⚡ Key takeaways
- AI start-ups raised about $102B in Q3 2026, 64% of all global venture funding (Crunchbase, reported).
- A record 27 rounds of $1B or more; the largest went to infrastructure and data platforms that sell to the labs.
- CB Insights counts $103.3B for Q3, down 33% from Q2: money is concentrating in fewer, larger rounds.
- Coding agents are the most crowded bet and the most exposed to the labs.
- Robotics, compute and verification are the least exposed; verification has few well-funded independents.
🗺️ The money
Thesis 1: The largest cheques went to companies that sell to the labs.medium confidence
| Round (as reported) | Amount | Theme |
|---|---|---|
| OpenAI | $10B | frontier lab |
| Databricks | $5B | data platform |
| Safe Superintelligence (NVIDIA-led) | $5B | neolab |
| Crusoe | $3.9B | neocloud |
| Cognition | $2B | coding agents |
| River AI (Series A) | $1.1B | applications |
| Sierra (May) | $950M | vertical agents |
| Walden Robotics (seed) | $300M | robotics |
Q3 2026 unless dated. Sources disagree on several figures; we show those reported by Crunchbase or CB Insights (via Lexology) or by more than one outlet.
So what: Infrastructure is where the money is safest from the labs, because the labs are its customers.
📊 Crowded and quiet bets
Thesis 2: Coding agents are crowded; compute, robotics and verification are quiet.medium confidence
🔧 Reproduce this: entrant_map.py
entrant_map.py · reads data/ai-entrants-2026.json
"""entrant_map.py — where new AI companies cluster, and how exposed they are to the labs.
Reads data/ai-entrants-2026.json (a hand-curated, sourced sample, not a census) and
groups companies by layer: how many, how many valuations are reported, and the
average 'lab overlap' (our 0-3 judgement of how directly frontier labs compete).
Standard library only.
"""
import json, os
d = json.load(open(os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "..", "ai-entrants-2026.json")))
by = {}
for x in d["items"]: by.setdefault(x["layer"], []).append(x)
print(f"{'layer':11} {'companies':>9} {'avg lab overlap':>15} examples")
for layer, xs in sorted(by.items(), key=lambda kv: -sum(x["overlap"] for x in kv[1]) / len(kv[1])):
ov = sum(x["overlap"] for x in xs) / len(xs)
print(f"{layer:11} {len(xs):9d} {ov:15.1f} {', '.join(x['name'] for x in xs[:3])}")
hi = [x["name"] for x in d["items"] if x["overlap"] >= 3]; lo = [x["name"] for x in d["items"] if x["overlap"] == 0]
print(f"most exposed to the labs (3/3): {len(hi)}; least exposed (0/3): {len(lo)} -> {', '.join(lo)}")$ python3 entrant_map.py
layer companies avg lab overlap examples
harness 2 3.0 Cognition, Cursor (Anysphere)
models 6 2.7 Thinking Machines Lab, Safe Superintelligence (SSI), Mistral AI
frameworks 2 2.0 LangChain, CrewAI
builders 2 1.5 n8n, Zapier
apps 3 1.3 Harvey, Sierra, River AI
data 1 1.0 Databricks
runtimes 1 1.0 Temporal
compute 4 0.0 Crusoe, CoreWeave, Groq
physical 2 0.0 Unitree Robotics, Walden Robotics
most exposed to the labs (3/3): 6; least exposed (0/3): 6 -> Crusoe, CoreWeave, Unitree Robotics, Walden Robotics, Groq, Cerebras
Average lab overlap by layer (0 = labs are customers, 3 = head-on)
| Harness (coding agents) | |
| Models (neolabs, open labs) | |
| Frameworks | |
| Builders | |
| Applications | |
| Compute & robotics |
Real script output on a 23-company sourced sample (not a census); overlap scores are our judgement.
So what: A start-up's risk is less about its funding than about whether the next lab release does its job.
🧭 The wrapper test
- Would the next model release make this company unnecessary?
- High risk: thin prompts and interfaces over a model the lab also sells.
- Lower risk: value in data, workflow, distribution, hardware or trust the lab does not have.
📄 Method and sources
- Totals and rounds as reported by Crunchbase and CB Insights (via Lexology) and press; secondary ⚠️.
- Entrant sample:
data/ai-entrants-2026.json, 23 companies, each with a source. - Lab-overlap scores are our judgement.
Facts are sourced; figures marked ⚠️ come from secondary reports or company statements. Scores, weights and predictions are our judgement and are labelled. Disclosure: Claude, made by Anthropic, helps write this site. Anthropic is one of the companies analysed here, under the same rules as the others.
🧭 Foundations: the AI market, Nov 2026
Live data: AI Market Map & Investments · AI Investment Tracker.
Read
Funding & new entrants
You are here
Read
Read