AI Startups Tracker
Week 0 baselines for 8 AI startups — Embodied, Infra, Dev Tools — plus append-only weekly signal diffs across blog, GitHub, careers, and press. Built to compound: today's entry is tomorrow's input.
8 companies · baseline 2026-04-20 · next checkpoint Week 12
01 EMBODIED AI
4 companies1X Technologies
Humanoid = hardware+data flywheel: ship early (teleop-assisted) → collect interaction data → train world model → close autonomy loop. Neural-drive tendon actuation + soft body =…
Figure AI
Vertical-integration bet: owning hardware + VLA model + factory is required to ship humanoids at consumer scale. Feb 2025 OpenAI split formalized this — Adcock: "we can't…
Physical Intelligence
One generalist policy can control many robots across many tasks — bet that a VLA + large cross-embodiment data + RL will reach an LLM-style capability inflection. π0.7's…
Skild AI
Robotics data is the bottleneck, not hardware. A single unified policy trained on internet-scale human video + sim can generalize across embodiments, mirroring the LLM "one model,…
02 AI INFRASTRUCTURE
2 companiesModal
Bet: the serverless-GPU layer becomes the default abstraction for AI-native developer teams who refuse to manage K8s/GPU fleets. Modal wins on DX + cold-start perf while…
Together AI
"AI-native cloud" — verticalize the stack (kernels → schedulers → inference runtime → datasets) so open-model inference beats hyperscaler price/perf, and capture enterprises…
03 AI DEV TOOLS
2 companiesCursor (Anysphere)
Own the AI-native coding surface by (a) controlling the IDE layer (VS Code fork, lock-in via agents/rules/MCPs), (b) vertically integrating proprietary models (Composer +…
Replit
Agent pivot collapsed 9 years of IDE grind into a category-leading AI app-builder. $150M ARR Sept 2025, $240M FY2025 revenue, Amjad public target $1B ARR end-2026 (S-8-005). Wedge…