Interview
Perplexity CEO: Micron Will Be More Valuable Than Meta & How Export Controls Helped Not Hurt China
- Public resistance regarding job losses, wealth inequality, and environmental concerns is expected to worsen, currently halting 40% of proposed data center developments, with power supply and permitting speed identified as the primary bottlenecks for construction over the next three years.
- Industry revenue models are projected to shift toward subscription and usage-based non-advertising streams, specifically driven by deep research and autonomous agent loops that handle complex tasks like software engineering, chip design, and drug discovery rather than simple conversation.
- Financial forecasts predict Micron will surpass Meta in value within six to 12 months, while OpenAI is deemed unready for an IPO, and Perplexity aims to go public around 2028 with a valuation potentially reaching $200 billion with 4,000 staff or $2 trillion with 10,000 staff.
- AI agent traffic is already surpassing human traffic and will increasingly drive costs, with power becoming the dominant constraint over memory by 2028 and token prices expected to decline in 24 months despite higher usage by power users.
- The competitive landscape will favor companies solving the orchestration problem to balance local and server-side models, with open-source models potentially matching frontier capabilities like Opus for eight at 10 times the cost within 12 months.
- Geopolitical risks involving export controls are expected to force China to build a sovereign tech stack, potentially creating a more potent competitor in physical AI and data center build-outs, while the US maintains short-term advantages in frontier capabilities.
- Business sustainability for infrastructure providers depends on offering software orchestration rather than just server rental, and a "deep seek moment" involving more efficient model architectures could threaten companies that have overbuilt capacity.
- Future company structures are expected to evolve with smaller software teams utilizing hundreds of thousands of agents, creating opportunities for new billionaires within 12 months and potentially 100,000 new multi-billion dollar entities through initiatives like the "billion dollar build" program.
- The "frontier" of AI is predicted to move toward solving infinite economic value problems such as curing cancer and understanding the universe, driven by physical infrastructure investments in fabs by companies like TSMC, Intel, and NVIDIA.
- Consumer interfaces will shift from conversational chat to exploration-based search, where advertising remains viable for subjective decisions but agent-based systems will dominate objective transactions, necessitating a focus on value per watt per user as the key long-term metric.