645: How China's AI Punches Above Its Capex, Google's Brain Drain, Neural Software, SpaceX's $6.3B Compute Deal, A24, Five Gens of TPUs, Amazon's Robots Learn English, and Three Beams of Light

2026-06-24 · Source: Liberty’s Highlights · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation, Economic Analysis & Policy · Depth: Fundamental Awareness, extended

Summary

China's AI sector exhibits a "fast-follower advantage," achieving near-frontier model performance, like DeepSeek R1 and GLM-5.2, with significantly lower capital expenditure than US counterparts. This efficiency comes from leveraging pioneers' discoveries and focusing resources on proven approaches. Google, however, faces a potential "brain drain" as key AI talent, including AlphaFold co-winner John Jumper and Transformer co-inventor Noam Shazeer, depart for competitors. In compute, SpaceX secured a \$6.3 billion deal to supply Nvidia GB300s to Reflection, an open-source AI lab. The concept of "neural software" proposes large language models directly generate application outputs, potentially transforming software development within 10-15 years. Google also invested \$75 million in A24 for AI tools in film production. Its TPU development from v2 to Ironwood shows a 100X performance increase and 30X power efficiency gain over eight years, driven by architectural stability.

Key takeaway

For AI/ML Directors evaluating compute investments or market strategies, recognize that "fast-follower" approaches can yield competitive AI capabilities with reduced capital expenditure by focusing on proven methods. Your teams should prioritize architectural stability in hardware development, as Google's TPU evolution demonstrates, to enable long-term software optimization despite shifting workloads. Additionally, consider the long-term implications of "neural software" on application development and value capture, preparing for a future where LLMs are the software.

Key insights

Fast-follower strategies enable rapid AI advancement with lower capex by exploiting pioneer discoveries and stable hardware architectures.

Principles

• More output increases both masterpieces and duds.
• Second movers learn from pioneers' expensive mistakes.
• Architectural stability aids software optimization.

In practice

• Utilize free, local voice-to-text apps like Fluid Voice.
• Explore neural software for non-deterministic applications.

Topics

• AI Strategy
• Compute Infrastructure
• Neural Software
• TPU Development
• Voice-to-Text
• Robotics Automation
• Film Production AI

Code references

• altic-dev/FluidVoice

Best for: CTO, VP of Engineering/Data, Entrepreneur, Director of AI/ML, Investor, Tech Journalist

Related on AIssential

• 633: AI’s All-You-Can-Eat Era is Ending, Google's Two-Chip TPU Strategy, OpenAI & Microsoft's New Deal, Musk's AI Reboot, SpaceX AI & Cursor, My Thoughts on GPT-5.5, and Pancreatic Cancer Vaccine — AI models are advancing rapidly, driving shifts in business models and hardware strategies.
• AI might make us dumber, but it will make companies a lot smarter — AI enables firms to compound institutional knowledge, creating a first-mover advantage unlike previous tech shifts.
• Amazon's $200B CapEx Spend Dominates AI Race — Tech giants are in an AI compute arms race, investing heavily in infrastructure despite investor skepticism.
• ChinAI #358: Around the Horn (25th episode) — China's AI ecosystem is dynamic, marked by rapid startup valuations, evolving business models, and significant talent and regulatory challenges.
• Codex, Cursor and the Great Beyond — Intense competition and strategic acquisitions are rapidly reshaping the AI market, challenging early leaders.
• This Week in AI: Your Recap — OpenAI faces critical financial strain and technological limits, challenging its market leadership and long-term viability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Liberty’s Highlights.