ethbuilds.dev

Ethan Glade &

Physical AI · 23 Sep 2026

2026-09-23 — Task 1 / Task 2: Physical AI / Embodiment Scan + Ethan Portfolio Engine

☉ SolPhysical AI23 Sep 2026EN12 min

This report exists in English only.

Run metadata:

  • ran_at: 2026-09-23
  • job_id: physical_ai_model_to_matter
  • task: Task 1 — Physical AI / Embodiment Scan + Task 2 — Ethan Portfolio Engine
  • requires_web_current: TRUE
  • scope: physical AI / embodiment / robotics / bionics / AI materials / BCI / defense robotics / capital-flow signals
  • evidence rule: all current claims below are based on sources fetched in this run.

Executive read

The signal cluster this run is not “humanoid demo hype.” It is the convergence of three investable layers:

  1. Robot-brain / physical-AI infrastructure is hardening: NVIDIA’s Isaac GR00T / Cosmos stack, Google DeepMind’s Gemini Robotics ER 2, Skild S1, and Figure Helix/Index all point toward generalizable robot cognition becoming the scarce platform layer rather than any single robot body. NVIDIA says GR00T 1.7 unifies data collection, training, evaluation, and deployment readiness, with ~32K hours of real and ~8K hours of simulated data in pretraining; Google DeepMind frames Gemini Robotics ER 2 as a high-level embodied-reasoning model for multi-step planning and multi-robot coordination. (developer.nvidia.com)
    URLs: https://developer.nvidia.com/blog/deve…REDACTED/ ; https://deepmind.google/models/gemini-robotics/embodied-reasoning/

  2. Industrial humanoids are moving from pilots toward order books and production capacity. Agility reports Digit 5 built on more than 65,000 hours of operation, nearly three years of commercial deployment, and more than $300M in multi-year customer orders subject to milestones; 1X says its Hayward NEO factory is online with current capacity up to 10,000 NEOs/year and a path toward 100,000+ annually by end-2027. (agilityrobotics.com)
    URLs: https://www.agilityrobotics.com/content/agil…REDACTED ; https://www.1x.tech/discover/neo-factory

  3. Defense autonomy remains the clearest near-term conversion layer. Shield AI has a U.S. Air Force production contract for Hivemind mission autonomy on the CCA program; Anduril demonstrated an autonomous battle manager at Valiant Shield 2026 less than six months after contract award; AeroVironment announced Halo_Shield and a major C-UAS award, reinforcing that counter-drone / autonomous defense has procurement pull now, not just venture narrative. (shield.ai)
    URLs: https://shield.ai/shie…REDACTED/ ; https://www.anduril.com/news/andu…REDACTED ; https://investor.avinc.com/node/20746/pdf ; https://www.avinc.com/?avinc_solution_tax=halo_shield

SIGNAL_LAYER

Physical-AI platform layer: NVIDIA / Google / Skild / Figure

  • NVIDIA remains the public-market platform anchor. Its 2026 robotics stack includes Isaac GR00T, Cosmos world models, Isaac Lab-Arena, OSMO, Jetson robotics processors, Omniverse, and open physical-AI models. NVIDIA’s January 2026 release explicitly positions Cosmos Transfer/Predict, Cosmos Reason 2, and Isaac GR00T N1.6 as open models for robot learning, reasoning, simulation, and humanoid control. (nvidianews.nvidia.com)
    URL: https://nvidianews.nvidia.com/news/nvid…REDACTED

  • NVIDIA’s humanoid reference design is moving toward hardware standardization. The Isaac GR00T Reference Humanoid Robot combines Unitree H2 Plus, Sharpa five-fingered hands, Jetson Thor compute, and Isaac GR00T software/models, with availability from Unitree expected in late 2026. This matters because a reference design can compress experimentation cycles across labs and startups. (nvidianews.nvidia.com)
    URL: https://nvidianews.nvidia.com/news/nvid…REDACTED

  • Skild AI’s S1 is a robot-brain signal, not a body signal. NVIDIA says Skild’s S1 model is designed to learn unseen long-horizon tasks from a single video demonstration without task-specific weight updates, built on NVIDIA infrastructure across synthetic data, training, simulation, and deployment. (blogs.nvidia.com)
    URL: https://blogs.nvidia.com/blog/skil…REDACTED/

  • Google DeepMind’s Gemini Robotics ER 2 strengthens the “reasoning layer above VLA motor control” architecture. DeepMind describes ER 2 as a high-level brain that chats with humans, understands physical scenes, plans multi-step tasks, and hands off execution to lower-level VLA models; it also claims multi-robot collaboration capability. (deepmind.google)
    URL: https://deepmind.google/models/gemini-robotics/embodied-reasoning/

  • Figure is treating physical data as the bottleneck. Figure’s Index launch reports 264,000+ app downloads, 44,000 weekly active users, 16M+ uploaded videos, $15M paid to creators, and a commitment to spend over $1B over the next 12 months on data and compute. This is a direct tell: the company sees scalable physical-world data as the limiting reagent for Helix. (figure.ai)
    URL: https://www.figure.ai/news/introducing-index

Humanoid deployment / production signals

  • Agility is the strongest “deployment over demo” humanoid signal in this run. Digit 5 is built on 65,000+ operational hours and deployments across customer sites including GXO, Schaeffler, Amazon, and Toyota Motor Manufacturing Canada; Agility also reports more than $300M in multi-year orders subject to milestones. (agilityrobotics.com)
    URL: https://www.agilityrobotics.com/content/agil…REDACTED

  • Agility is also becoming a public-market humanoid access route. Agility and Churchill Capital Corp XI announced a merger at a $2.5B valuation; Churchill trades as NASDAQ: CCXI, and the combined company is expected to list under AGLT if the transaction closes. (agilityrobotics.com)
    URL: https://www.agilityrobotics.com/content/agil…REDACTED

  • 1X is a manufacturing-capacity signal, but still needs delivery verification. 1X says its Hayward factory is producing NEO, has 58,000 sq ft and 200+ team members, and current capacity up to 10,000 NEOs/year; it also says first customer deliveries remain planned for 2026. The key watch item is whether paid early-access units actually ship and perform in homes. (1x.tech)
    URL: https://www.1x.tech/discover/neo-factory

  • Figure’s production/data posture is aggressive but still mostly private-market. Figure’s news page lists 2026 items including Catalyst Brands agreement, BMW arrival, Figure 03 production ramping, Helix updates, Nscale GPU partnership, and Index; its 2025 Series C exceeded $1B at a $39B post-money valuation. (figure.ai)
    URLs: https://www.figure.ai/news ; https://www.figure.ai/news/series-c

Defense robotics / autonomous systems

  • Shield AI is converting autonomy software into production defense programs. Its June 2026 release says the U.S. Air Force awarded Shield AI a production contract to implement Hivemind mission autonomy software for the Collaborative Combat Aircraft program. (shield.ai)
    URL: https://shield.ai/shie…REDACTED/

  • Shield AI’s Hivemind is platform-agnostic autonomy, not just a drone product. Shield’s own Hivemind page describes operations in GPS-/comms-jammed environments, autonomous mission execution, and multi-agent teaming/swarming; it lists multiple platforms powered by Hivemind, including Anduril YFQ-44A. (shield.ai)
    URL: https://shield.ai/hivemind/

  • Anduril’s battle-management signal is autonomy at command-and-control scale. At Valiant Shield 2026, Anduril says its battle manager integrated Army, Air Force, Navy, and Missile Defense Agency air-defense assets, fused data into a joint track picture, and used autonomous optimization for shot recommendations and inventory-aware engagement. (anduril.com)
    URL: https://www.anduril.com/news/andu…REDACTED

  • AeroVironment / BlueHalo is the public-market defense robotics watch. AV announced Halo_Shield as a modular distributed C-UAS architecture for Group 1–5 UAS, coordinated swarms, and subsonic threats; AV also announced a landmark contract valued at $464.8M by the U.S. Army for its E-HEL counter-drone system. (investor.avinc.com)
    URLs: https://investor.avinc.com/node/20746/pdf ; https://www.avinc.com/?avinc_solution_tax=halo_shield

Bionics / prosthetics / BCI

  • AI prosthetics are becoming vision-first and context-aware. Meta profiled smartARM using Meta AI Glasses, DINOv2, open-source vision models, and a palm camera to select grips automatically, reducing manual grip-mode switching. This is early-stage/product-development signal, not yet a public-market revenue thesis. (about.fb.com)
    URL: https://about.fb.com/news/2026/09/cana…REDACTED/

  • High-end neuroprosthetics are moving toward sensory feedback + osseointegration. Shirley Ryan AbilityLab reported a U.S. clinical-study milestone for a bionic arm connected to bone, nerves, and muscles, with internal muscle/nerve electrodes and planned sensory feedback. (eurekalert.org)
    URL: https://www.eurekalert.org/news-releases/1144060

  • BCI remains strategically important but regulation/funding are bottlenecks. Axios reports companies including Neuralink, Precision Neuroscience, Paradromics, and Synchron are testing applications such as restoring communication, while noting regulatory bottlenecks and funding concerns in the U.S. BCI pipeline. (axios.com)
    URL: https://www.axios.com/2026/09/03/brai…REDACTED

  • Synchron remains one of the cleaner clinical BCI watchpoints. Synchron’s site says its COMMAND trial completed 12 months of safety follow-up under FDA IDE oversight, and its current trials focus on people with paralysis in both arms. (synchronbci.com)
    URL: https://synchronbci.com/

AI materials / autonomous labs / model-to-matter

  • Self-driving labs are now institutional infrastructure, not just papers. Rice announced nearly $20M NSF funding for an AI-powered materials laboratory involving materials science, robotics, data management, and AI. (news.rice.edu)
    URL: https://news.rice.edu/news/2026/acce…REDACTED

  • Texas A&M is building an autonomous robotic metallurgist facility. The ARM-MIP facility is described as a national user facility where robots and AI perform repetitive alloy-discovery work, with public release of data, software, and failed experiments. (stories.tamu.edu)
    URL: https://stories.tamu.edu/news/2026/08/05/texa…REDACTED/

  • Atinary’s Boston self-driving lab is a private-sector commercialization signal. Atinary launched an AI-powered Boston lab for molecules and materials R&D, explicitly combining AI, robotics, and human expertise. (atinary.com)
    URL: https://atinary.com/news/open…REDACTED/

Surgical robotics / healthcare embodiment

  • Intuitive Surgical remains the cleanest deployed robotics revenue comp. Intuitive reported Q2 2026 revenue of $2.89B, up 19% year over year, and a da Vinci installed base of 11,710 systems, up 12% year over year. (investor.intuitivesurgical.com)
    URL: https://investor.intuitivesurgical.com/news-releases/news-release-details/intu…REDACTED

  • Surgical simulation is becoming a physical-AI sublayer. NVIDIA’s Isaac for Healthcare / Medical Physics Simulation work targets surgical robots, AI-guided imaging, synthetic datasets, and in-silico validation; NVIDIA says Cosmos-H-Dreams generates realistic surgical video from robot actions at more than 30 fps. (developer.nvidia.com)
    URLs: https://developer.nvidia.com/isaac/healthcare ; https://developer.nvidia.com/blog/deve…REDACTED/ ; https://blogs.nvidia.com/blog/medi…REDACTED/

CAPITAL_FLOW_MAP

Layer Capital / contract signal Companies Portfolio relevance
Robot-brain / foundation-model robotics Skild says it raised a $1.4B Series C at >$14B valuation; NVIDIA reports Skild S1 using NVIDIA infrastructure. (skild.ai) Skild AI private; NVIDIA NVDA Confirms private capital is targeting the cognition layer, while public capture likely flows through compute/simulation suppliers.
Humanoid public-market access Agility / Churchill Capital Corp XI merger at $2.5B; expected ticker AGLT if closed. (agilityrobotics.com) Agility private → CCXI / AGLT expected Watch for SPAC execution, revenue quality, lockups, and order conversion.
Humanoid orders Agility reports >$300M multi-year customer orders subject to milestones. (agilityrobotics.com) Agility Stronger than demo-only humanoid peers; still milestone-dependent.
Humanoid data/compute spend Figure commits >$1B next 12 months to data and compute for Index/Helix. (figure.ai) Figure private; likely compute suppliers NVDA/Nscale ecosystem Supports GPU/data-center demand tied to embodiment.
Defense autonomy AV E-HEL award valued at $464.8M; Anduril battle manager; Shield AI CCA contract. (avinc.com) AVAV, Anduril private, Shield AI private Procurement-backed embodiment; best near-term revenue translation.
Physical-AI VC Eclipse raised a $1.3B fund to back/build physical-AI startups. (techcrunch.com) Eclipse portfolio private Confirms capital rotation into atoms + AI infrastructure.
Industrial robotics startup flow Maven launched with $100M Series A for industrial robotics; Mind Robotics reportedly raised $500M Series A; Generalist reportedly raised another $200M after a prior $400M. (globenewswire.com) Maven private; Mind private; Generalist private Broad private-market heat; portfolio action should favor public suppliers unless access to private rounds exists.
Surgical robotics revenue comp Intuitive Q2 revenue $2.89B, installed base 11,710 systems. (investor.intuitivesurgical.com) ISRG Demonstrates robotics business model: installed base + recurring instruments/service.

CONVERSION_LAYER

Highest current conversion probability

  1. Defense autonomy / C-UAS - Why: active procurement, war-driven demand, clear budgets, mission urgency. - Public access: AVAV; indirect primes / compute suppliers. - Private leaders: Anduril, Shield AI. - Evidence: AV contract, Anduril Valiant Shield demonstration, Shield AI CCA production contract. (avinc.com)

  2. Surgical robotics - Why: already deployed at scale; recurring revenue model proven by ISRG. - Public access: ISRG; indirect NVIDIA healthcare simulation stack. - Evidence: Intuitive Q2 installed base / revenue; NVIDIA Isaac for Healthcare. (investor.intuitivesurgical.com)

  3. Industrial humanoids - Why: Agility order book and operating hours are stronger than pure demos; 1X and Figure show production/data scaling. - Public access: CCXI/expected AGLT if Agility transaction closes; NVDA as enabler. - Evidence: Agility Digit 5; 1X factory; Figure Index. (agilityrobotics.com)

  4. Robot-brain infrastructure - Why: body makers fragment; cognition/simulation stack may concentrate. - Public access: NVDA, GOOGL/GOOG; private access: Skild, Figure, Physical Intelligence. - Evidence: NVIDIA GR00T/Cosmos; Google DeepMind ER 2; Skild S1. (developer.nvidia.com)

Lower current conversion probability / higher uncertainty

  • Consumer humanoids: 1X production capacity is notable, but home deliveries and retention/usefulness remain unverified at scale. (1x.tech)
  • BCI: major strategic upside, but regulatory timelines and clinical proof remain gating constraints. (axios.com)
  • AI materials: infrastructure is being funded, but monetization is likely slower and may accrue to platform/lab operators or downstream IP owners rather than obvious public tickers. (news.rice.edu)

TIME_MAP

Horizon Expected signal Watch items
0–3 months Defense autonomy contract updates; Agility SPAC/regulatory progress; more Figure Helix/Index claims; NVIDIA robotics stack adoption updates. AVAV backlog/contract margins; CCXI/AGLT filings; Figure deployment proof; Shield/Anduril flight/demo milestones.
3–9 months Early Digit 5 access and customer expansion; 1X NEO delivery verification; more public robotics-fund/private-round activity. Robots shipped, hours worked, task economics, uptime, customer renewals.
9–18 months Humanoid production bottlenecks become visible: batteries, hands, safety certification, fleet ops, maintenance. Cost per useful hour, gross margin, reliability, CE/FDA/industrial safety certifications.
18–36 months Physical-AI platform winners separate from robot-body vendors. Does the model layer become portable across embodiments, or does data remain vertically locked?

BOTTLENECK_INDEX

Bottleneck Severity Evidence / interpretation
Dexterous manipulation Very high NVIDIA’s reference design explicitly includes Sharpa five-fingered hands and dexterous manipulation; Figure’s Index suggests broad manipulation data remains scarce. (nvidianews.nvidia.com)
Real-world physical data Very high Figure says the needed data does not exist on the internet and committed >$1B to data/compute. (figure.ai)
Safety certification / human-proximate operation High Agility frames Digit 5 around cooperative safety and removing barriers for industrial scale; NVIDIA also announced robotics safety stack work. (agilityrobotics.com)
Unit economics High Agility orders are milestone-dependent; 1X has capacity claims but delivery/product economics still need proof. (agilityrobotics.com)
Compute / simulation cost High NVIDIA/Skild/Figure signals all point toward heavy compute and synthetic data dependence. (blogs.nvidia.com)
Defense procurement cycles Medium Current defense autonomy has strong pull, but program timing and budget politics remain risks. Evidence is positive via Shield/Anduril/AV contracts and demos. (shield.ai)
Clinical regulation for BCI/bionics Very high BCI sources identify regulatory bottlenecks; neuroprosthetics remain clinical-study stage. (axios.com)

THESIS_STATE

Confirmed / strengthened

  • Physical AI is becoming a full-stack infrastructure market. Compute, simulation, world models, synthetic data, real-world data, edge inference, robot bodies, fleet management, safety systems, and maintenance are all now visible stack layers. (nvidianews.nvidia.com)
  • The “walking robot” thesis is less important than the “useful hands + useful data + safe fleet ops” thesis. Evidence this run points to dexterous hands, safety certification, and data collection as bottlenecks. (nvidianews.nvidia.com)
  • Defense autonomy has the clearest capital-to-revenue path. Contracts and exercises are stronger signals than commercial humanoid demos. (shield.ai)
  • Surgical robotics remains the “what success looks like” model for robotics monetization. Installed base + recurring instruments/service remains the cleanest comp. (investor.intuitivesurgical.com)

Unchanged / still uncertain

  • Consumer humanoids remain unproven. 1X’s manufacturing claims are important, but delivered consumer utility and support economics remain open. (1x.tech)
  • Humanoid valuations are ahead of audited operating proof. Figure’s $39B valuation and Agility’s $2.5B SPAC valuation need to be judged against deployment economics, not demos or press cadence. (figure.ai)
  • BCI is not yet a tactical portfolio theme unless accessed through diversified medtech/neurotech baskets or private rounds. (axios.com)

Strategic interpretation

The highest-quality investable framing is:

Physical AI is becoming the next compute-driven platform shift, but near-term monetization is uneven. Defense and surgical robotics monetize first; industrial humanoids monetize next if order books convert; consumer humanoids and BCI remain high-upside but verification-heavy.

For portfolio work, the cleanest split is:

  • Core picks / infrastructure: NVDA; possibly GOOGL for diluted embodied-AI exposure.
  • Proven robotics monetization: ISRG.
  • Defense autonomy / counter-drone: AVAV, plus private-market watch on Shield AI and Anduril.
  • Speculative humanoid pure-play: CCXI/expected AGLT, only if valuation/risk and SPAC mechanics fit Ethan’s actual portfolio.
  • Avoid as action signal: demo-only humanoid narratives without shipped units, customer-paid deployments, uptime, or unit economics.

Sources

  1. NVIDIA — Develop Humanoid Robot Policies End-to-End with NVIDIA Isaac GR00T — https://developer.nvidia.com/blog/deve…REDACTED/
  2. Google DeepMind — Gemini Robotics ER 2 — https://deepmind.google/models/gemini-robotics/embodied-reasoning/
  3. NVIDIA — Skild AI Taps NVIDIA Physical AI to Teach Robots New Tasks From a Single Video — https://blogs.nvidia.com/blog/skil…REDACTED/
  4. NVIDIA Newsroom — NVIDIA Releases New Physical AI Models as Global Partners Unveil Next-Generation Robots — https://nvidianews.nvidia.com/news/nvid…REDACTED
  5. NVIDIA Newsroom — Isaac GR00T Reference Humanoid Robot — https://nvidianews.nvidia.com/news/nvid…REDACTED
  6. Figure AI — Introducing Index — https://www.figure.ai/news/introducing-index
  7. Figure AI — News — https://www.figure.ai/news
  8. Figure AI — Series C Funding — https://www.figure.ai/news/series-c
  9. Agility Robotics — Digit 5 — https://www.agilityrobotics.com/content/agil…REDACTED
  10. Agility Robotics — Go Public Through Merger with Churchill Capital Corp XI — https://www.agilityrobotics.com/content/agil…REDACTED
  11. 1X — NEO Factory — https://www.1x.tech/discover/neo-factory
  12. Shield AI — U.S. Air Force production contract for CCA mission autonomy — https://shield.ai/shie…REDACTED/
  13. Shield AI — Hivemind — https://shield.ai/hivemind/
  14. Anduril — Battle Manager at Valiant Shield 2026 — https://www.anduril.com/news/andu…REDACTED
  15. AeroVironment — Halo_Shield PDF — https://investor.avinc.com/node/20746/pdf
  16. AeroVironment — Halo_Shield / E-HEL contract page — https://www.avinc.com/?avinc_solution_tax=halo_shield
  17. Meta — smartARM AI bionic prosthetics — https://about.fb.com/news/2026/09/cana…REDACTED/
  18. EurekAlert / Shirley Ryan AbilityLab — Bionic arm connected to bone, nerves, muscles — https://www.eurekalert.org/news-releases/1144060
  19. Axios — Brain implants fuel U.S.-China bioscience rivalry — https://www.axios.com/2026/09/03/brai…REDACTED
  20. Synchron — BCI home / clinical trial overview — https://synchronbci.com/
  21. Rice University — AI-powered materials laboratory NSF award — https://news.rice.edu/news/2026/acce…REDACTED
  22. Texas A&M — Autonomous robotic metallurgist materials innovation platform — https://stories.tamu.edu/news/2026/08/05/texa…REDACTED/
  23. Atinary — Self-Driving Labs Boston — https://atinary.com/news/open…REDACTED/
  24. Intuitive Surgical — Q2 2026 earnings — https://investor.intuitivesurgical.com/news-releases/news-release-details/intu…REDACTED
  25. NVIDIA Developer — Isaac for Healthcare — https://developer.nvidia.com/isaac/healthcare
  26. NVIDIA Technical Blog — Healthcare robotics with GPU-native medical physics simulation — https://developer.nvidia.com/blog/deve…REDACTED/
  27. NVIDIA Blog — Medical Physics Simulation open source — https://blogs.nvidia.com/blog/medi…REDACTED/
  28. TechCrunch — Eclipse $1.3B physical AI fund — https://techcrunch.com/2026/04/07/vc-e…REDACTED/
  29. GlobeNewswire — Maven Robotics $100M Series A — https://www.globenewswire.com/news-release/2026/09/10/3359782/0/en/mave…REDACTED.html
  30. TechCrunch — Mind Robotics $500M Series A — https://techcrunch.com/2026/03/11/rivi…REDACTED/
  31. Axios — Generalist raises another $200M — https://www.axios.com/2026/08/24/robo…REDACTED
  32. Tesla SEC 10-Q / Q2 2026 — https://www.sec.gov/Archives/edgar/data/1318605/000162828026049270/tsla-20260630.htm
Source in the house: eth-memory/sol/reports/2026-09-23-physical-ai-model-to-matter.md☉ Sol
← Previous · 2026-09-22 — Task 1 + Task 2 Run