#AI horizons 25-06 – Robotics Reignited

Executive Summary

Robotics is back in the spotlight, and this time it’s industrial-scale. SoftBank’s $1T bet on a new Arizona-based AI robotics hub, Google’s cloud-free Gemini Robotics model, Meta’s self-learning V-JEPA 2, and NVIDIA’s surging robotics revenue all point to a new phase of automation. From common-sense world models to dexterous bi-arm robots, 2025 may be the year robotics gets its AI renaissance—with strategic implications for infrastructure, chip manufacturing, and the U.S. industrial base.

Key Points

• SoftBank plans a $1 trillion AI robotics complex in Arizona, “Project Crystal Land,” inspired by Shenzhen.
• Masayoshi Son is courting TSMC, Samsung, and U.S. government support, including possible tax breaks.
• SoftBank’s AI investments include a $40B round in OpenAI and $6.5B acquisition of Ampere, alongside Project Stargate.
• Google’s Gemini Robotics On-Device matches its cloud-based version and supports high-dexterity tasks on-device.
• Meta’s V-JEPA 2 is a new open-source “world model” trained on real-world physical data, enabling self-learning robotics.
• NVIDIA’s robotics division generated $567M in Q2 2025, confirming robotics as one of its two fastest-growing sectors.

In-Depth Analysis

SoftBank Bets Big on U.S. Robotics Industrialization

After months of silence in the robotics sector, SoftBank’s Project Crystal Land signals a massive reawakening. The $1 trillion Arizona project aims to transform the U.S. into a high-tech manufacturing epicenter for AI-powered industrial robots. Inspired by China’s Shenzhen, this initiative reflects Masayoshi Son’s longstanding ambition to lead the next industrial AI wave.

To realize this vision, Son is in early-stage talks with Taiwan Semiconductor Manufacturing Co. (TSMC), Samsung, and the Trump administration. While TSMC already has a $165B AI chip expansion underway in Phoenix, it has yet to publicly commit to Crystal Land. Meanwhile, SoftBank is also lobbying U.S. officials for tax incentives and inviting Vision Fund portfolio companies to co-invest or establish production lines.

Project Crystal Land is not an isolated effort. It follows SoftBank’s broader infrastructure strategy, including its $500B-backed Project Stargate—a data center network developed in partnership with OpenAI, Oracle, and Ampere. SoftBank’s aggressive push aligns with its recent $40B OpenAI investment and Ampere acquisition, consolidating its position as a keystone investor in AI industrial scale-ups.

Google Delivers On-Device Robotics AI

Google has launched a version of its Gemini Robotics AI that operates entirely on-device—without needing cloud access. Gemini Robotics On-Device matches the performance of its cloud-based sibling and supports fine-tuning, marking a leap forward in local AI execution.

The model is designed for bi-arm robots and excels in highly dexterous tasks such as folding clothes or unzipping bags. By embedding intelligence locally, Google enables lower-latency, higher-privacy robotic operations, making the technology more viable for industrial and consumer environments alike. Its release of an SDK further signals intent to expand developer adoption and ecosystem maturity.

Meta’s V-JEPA 2 Ushers in “Common Sense” Robotics

Meta’s new open-source world model, V-JEPA 2, is a milestone in autonomous robotic cognition. Trained on over 1 million hours of physical-world data, it helps AI agents reason through common-sense physical scenarios—like a baby learning through observation. For instance, V-JEPA 2 can predict that a ball rolling off a table will fall, or that flour follows eggs when baking.

What makes V-JEPA 2 exceptional is its self-learning capability: it improves over time without explicit instruction. Meta positions this as a breakthrough for robotics in logistics, delivery, and self-driving contexts. The announcement comes as part of a broader industry trend—World Labs (founded by Fei-Fei Li) and Google’s Genie 2 are also investing in real-world simulation models to better connect AI with physical environments.

NVIDIA’s Quiet Robotics Success

While others grab headlines, NVIDIA continues to scale quietly but significantly in robotics. Its robotics unit pulled in $567 million in Q2 2025 alone—1% of total revenue—confirming its central role in AI-powered hardware. With demand for AI chips rising and robotics flagged as one of its top two growth areas, NVIDIA is poised to dominate the AI-robotics stack from silicon to software.

Business Implications

Opportunities

• U.S. manufacturing revival: If Project Crystal Land succeeds, it will shift the robotics supply chain westward.
• Cloudless AI adoption: On-device models like Gemini reduce latency, enhance data privacy, and cut infrastructure costs.
• Self-learning robotics: V-JEPA 2 may eliminate the need for excessive data labeling or rigid task-specific training.

Risks

• Dependence on geopolitical stability: Partnerships with TSMC and Samsung could be vulnerable to East Asia tensions.
• Capital intensity: With over $1.5T in cumulative investment across Stargate, Crystal Land, and chip fabs, any delay could create massive sunk costs.
• AI safety and regulation: As robots begin making “common sense” decisions, questions arise around reliability, liability, and legal frameworks.

Market Impact

• Chip and AI convergence: Companies like Ampere, NVIDIA, and Samsung are becoming critical nodes in the AI robotics supply chain.
• AI infrastructure race: The U.S. is accelerating efforts to counterbalance China’s Shenzhen hub and localize high-end tech production.
• Startups and developers: With SDKs, open-source world models, and new infrastructure, smaller players can now enter the robotics space faster and cheaper.

Why It Matters

The robotics resurgence is not just hype—it’s structural. With national governments, trillion-dollar firms, and foundational AI labs all aligned, robotics is transitioning from lab demos to industrial deployment. Executives should monitor infrastructure policy, invest in hardware-AI synergies, and explore partnerships in open-source robotics platforms. As physical AI agents become real-world actors, the businesses that lead in dexterity, autonomy, and reliability will dominate the next automation wave.