Winning the AI Race Part 3: Jensen Huang, Lisa Su, James Litinsky, Chase Lochmiller

MP Materials & Rare Earth Strategy

• Current Status: MP Materials is the sole U.S. supplier and refiner of rare earth materials and the only domestic manufacturer of rare earth magnets.
• DoD Partnership: Announced a $400 million public-private partnership with the Department of Defense (DoD) involving equity warrants and a 50/50 profit-sharing model for future capacity expansion.
• Price Floor Mechanism: The DoD deal includes a price floor guarantee to counteract Chinese "mercantilism" (selling below production costs), ensuring the commodity price never falls below the cost of production.
• Capacity Expansion: Accelerating the build-out of a new Texas magnetics facility intended to be 10x current capacity, with the DoD acting as the 100% off-take partner for the new volume.
• Apple Deal: Announced a separate major partnership with Apple to supply magnets, expanding their U.S. production footprint.
• Investment Timeline: Invested approximately $1 billion over eight years since acquiring assets from the bankrupt predecessor, Mollicorp, in 2017.
• Workforce Challenges: U.S. mining graduates number only 200 annually compared to China's vast output; MP Materials reports hiring 850 employees currently, with plans to add thousands more for Apple and DoD projects.
• Compensation: Median employee wage is approaching $100,000, with entry-level operators starting at $40,000–$60,000 and electricians/maintenance staff earning six-figure salaries.
• Strategic Vision: CEO Jim Lutensky predicts the DoD deal will serve as a blueprint for other critical sectors (shipbuilding, pharma, industrial diamonds) where public-private partnerships are needed to secure U.S. supremacy.
• National Security Logic: The arrangement mitigates risk by allowing the government to "hold feet to the fire" on execution timelines and costs while removing external threats like predatory pricing.

Semiconductor & AI Hardware (Lisa Su, AMD)

• TSMC Arizona Progress: AMD and TSMC achieved first silicon output (4nm chips) at the Arizona facility; yields in Arizona are now equivalent to TSMC's Taiwan operations.
• Cost Differential: Manufacturing AI chips in the U.S. is projected to cost 15–20% (low double digits) more than in Taiwan, justified by supply assurance and geopolitical risk mitigation.
• Workforce Gap: TSMC faced initial hurdles with qualified local labor, requiring recruitment and training, but has since scaled operations with significant yield success.
• Market Forecast: The accelerator chip market is projected to exceed $500 billion in the coming years; total demand is expected to grow exponentially to meet the needs of hyperscalers and sovereign AI initiatives.
• Product Diversity: Predicts a "Cambrian explosion" of chip types (ASICs, GPUs, NPUs) tailored to specific use cases (training, inference, edge, personal AI) rather than a single standard chip.
• Physical AI Timeline: Estimates the "physical AI" market (chips for robotics, drones, EVs) will equal or surpass the data center chip market within 5+ years.
• Local AI Deployment: Anticipates significant adoption of AI on local devices (laptops, PCs) to protect user privacy and reduce latency, though cloud remains essential for massive scale.
• Innovation Strategy: Emphasizes the need to "shoot ahead of the duck" by investing in future technologies years before market inflection points; acknowledges China as a highly competitive rival investing heavily in the sector.
• Education Reform: Advocates for revitalizing STEM curriculums and inspiring youth early to address the talent shortage in science and technology fields.

AI Infrastructure & Energy (Chase Iron Eyes, Crusoe)

• Infrastructure Scale: AI is creating a "new infrastructure of intelligence" requiring massive capital investment; data centers are projected to consume 10% of U.S. power by 2030, up from 2.5% today.
• Economic Impact: IDC estimates AI will generate $20 trillion in economic impact by 2030, with $4.60 return for every $1 invested in business-related AI.
• Crusoe Operations: Operating a 1.2 gigawatt AI factory in Abilene, Texas, utilizing 400,000 NVIDIA GPUs, employing ~4,000 workers daily during construction/operation.
• Modular Construction: Utilizes modular, off-site manufacturing ("Lego blocks") to accelerate the build speed of gigawatt-scale facilities.
• Partnerships:
  • Redwood Materials: Built the largest U.S. microgrid (60 MWh battery, 20 MW solar) for an AI factory.
  • GE Vernova: Partnership for 4.5 GW of new gas generation capacity.
  • Tallgrass Energy: New partnership in Wyoming to power 1.3 GW of compute load (scaling potential to 10 GW).
• Energy Sources: Pipeline spans 40 GW of capacity across SMRs, renewables, and natural gas; actively seeking sites in Texas, Wyoming, and potentially California despite regulatory hurdles.
• Labor Constraints: Identifies labor as a primary bottleneck, necessitating a multi-state recruitment strategy and significant re-skilling of oil/gas and construction workers.

NVIDIA & Industry Leadership (Jensen Huang)

• AI as Manufacturing: Defined AI as a manufacturing process where "intelligence" (tokens) is produced continuously, creating a new industrial sector rivaling energy production.
• H100/Blackwell Demand: Elon Musk's projection of 50 million H100 equivalents in five years represents a multi-trillion dollar infrastructure build-out; NVIDIA is scaling production to meet this "factory" demand.
• Performance Economics: Moving from a "one-shot" model to "reasoning" models (e.g., DeepSeek R1, Kimi K2); increasing performance-per-watt and performance-per-dollar reduces costs and drives revenue.
• Asset Longevity: Hopper GPU residual value remains 75–80% after one year; performance improves over time via software updates (CUDA stack), extending utility beyond initial hardware amortization.
• Allocation Strategy: Shifted from reactive sales to a proactive annual roadmap disclosure, allowing partners to plan power and capital expenditure years in advance.
• Physical AI Integration: Every industrial company will eventually operate two factories: a machine factory and an AI factory; "everything that moves will be autonomous."
• U.S. Competitiveness: Credits President Trump's administration for recognizing the symbiotic link between energy independence and AI leadership; asserts the U.S. is the most technology-rich nation globally.
• China's Role: Views Chinese open-source models (DeepSeek, Qwen) as a "win" for the U.S. ecosystem because they run on American tech stacks, reinforcing global developer reliance on U.S. platforms.
• Compensation Philosophy: NVIDIA uses machine learning to analyze compensation for 42,000 employees, consistently increasing OpEx to ensure retention and reward high performers with RSUs.
• Small Team Efficacy: Notes that elite AI teams (e.g., DeepSeek, OpenAI) often succeed with small groups (approx. 150 researchers) backed by massive infrastructure and funding.