AI
Perplexity Launches AI Research Unit to Build Automated Scientist
The new Frontier Research division deploys 10,000 GPUs to develop systems that can run the entire scientific discovery process autonomously.
Key takeaways
- Perplexity has launched Frontier Research, a new division dedicated to building AI systems that can conduct scientific research autonomously.
- The company has committed at least 10,000 GPUs to support large-scale experiments in pre-training, reinforcement learning, and continual learning.
- Multi-agent orchestration—coordinating specialized AI agents into a unified research workflow—forms a central technical approach.
- Andrew Gordon Wilson leads the initiative, which is hiring across four cities and promises to publish papers, models, and data openly.
- The move represents a strategic expansion from Perplexity's consumer search tools toward foundational AI research capabilities.
Perplexity pivots from search to scientific automation
Perplexity has established a dedicated research division aimed at creating AI systems capable of conducting scientific research without human intervention. The new unit, called Perplexity Frontier Research, represents a significant expansion beyond the company's consumer-facing search and research tools.
The initiative comes with substantial computational backing: Perplexity has allocated at least 10,000 GPUs to support the effort. Andrew Gordon Wilson leads the division, which is actively recruiting researchers across offices in San Francisco, Palo Alto, New York, and London.
Technical approach spans multiple AI disciplines
The research agenda centers on three core computational areas: pre-training, reinforcement learning, and continual learning. These form the foundation for experiments designed to push toward fully autonomous research capabilities.
The team has outlined several technical pillars for its work. Multi-agent orchestration stands out as particularly central to the vision. This approach involves coordinating multiple specialized AI agents, each responsible for different aspects of the research process, then integrating their outputs into a cohesive workflow.
Other focus areas include developing new pre-training architectures, advancing reinforcement learning techniques, building systems infrastructure, and creating specialized evaluation frameworks to measure progress toward autonomous research capabilities.
Why it matters
The move signals a strategic shift for Perplexity from building consumer tools to pursuing foundational AI research. If successful, automated research systems could accelerate scientific discovery across fields by handling hypothesis generation, experiment design, data analysis, and result interpretation without human bottlenecks. The 10,000-GPU commitment also indicates serious capital investment in a space where compute resources often determine research feasibility. For the broader AI industry, this represents another major player entering the race to build more capable, autonomous AI systems.
From product features to research infrastructure
Perplexity previously released products like Deep Research and Computer, which provided AI-assisted capabilities for information gathering and task completion. The new research unit marks an evolution from enhancing human workflows to attempting to replace them entirely in the scientific domain.
The company has committed to open science practices, stating it will regularly publish research papers, release models, and share datasets from the initiative. This approach could accelerate progress across the field while building credibility for Perplexity's research capabilities.
Perplexity positions the effort as potentially transformative for how scientific discovery operates, though the timeline and specific milestones for achieving autonomous research capabilities remain unspecified.
These details were first reported by Crypto Briefing.