Why Hardware Is the Way to Win the AI Race

In this episode of Live Players, Samo Buja and Erik Torenberg discuss the pivotal role of hardware in the advancement of AI, the strategic importance of semiconductor manufacturing, and the intricate geopolitical dynamics between the US, China, and Taiwan.

—

📰 Be notified early when Turpentine's drops new publication: https://www.turpentine.co/exclusiveaccess

—
SPONSORS:
☁️More than 41,000 businesses have already upgraded to NetSuite by Oracle, the #1 cloud financial system bringing accounting, financial management, inventory, HR, into ONE proven platform. Download the CFO's Guide to AI and Machine learning: https://netsuite.com/102

—
LINKS:
Bismarck Analysis: https://brief.bismarckanalysis.com/

—
X / TWITTER:
@samoburja
@eriktorenberg
@turpentinemedia

—
HIGHLIGHTS FROM THE EPISODE:
• The thesis is that hardware is crucial to winning the AI race because capital accumulation leads to more data gathering, which has been happening since before AI through platforms like social media, smartphones, and self-driving cars.
• Two critical hardware aspects are data collection platforms (sensors, cameras, mapping) and computational infrastructure (data centers, chips), where proximity to chip manufacturing provides competitive advantages.
• NVIDIA's dominance stems partly from founder Jensen Huang's Taiwanese-American background and personal relationships with TSMC leadership, creating what could be called a "Taiwan lobby" in Silicon Valley.
• Taiwan's economic interests align with mainland China (half their manufacturing is there) while their political interests align with the US, creating strategic tensions.
• Export controls on semiconductors to China came too late and actually incentivized China to develop their full semiconductor stack independently.
• The US should create a special industrial zone near San Francisco with cheap energy, reduced regulations, and government funding to compete with China in chip manufacturing.
• China no longer has cheap labor but rather very skilled, expensive labor, especially in coastal areas where wages can reach $100-200K annually.
• A second-best option for the US would be building semiconductor manufacturing in Texas near existing infrastructure like Starbase.
• The recommended strategy includes targeting allies' semiconductor companies for replacement, implementing talent acquisition programs, and using defense/national security arguments to justify the investment.
• The US should start with purely American-made chips for sensitive government applications (NSA, CIA) as a national security imperative.
• Co-locating chip fabs, data centers, and AI labs would create a powerful feedback loop where 15-minute drives replace 15-hour flights for collaboration.
• On automation, AI will likely automate white-collar knowledge work first since digital activities are easier to train on than physical world tasks.
• Automating knowledge work could lead to dangerous political consequences, as displaced white-collar workers might turn to activism and push for protectionist policies.
• There's a risk that AI automation could lead to negative economic growth if it only automates 95% of white-collar work while leaving physical jobs intact.
• The "laptop class" poses more political danger than physical workers because they have better tools for organizing and influencing policy.
• Physical automation will likely require specialized companies for each domain (baristas, waiters, etc.) rather than general-purpose robots, leading to gradual economic growth.
• On immigration, while top talent is desirable, roughly half of crucial industry leadership should be domestic citizens to ensure patriotic alignment with national interests.
• Having homogeneous cultural groups within companies (speaking only Hindi, Mandarin, etc.) can create unfair advantages and isn't true meritocracy.
• Immigration works best with diversity from many countries rather than concentration from single nations.