Interview
Perplexity CEO: SMBs Could Drive the Next Economic Boom
- AI is expected to become the primary accelerant of human curiosity by shifting the role of human intellect from seeking answers to formulating the next set of questions.
- The company aims to build a "trillion-dollar" entity over a half-decade to decade timeframe by serving American small and mid-market companies that are becoming AI-native and capital efficient.
- Future American GDP growth is projected to stem primarily from smaller and mid-market enterprises leveraging AI tools to create value faster than traditional large enterprises.
- Revenue growth is predicted to be more pronounced in the enterprise segment this year as the product unlocks real business operations, though usage is expected to rise among existing users as serving costs decrease due to model competition.
- The business strategy prioritizes remaining a model-neutral platform that orchestrates both frontier and open-weight models, betting on the latter for local execution and the former for cloud orchestration.
- Users will increasingly demand the ability to orchestrate between local hardware and the cloud to ensure privacy and data sovereignty, driven by a need to protect against bottlenecks in data center build-outs caused by permitting, interconnection queues, and gigawatt requirements.
- Cost efficiency gaps between cloud and local hardware are expected to narrow to a 3x or 1x ratio as context windows expand, making local hardware a viable alternative for long-running autonomous agents.
- The company plans to incentivize tapping into existing power and RAM in homes and offices rather than building its own data centers, while simultaneously increasing demand for third-party data centers.
- Product positioning will focus on solving problem-solving, clarity of thought, and deep research analysis challenges rather than code writing, using multi-model orchestration to demonstrate trust and help other businesses succeed.
- Security measures will include developing sandboxes with background guardian agents, utilizing open-source releases like "Bumblebee" for devices and "Numbat" for servers to audit trails and detect multi-agent threats.
- Long-horizon research tasks and real-time hypothesis testing are planned to enable users to function as "scientists," despite current uncertainties regarding the optimal interface for such autonomous tools.
- Rollout times for new AI models are predicted to slow significantly as single-agent testing fails to identify multi-agent threats, necessitating the described security protocols.
- Sovereign deployments are anticipated to grow as countries and sovereign funds push for systems running on fully owned hardware with fine-tuned open-weight models.
- The business expects to maintain a unique position as a neutral model orchestrator for knowledge work, where other model developers act as partners rather than competitors.
- Material risks include potential data center infrastructure bottlenecks and the challenge of creating intuitive interfaces for long-horizon autonomous research products.