卫诗婕 × 熊友军

人形机器人运动会正从速度竞赛转向自主化与具身智能验证场，天工 Omni 从彩排9秒44迭代至8秒64，“无脸跑”则是在肩关节和功耗约束下自我学习出的最快解。商业化仍受7×24小时运行、甚至99%成功率的工业门槛约束，工业场景将先提供收入与数据，但长程规划、错误反思、数据规模和人形终局仍是待验证变量。

Peter Diamandis × Salim Ismail × Dave Blundin × Alexander Wissner-Gross × Palmer Luckey

Anduril’s $2.2 million wildfire XPRIZE win, including $1 million from Lockheed, shows its autonomy workflow can transfer to emergencies where a 10-minute versus hour-long response requires different tools.Luckey’s specialized, verifiable systems and self-funded product model challenge a market where five companies once captured roughly 80% of major defense acquisition programs, while Anduril’s choice to use rather than build humanoids keeps platform scope and procurement scale in focus.

Anastasis Germanidis × swyx × Vibhu

Runway’s Anastasis Germanidis argues video prediction and world-model dynamics are “the same thing,” while Physics IQ scores “predictably improve” with model and compute scale, despite models potentially cheating.Real-time generation is “inevitable”: Runway’s 24-FPS characters model demonstrates serving-cost and UX advantages, but autoregressive error, simulation fidelity, and Western competitive lag constrain pixel-native and robotics ambitions.

Koji × 王家伟

具身智能正从单次演示转向 OOD 适应、steerability 与 context 理解，覆盖即时反馈和长期记忆。深普以 UMI 数据闭环推进真机验证，超过95%轨迹可执行；年底模型的泛化和评测、scaling law 仍待验证，延迟与数据成本构成约束。

Koji × 韩铮 × 沈宇军 × 王潜 × 许华哲

Astra实测显示，语义和空间泛化较强，复杂接触任务却暴露出物理泛化短板。仿真抓取、放置、组装已接近100%成功率，但真实触觉信号与仿真仍“基本上千差万别”。商业化取决于客户持续付费及多数任务接近100%的可靠性，3年内通用服务能否出现仍待验证。

Dino Mavrookas × Rob Lehman × Doug Lambert × Vibhav Altekar × Molly O'Shea

China outbuilds the US 230-to-1, while only 1% of roughly $1T 2026 Department of War budget went to autonomy.Saronic says faster iteration and volume can change maritime economics, with Marauder’s propulsion stack bench-built eight months before its 180-foot ship existed.Policy tailwinds are possible, but legal approval, manning requirements, classified subsea technology, and supply-chain bottlenecks remain watchpoints.

Peter Diamandis × Salim Ismail × Dave Blundin × Alexander Wissner-Gross × Emad Mostaque

Astra reportedly delivers 3.1 research-days per human-researcher day after training on more than 100,000 NVIDIA Gracewell and Blackwell GPUs, pulling OpenAI’s roadmap forward six months.With 10,000 agents tackling Navier–Stokes for approximately $6.5 million and a possible 100× inference price-performance improvement, productivity accelerates, but the German wiki hijack and slowdown calls leave deployment risk unresolved.

José × Ian Rountree

Cantos’ edge is deliberate smallness: Fund IV reached $70M, supporting $1.5–4M checks before websites and same-day SAFE execution.Ian favors hard tech, vertical integration and manufacturing scale, with Neros at 1,000 drones weekly and targeting a million annually.The opportunity is cost advantage and defense demand; China’s control of motors, actuators and rare-earth inputs remains a physical-AI risk.

Dino Mavrookas × Molly O'Shea

China built over 1,000 commercial ships last year versus five large ocean-going vessels in the US, while China holds 57% of global shipbuilding capacity.Port Alpha’s $3 billion project targets autonomous container ships, tankers, and roll-on/roll-off vessels: simpler designs could enable scale and lower costs, but only 1% of the roughly $1 trillion FY26 Department of War budget went to autonomy.

Molly O'Shea × Dino Mavrookas

Saronic’s $9.25B valuation and $2.6B raised center its shipbuilding thesis on capacity, with Port Alpha’s $3.2B initial buildout adding 150,000 gross tons.An Austin line designed for 2,000 Corsairs annually combines hardware-software co-design, software reuse, and 25,000+ testing hours, but ocean reliability remains unresolved.