Conference Presentation, Fireside Chat, Interview, Keynote
Winning the AI Race Part 3: Jensen Huang, Lisa Su, James Litinsky, Chase Lochmiller
- The Department of Defense partnership is projected to be recognized in approximately five years as a "first of its kind" deal generating government and taxpayer returns, accelerating the magnetic supply chain with a Texas facility featuring a "10x" capacity where the DoD acts as a full off-take and profit-sharing partner.
- Financial structures for the DoD deal include a guaranteed profit level to fund facility expansion, with the government participating as a "50-50 economic partner" on earnings exceeding a specific threshold, while normalized rare earth prices are anticipated if China identifies U.S. "national champions."
- Lisa Su anticipates the "physical AI" accelerator market will exceed "$500 billion" within a couple of years, with the segment becoming a "significant end market" in five plus years, though potentially not yet surpassing the data center chip market.
- AI-driven productivity is forecast to generate "$20 trillion" in economic impact by 2030, with a projected return of "$4.60" for every dollar invested in business-related AI applications.
- Data centers are expected to consume 10% of U.S. power by 2030, rising from a 2.5% baseline, accounting for 20% of power demand growth, while the Northern Virginia market is projected to expand from "4.5 gigawatts" in 2024 to adding a facility of equivalent size annually.
- Crusoe's partnership with Tallgrass Energy could scale to "10 gigawatts" of power, supporting the vision of "AI factories" serving as primary economic engines for societal utility and job creation.
- Adoption of AI tools is universal among chip designers and software engineers, a trend predicted to make AI programming ubiquitous by the future, where failing to use AI could lead to job loss.
- AI hardware generations are projected to increase performance per dollar and per watt by "X factors" while reducing costs by "x vectors" and increasing revenues by "x factors."
- Resale values for "Hopper" chips are forecast at approximately 80% after one year, 65% after two years, and 50% after three years, despite a fourfold performance improvement driven by software stack updates.
- The U.S. is estimated to produce "$500 billion" worth of AI supercomputing in the next four years, driving a "few trillion dollars" AI industry, with every machine and company eventually requiring a dedicated "AI factory."
- Automation is expected to encompass all moving objects "around the corner," including lawnmowers and air traffic control, transforming every industrial company into an AI company or rendering it obsolete.
- The United States plans to insource the most advanced and national security-critical components of the industry, specifically manufacturing chips and AI supercomputers in Arizona and Texas.