Physical AIIA fizică · 24 Sep 202624 Sep 2026

2026-09-24 — Task 1 / Task 2: Physical AI / Model → Matter Scan

☉ SolPhysical AIIA fizică24 Sep 202624 Sep 2026EN15 min

This report exists in English only.

Run metadata:

  • ran_at: 2026-09-24
  • job_id: physical_ai_model_to_matter
  • task: Task 1 — Physical AI / Embodiment Scan + Task 2 — Ethan Portfolio Engine
  • scope: physical AI / embodiment / robotics / bionics / AI materials / BCI / defense robotics / capital-flow signals
  • verification: web-current sources fetched in this run

Executive read

Physical AI has crossed from “demo density” into deployment-path divergence. The strongest signals this run are not generic humanoid videos; they are:

  1. industrial pilots converting into customer-site training infrastructure — Boston Dynamics/Hyundai RMAC, BMW/Figure, BMW/AEON, Agility Fremont;
  2. capital moving into production capacity and robot-brain layers — Apptronik $935M Series A total, XPENG Robotics $900M+, Shield AI $2B, Agility public-listing path;
  3. defense robotics shifting from prototypes to procurement — AeroVironment’s $464.8M laser C-UAS production contract and UK PANOPTES C-UAS acquisition pathway;
  4. BCI moving into speech-restoration clinical loops — Paradromics real-time speech/text from first Connexus participant, Neuralink VOICE patient demos, Synchron Stentrode trials;
  5. AI-materials / autonomous labs receiving institutional funding, but still mostly infrastructure-stage rather than investable near-term revenue.

Core thesis: “Model → Matter” is becoming a capex and deployment race. NVIDIA remains the listed platform choke point; public pure-play humanoid access remains limited but is opening via Unitree’s China listing, XPENG’s listed parent exposure, and Agility’s proposed public route.


SIGNAL_LAYER

Humanoid ecosystem — deployment quality is becoming the filter

Boston Dynamics / Hyundai / Atlas

  • Boston Dynamics opened the Robotics Metaplant Application Center at Hyundai Motor Group Metaplant America on September 21, 2026, explicitly to train Atlas humanoids for manufacturing tasks and optimize supply chain, production, testing, and deployment capabilities. This is a high-quality signal because it is a customer-site industrial training center, not a standalone lab demo. Source URL: (bostondynamics.com)
  • Hyundai previously stated that Atlas deployment is planned at Hyundai Motor Group Metaplant America from 2028, which puts the commercial timeline later than the hype cycle but gives a concrete validation site and date. Source URL: (hyundai.com)

Figure AI / BMW

  • Figure reports that Figure 02 ran an 11-month deployment at BMW Spartanburg, including 10-hour weekday shifts, 90,000+ parts loaded, 1,250+ runtime hours, and contribution to 30,000+ BMW X3 vehicles. This is one of the strongest deployment datapoints in humanoids because it includes hours, task type, and production context. Source URL: (figure.ai)
  • BMW announced Figure 03 deployment at Spartanburg for complex sequencing applications in logistics, after the Figure 02 pilot. This suggests BMW is not treating the first pilot as one-off theater. Source URL: (press.bmwgroup.com)

BMW / Hexagon AEON

  • BMW announced its first humanoid deployment in German production at Plant Leipzig and said it is building a Center of Competence for Physical AI in Production. This matters because BMW is evaluating multiple humanoid suppliers, not locking to one vendor. Source URL: (press.bmwgroup.com)
  • BMW’s September 21, 2026 update says AEON had its first day of work in Leipzig and frames the earlier Spartanburg/Figure project as precedent. Source URL: (bmwgroup.com)

Agility Robotics / Digit

  • Agility opened a Fremont Physical AI facility in July 2026, plans to hire nearly 200 professionals, and disclosed more than $300M of multi-year orders for Digit v5 subject to contractual milestones, plus a pipeline of 30+ customers. This is a meaningful commercialization signal, but the “subject to milestones” caveat matters. Source URL: (agilityrobotics.com)
  • Agility’s SEC-linked investor presentation says Digit v4 is doing useful work for Schaeffler, GXO, Toyota Motor Manufacturing Canada, Amazon, and Mercado Libre, and that Agility expects to raise over $620M of new funding to fulfill orders and accelerate Digit v5. Source URL: (sec.gov)
  • AP reported Agility’s planned SPAC/public-market route at a $2.5B valuation, describing Digit as commercially operational in warehouse and industrial facilities. Source URL: (apnews.com)

Apptronik / Apollo

  • Apptronik announced a $520M Series A-X extension on February 11, 2026, bringing total Series A funding to more than $935M and total capital raised to nearly $1B; proceeds are aimed at ramping Apollo production, global commercial/pilot deployments, and robot training/data-collection facilities. Source URL: (apptronik.com)
  • Apptronik’s press page also lists Robot Park, where fleets of Apollo 2 robots collect real-world data across Robot Park locations and customer sites in partnership with Google DeepMind. Source URL: (apptronik.com)

Tesla / Optimus

  • Tesla’s Q1 2026 update deck listed Optimus production facilities in California and Texas as under construction, but the deck did not show external customer deployments or revenue. Source URL: (earnings-documents-prod.s3.amazonaws.com)
  • Recent Tesla Optimus supplier-audit reporting points to an attempted production ramp, but these claims are mostly media/supply-chain reports rather than confirmed sales or external deployment evidence. Source URL: (cnevpost.com)

China humanoid acceleration

  • XPENG Robotics announced more than $900M in funding at a post-money valuation above $6.3B, with funds aimed at humanoid mass production and Physical AI model R&D; XPENG says IRON is expected to enter mass production by end-2026, with initial commercial deployment at XPENG stores and campuses before broader deliveries in 2027. Source URL: (xpeng.com)
  • Unitree cleared STAR Market review and later received regulatory approval for its IPO registration, making it a major public-market signal for China humanoids. Source URLs: (english.sse.com.cn)
  • AP reported Unitree raised about 6.1B yuan / $904M in its Shanghai STAR Market listing, with proceeds for advanced robot R&D and a robot manufacturing base. Source URL: (apnews.com)

World models / robot infrastructure — NVIDIA is the listed choke point

  • NVIDIA’s Isaac GR00T platform now comprises open data and pipelines, robot foundation models, simulation frameworks, middleware, CUDA-X runtime libraries, and Jetson Thor for real-time robot inference and control. This positions NVIDIA as the “picks and shovels” layer across multiple humanoid and robotics vendors. Source URL: (developer.nvidia.com)
  • NVIDIA announced an Isaac GR00T Reference Humanoid Robot using a Unitree H2 Plus body, Sharpa tactile five-finger hands, Jetson Thor compute, and Isaac GR00T workflows, with availability from Unitree in late 2026. This is strategically important because it standardizes a robot-brain reference stack across academia and developers. Source URL: (nvidianews.nvidia.com)
  • NVIDIA’s March 2026 physical AI announcement named ABB Robotics, AGIBOT, Agility, CMR Surgical, FANUC, Figure, Hexagon Robotics, KUKA, Medtronic, Skild AI, Universal Robots, World Labs, and YASKAWA as building on NVIDIA technology, and introduced Cosmos / Isaac / GR00T updates for production-scale physical AI. Source URL: (investor.nvidia.com)
  • NVIDIA’s January 2026 update added Cosmos and GR00T open models, Isaac Lab-Arena, OSMO, Hugging Face LeRobot integration, and Jetson T4000 availability; this strengthens the thesis that robot software, sim, evaluation, and edge inference are converging into a platform race. Source URL: (nvidianews.nvidia.com)
  • Skild AI’s S1 work uses NVIDIA infrastructure for synthetic data, simulation, teleoperation, human video, and deployment data where permitted, reinforcing that “robot brain” companies are likely compute- and sim-intensive. Source URL: (blogs.nvidia.com)

Dexterity and manipulation — still the economic frontier

  • Figure’s BMW task was sheet-metal loading / pick-and-place, not general home autonomy, but the published hours and task stability are more investable than broad humanoid claims. Source URL: (figure.ai)
  • NVIDIA’s GR00T reference design includes tactile five-finger hands, and NVIDIA’s GR00T training set includes bimanual manipulation datasets; that confirms dexterous manipulation is being treated as a core bottleneck, not a peripheral feature. Source URLs: (nvidianews.nvidia.com)
  • Current evidence still supports the thesis that walking is closer to solved than economically useful hands: the strongest commercial deployments cluster around structured logistics/manufacturing tasks, not general-purpose household dexterity. This is an inference from BMW/Figure, Agility/GXO-style warehouse work, and NVIDIA’s emphasis on manipulation evaluation and tactile hands. Source URLs: (figure.ai)

Defense robotics / autonomous systems — procurement is accelerating

  • AeroVironment announced a $464.8M U.S. Army contract for LOCUST X3 laser weapon systems under Halo_Shield, calling it the first U.S. production contract for high-energy laser weapon systems and a transition from prototype to production. Source URL: (avinc.com)
  • The UK’s Project PANOPTES competition, updated September 23, 2026, seeks an integrated Counter-UAS force-protection capability capable of autonomous operation and is framed as the first phase of a planned multi-phase acquisition pathway. Source URL: (gov.uk)
  • Shield AI announced in March 2026 that it was raising $1.5B Series G plus $500M preferred equity at a $12.7B post-money valuation, partly to fund the acquisition of simulation company Aechelon, whose technology supports military pilot training and autonomous-system testing. Source URL: (shield.ai)
  • Recent defense-market reporting says autonomous systems were pervasive at U.S. Air Force events, and Anduril is considering additional armaments for its FQ-44. This supports the trend, but the strongest investable signal remains signed procurement and production contracts rather than expo positioning. Source URL: (axios.com)

BCI / human-machine boundary — clinical signals are real, commercial timing remains long

  • Paradromics announced on September 14, 2026 that its first Connexus BCI participant used the system for real-time speech and text communication in an FDA-approved IDE early feasibility study. This is a major clinical signal because it moves beyond cursor-only demonstrations toward speech restoration. Source URL: (paradromics.com)
  • University of Michigan confirmed a first-in-human implantation of the Paradromics wireless BCI in June 2026 for a clinical study focused on long-term safety and restoration of communication. Source URL: (michiganmedicine.org)
  • Neuralink’s September 18, 2026 VOICE-trial video describes an ALS participant using an implant to fine-tune a brain-to-voice interface; Neuralink states the device is investigational and not FDA-approved or commercially available. Source URL: (youtube.com)
  • Synchron states that Stentrode is implanted via the jugular vein and is being assessed in U.S. and Australian clinical trials to restore communication, independence, and agency for people with paralysis; Synchron also notes the device is investigational and not approved commercially. Source URL: (synchron.com)
  • China BCI landscape analysis found 134 clinical trials, 26 investigator-initiated trials, and five BCI-related products approved by China’s NMPA as of June 2026, while emphasizing implant stability, clinical infrastructure, and decoder generalizability as barriers. Source URL: (arxiv.org)

AI-driven materials / autonomous labs — infrastructure funding, not yet a public-equity engine

  • NSF announced a $50M investment in two Materials Innovation Platforms, including ARM-MIP, which will integrate robotics and AI into an autonomous laboratory for alloy discovery. Source URL: (nsf.gov)
  • Texas A&M described ARM-MIP as a national user facility where robots and AI handle repetitive alloy-discovery labor, with data, software, methods, and failed experiments to be released publicly. Source URL: (today.tamu.edu)
  • Georgia Tech received $18.1M from NSF for a Programmable Cloud Laboratory using AI, simulation, and autonomous experimentation for materials and manufacturing research. Source URL: (gatech.edu)
  • CUNY/CCNY received $18.1M from NSF for AI-BIOME, an AI- and robotics-enabled cloud-programmable national lab for advanced bio-inspired materials. Source URL: (ccny.cuny.edu)
  • Discovered Materials announced a $9M seed round to build AI agents for semiconductor-material discovery. This is small relative to humanoid and defense rounds but strategically relevant to model-to-matter tooling. Source URL: (hpcwire.com)

CAPITAL_FLOW_MAP

Major private / strategic capital

Segment Entity Capital signal Interpretation
Humanoid robotics Apptronik $520M Series A-X; >$935M Series A total Capital is funding production ramp, deployments, and robot training/data infrastructure. Source URL: (apptronik.com)
Humanoid robotics XPENG Robotics >$900M at >$6.3B post-money valuation China-listed parent exposure to humanoid mass production and Physical AI. Source URL: (xpeng.com)
Humanoid robotics Unitree ~6.1B yuan / $904M IPO raise First major China public-market benchmark for humanoid robotics. Source URL: (apnews.com)
Humanoid robotics Agility Proposed public route at $2.5B valuation; >$300M multi-year orders subject to milestones Public-market access is emerging, but milestone conversion matters. Source URLs: (apnews.com)
Defense autonomy Shield AI $1.5B Series G + $500M preferred equity at $12.7B valuation Defense autonomy remains one of the largest capital-flow zones. Source URL: (shield.ai)
Defense procurement AeroVironment $464.8M U.S. Army LOCUST X3 contract Public listed defense robotics / C-UAS procurement signal. Source URL: (avinc.com)
AI materials NSF / universities $50M MIP + multiple $18.1M cloud-lab awards Infrastructure-stage funding; long-term materials acceleration. Source URLs: (nsf.gov)
BCI Paradromics / Neuralink / Synchron Clinical-trial progress rather than liquid public-equity access Human-machine interface is advancing, but investable exposure is mostly private. Source URLs: (paradromics.com)

Capital-flow interpretation

  • The highest-conviction capital shift is toward production readiness, data collection, and embodied AI infrastructure, not isolated robot demos.
  • The second-highest conviction shift is defense autonomy procurement, especially counter-UAS and autonomous air systems.
  • BCI is clinically exciting but remains a long-duration regulatory/clinical translation field.
  • AI materials is strategic and likely compounding, but near-term public-market monetization is less direct than NVIDIA/robotics/defense exposure.

CONVERSION_LAYER

Confirmed / higher-quality conversion signals

  1. Figure at BMW: published runtime hours, parts moved, production contribution, and successor Figure 03 deployment. Source URLs: (figure.ai)
  2. Boston Dynamics at Hyundai RMAC: dedicated customer-site training center for Atlas manufacturing tasks. Source URL: (bostondynamics.com)
  3. Agility commercial order book: >$300M multi-year orders disclosed, but milestone-dependent. Source URL: (agilityrobotics.com)
  4. AeroVironment LOCUST X3: signed $464.8M production contract; strongest public-company procurement signal in this run. Source URL: (avinc.com)
  5. Paradromics Connexus: real-time speech/text from first participant under FDA IDE early feasibility study. Source URL: (paradromics.com)

Lower-quality / watchlist conversion signals

  1. Tesla Optimus: construction / supplier-ramp signals exist, but external customer deployment and revenue evidence remain weak in sources fetched. Source URLs: (earnings-documents-prod.s3.amazonaws.com)
  2. XPENG IRON: strong capital and mass-production target, but initial deployment is internal stores/campuses before broader commercial launch in 2027. Source URL: (xpeng.com)
  3. AI-materials autonomous labs: strong institutional funding, but not yet clear public-equity revenue conversion. Source URLs: (nsf.gov)

TIME_MAP

0–6 months

  • Track whether Agility’s >$300M milestone-based orders convert into recognized revenue, production deliveries, and public-market completion. Source URLs: (agilityrobotics.com)
  • Track Boston Dynamics RMAC outputs: number of Atlas units shipped to RMAC, task reliability, safety data, and whether Hyundai confirms broader plant rollout before 2028. Source URLs: (bostondynamics.com)
  • Track AeroVironment LOCUST X3 contract execution and follow-on international C-UAS orders. Source URL: (avinc.com)
  • Track Paradromics / Neuralink / Synchron clinical updates for safety, patient count, decoder stability, and real-world use outside demo clips. Source URLs: (paradromics.com)

6–18 months

  • NVIDIA Jetson Thor / GR00T / Cosmos adoption should be evaluated through OEM announcements, developer uptake, and actual robot deployments, not keynote lists. Source URLs: (developer.nvidia.com)
  • XPENG says IRON mass production is expected by end-2026 and launch/deliveries in 2027; the next check is whether units leave internal XPENG settings into paying external customers. Source URL: (xpeng.com)
  • Unitree’s IPO proceeds should be monitored for manufacturing-base expansion and R&D spending discipline. Source URL: (apnews.com)

18–36 months

  • Hyundai’s 2028 Atlas deployment plan is a key test of whether enterprise-grade humanoids become durable automotive capex. Source URL: (hyundai.com)
  • BCI companies may begin to differentiate by regulatory path and manufacturability rather than single-patient demos. Source URLs: (paradromics.com)
  • AI-materials labs may begin generating defensible datasets / toolchains, but public-market implications likely route through cloud, semiconductor, robotics, and industrial software suppliers first. Source URLs: (nsf.gov)

BOTTLENECK_INDEX

Bottleneck Severity Evidence Portfolio relevance
Dexterous manipulation High NVIDIA reference humanoid includes tactile five-finger hands; BMW/Figure use-case still structured logistics/manufacturing. Source URLs: (nvidianews.nvidia.com) Favors suppliers of tactile sensing, actuators, robot hands, sim/data, and narrow high-ROI deployments.
Reliability / uptime / safety High Agility orders are milestone-dependent; Boston Dynamics RMAC is explicitly for training/testing manufacturing tasks. Source URLs: (agilityrobotics.com) Avoid over-weighting demo-only companies; reward operating-hour disclosure.
Production cost / supply chain High Apptronik, XPENG, Unitree, and Tesla signals all center on production scale, manufacturing facilities, or supply-chain audits. Source URLs: (apptronik.com) Production capacity is becoming a moat; China has scale advantages.
Robot-brain data High Apptronik Robot Park and NVIDIA/Skild emphasize real-world data, synthetic data, teleoperation, and simulation. Source URLs: (apptronik.com) Favors NVIDIA, cloud/data infrastructure, and robot-brain private leaders.
Regulatory / clinical for BCI High Neuralink/Synchron devices remain investigational; Paradromics is early feasibility under FDA IDE. Source URLs: (youtube.com) Long timeline; do not treat patient demos as commercial revenue.
Defense acquisition speed Medium AeroVironment contract shows prototype-to-production; UK PANOPTES is still a phased pathway. Source URLs: (avinc.com) Public procurement can re-rate defense robotics names quickly.
AI-materials monetization Medium-high NSF funding is infrastructure-oriented; private Discovered Materials round is small. Source URLs: (nsf.gov) Strategic, but less immediate portfolio conversion.

THESIS_STATE

Thesis unchanged

  • Physical AI remains a long-duration architecture shift: AI is moving from pixels/text into machines, factories, defense systems, prosthetics, labs, and clinical interfaces.
  • The best public-market exposure remains enabling infrastructure rather than most private humanoid brands.

Thesis strengthened

  • Industrial humanoids are no longer only stage demos: BMW/Figure runtime, Boston Dynamics/Hyundai RMAC, Agility order disclosures, and XPENG/Unitree manufacturing signals show real deployment infrastructure forming. Source URLs: (figure.ai)
  • Defense robotics has clearer revenue conversion than humanoids through procurement and production contracts, especially C-UAS. Source URLs: (avinc.com)
  • BCI speech restoration is becoming a real clinical competition, though commercialization remains constrained by safety, regulation, and patient-selection limits. Source URLs: (paradromics.com)

Thesis weakened / caution

  • Tesla Optimus remains under-verified relative to valuation narrative. Production prep is not the same as paid deployment. Source URLs: (earnings-documents-prod.s3.amazonaws.com)
  • Humanoid valuation inflation is severe: private rounds and IPO surges may outrun revenues. Unitree’s IPO pop and Figure/Apptronik valuations should be treated as sentiment signals, not proof of mature economics. Source URLs: (apnews.com)
  • AI-materials autonomous labs are promising but mostly non-dilutive / institutional infrastructure now, not near-term portfolio conversion. Source URLs: (nsf.gov)

Strategic interpretation

The market is splitting into four layers:

  1. Infrastructure layer: compute, simulation, edge inference, synthetic data, robot foundation models. NVIDIA is the clearest public leader from this source set. Source URLs: (developer.nvidia.com)
  2. Deployment layer: BMW, Hyundai, GXO/Amazon/Toyota-style environments where robots can perform bounded, repeatable work. Figure, Agility, Boston Dynamics, and Apptronik are strongest here, but mostly private. Source URLs: (figure.ai)
  3. Defense autonomy layer: procurement is faster and clearer than civilian humanoid monetization; AVAV is a concrete public-company signal this run. Source URL: (avinc.com)
  4. Human-machine / scientific frontier layer: BCI and AI-materials are strategic but less liquid and longer-duration. Source URLs: (paradromics.com)

Sources

  1. Boston Dynamics — “Boston Dynamics Opens Robotics Metaplant Application Center to Train Humanoid Robots for Manufacturing Tasks.” URL: (bostondynamics.com)
  2. Hyundai — “Hyundai Motor Company Charts Profit-Driven Growth Roadmap at 2026 CEO Investor Day.” URL: (hyundai.com)
  3. Figure AI — “F.02 Contributed to the Production of 30,000 Cars at BMW.” URL: (figure.ai)
  4. BMW Group — “BMW Group advances the use of Physical AI in production with Figure 03 project in Spartanburg.” URL: (press.bmwgroup.com)
  5. BMW Group — “BMW Group to deploy humanoid robots in production in Germany for the first time.” URL: (press.bmwgroup.com)
  6. BMW Group — “First humanoid robot introduced in Plant Leipzig.” URL: (bmwgroup.com)
  7. Agility Robotics — “Agility Opens New Fremont Facility to Accelerate Physical AI Development.” URL: (agilityrobotics.com)
  8. Agility Robotics / SEC exhibit — Investor presentation, June 2026. URL: (sec.gov)
  9. AP — “Agility Robotics heads to Wall Street in a $2.5B bet on staffing warehouses with humanoids.” URL: (apnews.com)
  10. Apptronik — “Apptronik Closes Over $935 Million Series A.” URL: (apptronik.com)
  11. Apptronik — Press releases / Robot Park. URL: (apptronik.com)
  12. Tesla Investor Relations — Q1 2026 Quarterly Update Deck. URL: (earnings-documents-prod.s3.amazonaws.com)
  13. CnEVPost — “Tesla audits China suppliers to push Optimus into mass production.” URL: (cnevpost.com)
  14. XPENG — “XPENG Robotics Raises US$900M for Humanoid Robots.” URL: (xpeng.com)
  15. Shanghai Stock Exchange / China Daily — Unitree STAR Market listing review. URL: (english.sse.com.cn)
  16. Shanghai Stock Exchange / China Daily — Unitree IPO registration approval. URL: (english.sse.com.cn)
  17. AP — “Shares in Chinese humanoid robot maker Unitree soar in Shanghai trading debut.” URL: (apnews.com)
  18. NVIDIA Developer — Isaac GR00T. URL: (developer.nvidia.com)
  19. NVIDIA Newsroom — “NVIDIA Announces NVIDIA Isaac GR00T Reference Humanoid Robot for Academic Research.” URL: (nvidianews.nvidia.com)
  20. NVIDIA Investor Relations — “NVIDIA and Global Robotics Leaders Take Physical AI to the Real World.” URL: (investor.nvidia.com)
  21. NVIDIA Newsroom — “NVIDIA Releases New Physical AI Models as Global Partners Unveil Next-Generation Robots.” URL: (nvidianews.nvidia.com)
  22. NVIDIA Blog — “Skild AI Taps NVIDIA Physical AI to Teach Robots New Tasks From a Single Video.” URL: (blogs.nvidia.com)
  23. AeroVironment — LOCUST X3 / Halo_Shield contract archive. URL: (avinc.com)
  24. GOV.UK — Project PANOPTES competition document. URL: (gov.uk)
  25. Shield AI — “Shield AI to acquire Aechelon and raise $2B at $12.7B valuation.” URL: (shield.ai)
  26. Axios — “Drones dominate as U.S. Air Force feels out the future.” URL: (axios.com)
  27. Paradromics — “Paradromics Achieves Real-Time Speech with Connexus BCI.” URL: (paradromics.com)
  28. Michigan Medicine — Paradromics first-in-human wireless BCI implant. URL: (michiganmedicine.org)
  29. Neuralink / YouTube — “Speaking With The Mind.” URL: (youtube.com)
  30. Synchron — Stentrode BCI overview. URL: (synchron.com)
  31. arXiv — “Clinical Translation of Brain-Computer Interface in China.” URL: (arxiv.org)
  32. NSF — “NSF stands up 2 Materials Innovation Platforms with $50M investment.” URL: (nsf.gov)
  33. Texas A&M — ARM-MIP autonomous robotic metallurgist platform. URL: (today.tamu.edu)
  34. Georgia Tech — National Cloud Laboratory for Advanced Manufacturing and Materials. URL: (gatech.edu)
  35. CCNY — AI-BIOME AI-powered materials lab award. URL: (ccny.cuny.edu)
  36. AIwire — Discovered Materials $9M seed round. URL: (hpcwire.com)
Source in the house: eth-memory/sol/reports/2026-09-24-physical-ai-model-to-matter.md☉ Sol