Version 27

What

Why it matters

Open questions

• Brett Adcock argues training data scarcity is the sole remaining barrier to general robotics and projects home robots by end 2026[14][15][13] — does Figure's 17-hour factory shift[11] and BMW automotive deployment[10] represent capability close to generalizing, or is the unstructured home environment still a separate problem?

• Morgan Stanley has revised its China 2026 humanoid shipment forecast up to 50,000 units for the third time this year[5] — how much of China's deployment lead reflects industrial policy and state procurement versus genuine production cost advantages?

• Meta is discussing doubling AI glasses production to 20M units annually[21], and Apple is reported preparing a 2027 entry[24] — does Apple's entry compress Meta's window to lock in platform dominance, or does Meta's 3.2B-user distribution make the share lead durable?

• Meta's Brain2Qwerty v2 reaches 78% word accuracy using non-invasive MEG helmets rather than implanted electrodes, beating prior non-invasive 8% accuracy baselines via an LLM error-correction layer[25] — does performance improve fast enough to make non-invasive BCI a practical computing modality, or does the gap to surgical approaches remain too wide?

Narrative

Timeline

• 2026-01-01: Boston Dynamics Atlas debuts as production-ready at CES 2026; Hyundai Motor Group announces FieldAI partnership using Boston Dynamics' platform. [42][43]
• 2026-05-25: FieldAI closes $400M+ raise backed by Intel Capital, announces NVIDIA collaboration and Hyundai Motor Group deployment via Boston Dynamics. [53][54][52]
• 2026-06-03: NVIDIA at CVPR: Cosmos 3, GraspGen-X (zero-shot grasping trained on 2B simulated grasps), Physical AI Dataset at 15M+ downloads. [6][7]
• 2026-06-04: Ars Technica's Jeremy Hsu argues humanoid robot viral demos exploit anthropomorphic bias and reliable real-world performance lags behind what demos imply. [31]
• 2026-06-09: Google DeepMind launches a three-month London accelerator for 15 European robotics startups, offering Gemini robotics models and its full AI stack. [30]
• 2026-06-12: Prometheus raises $12B (total ~$18B, 150 employees); Bezos co-CEO frames the mission as an 'artificial general engineer' for the physical economy, explicitly not a robotics company. [40][26][28][27][55][41]
• 2026-06-16: NVIDIA XR AI enters public beta for AR glasses developers; Snap opens preorders for Specs smart glasses at $2,195 ahead of an autumn launch. [35][23]
• 2026-06-21: Hyundai acquires 100% of Boston Dynamics by purchasing SoftBank's remaining 9.65% stake for $325M. [18]
• 2026-06-24: Meta launches own-brand Meta Glasses at $299 powered by Muse Spark; SemiAnalysis publishes deep-dive on Unitree framing China's humanoid approach as applying AI scaling laws with geopolitical implications. [22][19]
• 2026-06-24: Agibot streams live Chinese factory floor humanoid robot operations. [20]
• 2026-06-25: Figure AI signs agreement with Catalyst Brands for humanoid warehouse deployment — its first publicly named retail/logistics customer. [8][9]
• 2026-06-25: Figure AI's robots are documented completing 17-hour shifts sorting 22,000+ packages; deployed robot count exceeds the company's human employee count. [39][11][12]
• 2026-06-26: BMW Group deploys Figure 03 humanoid robots at its Spartanburg plant; Figure CEO projects home robots capable of long-horizon tasks by end 2026. [10][56][14][15]
• 2026-06-27: PitchBook data: Q1 2026 robotics and physical AI investment reached ~$16B across ~500 deals, all-time highs and ~4.5x the 2021–2025 value run rate. [1][2]
• 2026-06-28: Meta disclosed selling 7 million AI glasses in 2025 — triple the combined 2023–2024 total, 80%+ share in a 322%-growth market — with production capacity discussions targeting 20M units annually by end 2026. [21]
• 2026-06-28: BMW's official Figure 03 announcement confirms substance of a 2025 Fortune story that Figure AI had threatened legal action over. [16][17]
• 2026-06-30: Meta open-sources Brain2Qwerty v2, a non-invasive brain-to-text system achieving 78% word accuracy for its strongest participant using MEG helmet recordings rather than implanted electrodes. [25]
• 2026-06-30: Reports circulate that Apple is preparing smart glasses to compete with Meta, with a launch expected in 2027. [24]
• 2026-07-01: Morgan Stanley raises China humanoid robot 2026 shipment forecast to 50,000 units — third upward revision of the year, tripling from 14,000 at year start. [5]

Perspectives

Tensions

• Demo-readiness vs. real-world reliability: Figure AI's BMW deployment, 17-hour factory shifts, and Adcock's claim that data is the only remaining barrier represent the case for near-term general capability; Ars Technica's Jeremy Hsu argues demos exploit anthropomorphic bias and reliable repeatable performance lags behind what milestones imply. [8][39][14][11][10][31][12][13]
• US vs. Chinese humanoid deployment pace: Figure AI, Hyundai/Boston Dynamics, and FieldAI lead the US commercial deployment narrative, while China holds 80%+ of global humanoid installations and Morgan Stanley's third upward forecast revision puts 2026 Chinese shipments at 50,000 units. [4][19][20][45][18][8][10][5]
• 'Artificial general engineer' ambition vs. no shipped product: Prometheus frames its mission as AGE infrastructure for the physical economy with $18B+ raised, while critics note it has shipped nothing at a multi-billion-dollar valuation. [26][28][49][50][41][51]
• Vertical integration vs. ecosystem partnership: Hyundai owns 100% of Boston Dynamics and is Atlas's primary customer, while NVIDIA, FieldAI, and Google DeepMind pursue ecosystem models that depend on third-party deployers. [18][32][33][52][30]
• Smart glasses market structure: Meta at $299 with 80%+ share and distribution through 3.2 billion social platform users vs. Snap at $2,195 and Apple expected in 2027 — whether Meta's pricing and distribution lock in the category before Apple arrives is unresolved. [22][23][21][24]
• Non-invasive vs. surgical BCI: Meta's Brain2Qwerty v2 achieves 78% word accuracy without surgery using MEG helmets, well above prior non-invasive 8% baselines, but whether the gap to surgical approaches narrows fast enough for practical hands-free computing remains open. [25]