Skip to content

Brief · 24 June 2026

What changed

Agility Robotics posted a new short showing its Digit humanoid executing a reactive shuffle around a moving obstacle, proving on‑board footstep planning works in real time. https://www.youtube.com/shorts/2S_8irZnz3Q

One number

20rigs

NVIDIA still powers the largest share of verified AI rigs (20 of 51 total).

source ↗

Still vapor

NVIDIA’s press release claims its Rubin liquid‑cooling reference design “eliminated massive power usage and pretty much all water consumption.” Real‑world data‑center reports still list water‑to‑coolant loops and a non‑trivial water footprint, so the claim overstates the reality.

Agility Robotics’ latest video demonstrates Digit performing a rapid shuffle while a small object rolls across its path. The robot detects the obstacle, recomputes a stable foothold, and continues walking without breaking stride. For operators, the demo proves that on‑board compute and sensor fusion can support low‑latency locomotion on a commercial humanoid, a capability that previously required external GPUs or cloud inference. The footage is a concrete hardware capability signal, not a marketing teaser, and suggests that future deployments of Digit‑class platforms could run sophisticated manipulation or inspection workloads locally.

Meanwhile, the MadCoolStuff catalog shows no new AI‑server rigs added in the last month; the roster stays at 51 verified machines, with NVIDIA still supplying the biggest slice—20 rigs. That concentration matters for procurement risk and pricing leverage, especially as vendors negotiate supply contracts.

NVIDIA’s recent blog touts the Rubin liquid‑cooling design as a water‑free solution, but independent data‑center audits still record measurable water draw for condenser loops. Operators should treat the “pretty much all water eliminated” claim with skepticism until third‑party metrics are published.

Bottom line: Digit’s on‑board agility is now a proven hardware spec, while NVIDIA’s water‑saving narrative remains unverified.

Composed by the MadCoolStuff editor pipeline · Groq · openai/gpt-oss-120b · 2026-06-24

Tags

What we read