Will the scale of on-chain AI data expand by 100 times? Mechanism Capital reveals key variables for 2026

According to the latest news, Mechanism Capital partner Andrew Kang recently stated that by 2026, the scale of physical AI data will expand by 100 times. This prediction reflects significant technological breakthroughs in the robotics and physical AI fields in 2025, with key advances in model architecture, training methods, and data collection.

2025 Technological Breakthroughs Paving the Way for Data Explosion

In 2025, the robotics field solved several long-standing core problems. According to Andrew Kang’s analysis, these breakthroughs include:

• Innovations in model architecture and training methods enabling AI systems to learn more efficiently
• Breakthroughs in data collection technology making large-scale data gathering feasible from an ideal state
• Progress in understanding data quality and data recipes, improving data usability
• Innovative applications of reinforcement learning technology, allowing companies like Figure, Dyna, and PI to achieve over 99% success rates in real-world scenarios
• Breakthroughs in memory technology, breaking the previous “memory wall” limitations

Key shift from theory to practice

What do these advances collectively point to? It is that AI companies now have confidence to invest in large-scale data collection. In other words, the technological progress in 2025 solves the question of “can it be done,” while the data explosion in 2026 addresses “how to do it at scale.”

Andrew Kang mentioned that technologies like NVIDIA’s ReMEmber, Titans, and MIRAS enable memory during testing, and more advanced visual language models (VLM) provide stronger spatial understanding for visual language action models (VLA). These advances mean systems can handle more data and extract deeper value from it.

Market Significance of 100x Data Scale Growth

Why is the number 100x so critical?

According to recent reports, by 2025, the market has already begun to see capabilities enabled by increased data scale, such as zero-shot mapping, visual sensitivity, and general physical reasoning. In other words, larger-scale data has started to demonstrate new capability dimensions. The expected 100x growth implies these capabilities will be released exponentially.

Echoes in AI Applications within the Cryptocurrency Field

Interestingly, this trend resonates with upgrades in AI applications within the crypto space. According to the latest information, Nansen AI will be upgraded to a full-stack on-chain trading product in 2026, supporting all on-chain transactions via AI. From data analysis to trade execution upgrades, this somewhat reflects the progress of physical AI in data processing and decision-making capabilities being applied across multiple domains.

Future Focus Areas

The potential 100x growth in physical AI data scale in 2026 suggests several possible development directions:

• Significant improvements in the versatility and adaptability of AI systems
• Increased efficiency in data annotation and processing workflows, potentially becoming new competitive focal points
• Further expansion and deepening of physical AI application scenarios
• A substantial rise in infrastructure and tool demands related to these developments

Summary

Andrew Kang’s prediction is not made out of thin air but is based on substantial technological breakthroughs in the physical AI field across multiple dimensions in 2025. From reinforcement learning to memory technology, and from data collection to data quality understanding, these advances collectively point toward an era of large-scale data collection and application. The projected 100x increase in data scale reflects a shift from “can it be done” to “how to do it at scale,” and this transition may gradually become evident across the entire AI industry chain by 2026.