JTCはもう終わり？ │米中vs日本 フィジカルAI最前線 中島聡が注目の業界は〇〇【伝説のエンジニア 中島聡 × 松尾研究所 金剛洙】

Overview

Main theme: the next big battleground is “physical AI” — humanoid and modular robots that pair hardware with base AI models.

This summary covers technology strategies, market dynamics, startup/funding issues, and national policy implications (US/China vs Japan). It highlights competing hardware/software approaches, organizational and funding challenges (particularly in Japan), infrastructure bottlenecks, and concrete recommendations for investors, startups, and policymakers.

---

Robotics & physical AI

Robot “base model” concept

• A base model for robots is analogous to large language models for text: a common robot “brain” that can be deployed across different hardware bodies to accelerate capabilities and adoption.

Two competing hardware/software approaches

1. Vertical integrated model (closed system)
   • Examples: Tesla, Figure.
   • Companies build hardware and AI together and ship finished products.
2. Modular/open model (open interfaces)
   • Examples: many Chinese efforts, Unity-like ecosystems.
   • Cheap modular hardware with open interfaces lets researchers and users attach their own brains, enabling rapid experimentation and global adoption.

Japan’s strategic opportunity

• Define open hardware interfaces (e.g., hand actuator standards).
• Compete on high-quality actuators and components.
• Develop safe, trusted AI models and export integrated, reliable systems.

Matsuo Research Institute initiatives

• Proposes an “AI Robot Church” / foundational model for physical AI.
• Plans to build an open-source “Generator I” to kickstart a humanoid robot market.

Dual-use and funding angle

• Suggests allocating portions of increasing defense budgets to fund humanoid and industrial robot development, enabling both military and civilian applications (elderly care, construction, etc.).

---

Product lifecycle & deployment lessons

• Early home/demo/beta robots are being released but often underperform functionally.
• Remote updates and operator support create feedback loops analogous to the GPT-3 era: usage-driven improvement accelerates capability gains.
• Data collection and iterative improvement are critical — real-world usage generates the feedback needed to make physical AI progressively more capable.

---

Hardware & infrastructure trends to watch

Short-term bottlenecks

• Power distribution (transformers, power lines) and availability of human operators.
• Some equipment already has waiting lists measured in years.

Memory hierarchy evolution

• Current trend: HBM (high-bandwidth memory) placed close to GPUs.
• Emerging alternatives:
  • External memory architectures (SAMRAM-like approaches).
  • Experiments using flash or other persistent memory.
  • Long-term bets: new SRAM designs or flash-based solutions.
• Fiber-optic interconnects and related infrastructure are key long-term strategic areas.

Investment profile

• These are high-risk, long-horizon areas but may be strategically important for physical AI and robotics deployment.

---

AI impact on white-collar work and business models

• Many white-collar roles (lawyers, system integrators, consultants, call centers) are being transformed by AI agents.
  • Example: an experienced lawyer may shift to an independent, managerial role while AI handles junior-level tasks.
  • Call centers can be automated by feeding manuals and field feedback into models.
• Emergence of new service models:
  • AI-powered boutique consultancies.
  • Forward-deployed engineers (FDEs) embedded in companies to implement AI automation.
• Business model shift:
  • Small, AI-native teams (high margins, low headcount) can outperform traditional labor-heavy scaling.
  • Investors may favor nimble AI-native firms over large legacy corporations.

---

Startup, funding and ecosystem analysis (Japan focus)

Current barriers in Japan

• Hardware-heavy startups require large capital; few have secured large funding rounds.
• Many Japanese VCs are risk-averse (bank-origin leadership) and favor revenue metrics over high-risk AI bets.
• Talent is improving, but funding is the main constraint.
• Structural constraints: bank/LP-driven VC models limit access to risk capital compared to the US and China.

Recommendations

• Create VCs with founder/operator experience and large funds that understand AI tech trends.
• Government or private initiatives to fund education and awareness programs, raising talent and entrepreneurial readiness.
• Encourage second-time founders and experienced operators to form VC LPs and invest in risky, strategic areas.

---

Organizational & human factors

• Hidden liabilities: large permanent workforces are balance-sheet liabilities that become more challenging as automation spreads.
• Organizational strengths remain crucial: business acumen, relationships, storytelling/sales ability, and leadership will still matter even as AI automates routine tasks.
• Collaborative initiatives: Singularity Society and Matsuo Research Institute aim to support startups, education, open-source projects, and VC-like activities.

---

Concrete tech / product items mentioned

• Humanoid robot base models (foundational physical AI)
• Modular robot hardware and standardized interfaces (hands/actuators)
• Remote update and operator support systems for deployed robots
• Memory technologies: HBM on-GPU, external memory (SAMRAM), flash alternatives, future SRAM designs
• Fiber-optic and power infrastructure investments
• AI agent tooling for legal work, system integration, and call centers
• Forward-deployed engineers (FDEs) and AI-native boutique consultancies
• Open-source “Generator I” project to stimulate the humanoid robot market

---

Actionable implications

• For investors
  • Consider long-horizon bets in memory, fiber, and power infrastructure.
  • Fund VCs with startup/operator experience who understand strategic AI hardware/software trends.
• For startups
  • Focus on modular hardware/software stacks, robust data collection loops, and operational feedback mechanisms.
  • Explore military/defense funding pathways where applicable to de-risk early capital needs.
• For policymakers
  • Fund education and standard-setting for open hardware interfaces.
  • Provide targeted grants to bootstrap physical AI ecosystems and encourage trusted exports.

---

Main speakers / sources

• 中島聡 (Satoshi Nakajima)
• 松尾研究所 金剛洙 (Matsuo Research Institute — 金剛洙)
• Additional referenced organizations and examples: Tesla, Unity, Figure, Matsuo Research Institute, Singularity Society

(End of summary.)