$4.68B
Share of all venture $
41.1%
AI companies funded
77
Biggest round
$1.5BBaseten
Tracked so far
$383.7Bacross 30 weeks

The week in brief

AI raised $4.68B across 77 companies, 41.1% of all venture dollars. Last week's $11.64B looked twice as big, but 76% of it was one China wave. This week China goes quiet and the money comes home: US AI funding nearly tripled, $1.34B → $3.74B, four of every five AI dollars. And the biggest checks all went to the same place, the infrastructure that runs AI. Baseten ($1.5B) and Groq ($650M) are both about running models, not building them. The app is the show. The money's going to the stage crew.

Of 201 funded companies this week, 77 (38.3%) were AI, taking $4.68B of the $11.37B total raised (41.1%). The honest read: it looks like AI funding fell off a cliff from last week's $11.64B, but that was a China week, $8.9B of it. Strip the geography and the opposite is true: US AI funding nearly tripled, from $1.34B to $3.74B. The drop is an illusion. The US engine surged.

Where the money went

  • United States$3.74B

    Owned both the dollars and the deal count this week. Four of every five AI dollars.

  • China$329M

    Carried by LibLib AI ($300M). Nearly absent after dominating last week.

  • Middle East & Africa$272M

    All Israel, Dream's $260M sovereign-AI round is nearly the whole bar.

  • Europe$230M

    The most companies after the US, but small checks. Led by Multiverse Computing ($122M, Spain).

  • Rest of world$65M

    (incl. Canada).

  • Asia-Pacific (ex$China) — $36M

    Lots of small rounds. Latin America: $0.

The top five AI rounds

  1. Baseten $1.5B

    Series F · San Francisco · $13B valuation · AI inference

    The plumbing that runs AI models inside other companies' products, fast and in real time, so they don't build their own data-center stack. The biggest check of the week is for running models, not training them. They raised it just months after their last mega-round.

  2. Groq $650M

    Venture round · San Jose · AI inference chips

    Designs its own chips to run AI models fast and cheap, an alternative to Nvidia's GPUs. The juicy part: this lands right after a $20B deal with Nvidia and a re-staffing.

  3. LibLib AI $300M

    Series B · Beijing · $2B+ valuation · AI image generation

    China's AI art platform, create and share AI images and models. The week's biggest non-US round, and proof China's consumer-AI scene stays well funded even in a quiet China week.

    Press-confirmed (Yicai, 36Kr), no official company release.

  4. Dream $260M

    Series C · Tel Aviv · $3B valuation · sovereign AI + cyber defense

    Sells AI that helps governments defend critical systems, power grids, national infrastructure, and turn sensitive data into fast decisions. A theme to watch: nations buying their own AI defense.

  5. Mirendil $200M

    Seed · United States · ~$1B valuation · AI research lab

    Ex-Anthropic and Google researchers building AI tools that help scientists create their own models. The anomaly of the week: a seed round of $200M at a billion-dollar valuation, a16z, Kleiner Perkins, AND Nvidia, all in before there's a product. Seed used to mean a few million. Now it can mean you already won.

    Press-confirmed (WSJ), no official release. Location not in Crunchbase; verified US lab.

Also notable: Upscale AI $190M (AI-data-center networking), Lightwheel $147M (robot simulation), Assort Health $120M (AI voice agents for clinics), Runpod $100M (GPU cloud).

  • The money is in the plumbing, not the apps. Baseten, Groq, Upscale AI, Runpod, Lightwheel, the biggest checks went to the infrastructure that runs AI. Picks-and-shovels are eating the funding.
  • Inference beat training. Both top US rounds are about running models cheaply, not building new ones. The market shifted from "build the model" to "serve it to millions."
  • Wide at the bottom, heavy at the top. 40 seed + 19 Series A rounds (59 of 77 AI deals), lots of new bets. But the dollars pool in a few late-stage infra rounds.
  • Seeds are getting absurd. Mirendil's $200M seed and Trase's $107M seed. "Seed" no longer means small.
  • US concentration is back. After a China-dominated week, 80% of AI dollars came home. The US AI engine nearly tripled.