Conference Presentation, Panel, Fireside Chat, Keynote
Why Top Founders Are Racing Into AI Infrastructure
- AI infrastructure demand is projected to necessitate a shift from copper to new power and system software solutions, driven by hardware bottlenecks "south of the model" that persist as model capabilities improve faster than infrastructure.
- Hyperscalers are expected to collectively reach a trillion dollars in capital expenditure next year, while GPU supply remains fully booked through 2028 with multi-day auctions for thousands of units.
- Token consumption per task is anticipated to increase by orders of magnitude due to a shift toward complex reasoning and agent-based workflows, with total token demand growing nearly 1,000% annually and outpacing supply growth.
- New data centers require 44 gigawatts of additional power by 2028, creating a deficit against grid additions of only 25 gigawatts, alongside a complete industry transition to liquid cooling and a shortage of DC power certified electricians.
- The market is expected to fragment into specialized hardware niches, with the proportion of deals targeting complex hardware problems rising from 5% to over 20% or 30% among top founders.
- Business success is forecasted to depend on efficiency and profitability rather than initial adoption, as margins shift toward hardware optimization and long-term compute investment extends for decades.
- Deployment of robots for server assembly is expected to increase, accompanied by the emergence of a new generation of hardware founders skilled in complex chip manufacturing supply chains.
- If the Machine Age Fund performs as anticipated, the U.S. is projected to win the infrastructure game by 2030 or within the next five to 10 years, resulting in abundant, super eco-friendly, and efficient data centers.