OpenAI Backs $650M AI Agent Swarm Startup Isara

A San Francisco startup developing software to orchestrate thousands of AI agents on complex analytical work has secured a $94 million investment, achieving a $650 million valuation just nine months after its founding. OpenAI is a notable participant in this funding round for Isara, a company with no commercial product yet on the market. The venture was launched in June 2025 by two 23-year-olds, Eddie Zhang, a former OpenAI safety researcher, and Henry Gasztowtt, an Oxford computer science student. Their academic collaboration, including a co-authored paper at ICML 2024 on AI systems cooperating for policymaking, forms the intellectual bedrock of the company. Other investors include Amity Ventures, former CAA chairman Michael Ovitz, and billionaire hedge fund manager Stanley Druckenmiller. Isara has since expanded its team, hiring roughly a dozen additional researchers from leading firms like Google, Meta, and OpenAI.

The company’s core proposition is multi-agent coordination at an unprecedented scale. While most current AI applications rely on a single model responding to a prompt, Isara’s architecture is designed to enable hundreds or even thousands of specialized agents to communicate, divide labor, align on objectives, and synthesize a unified output. The founders frame this as an evolution from isolated tools to intelligent, coordinated teams. In a demonstration, the platform coordinated approximately 2,000 agents working in concert to forecast the price of gold. The initial commercial focus is on providing predictive modeling software to investment firms, with biotechnology and geopolitical analysis identified as secondary markets. The long-term ambition is to train these agent swarms to track geopolitical shifts and forecast broad economic trends.

This ambition presents a formidable technical challenge. Achieving reliable performance from a single AI agent on a complex task is difficult. Coordinating thousands without cascading errors, conflicting goals, or compounding hallucinations is a problem academic literature has scarcely addressed at this magnitude. Existing frameworks like LangChain, CrewAI, and AutoGen typically manage small numbers of agents on relatively structured tasks. Isara’s goal of orchestrating thousands on open-ended analytical problems represents a different order of difficulty entirely.

Isara exemplifies a broader trend in artificial intelligence investment known as the neolab phenomenon. These are research-intensive startups, often founded by alumni of elite labs like OpenAI, DeepMind, and Google Brain, that operate more like privately funded research institutions than traditional companies. In a recent one-month span, investors poured or discussed $2.5 billion into five such ventures. Industry analysts estimate over $10 billion has now flowed into this category, reflecting a major bet that the next significant AI breakthroughs will stem from architectures fundamentally different from today’s dominant large language models.

The pattern is consistent: a small team with elite credentials and top-tier publications raises capital at a valuation in the hundreds of millions long before generating revenue. The investor thesis holds that foundational research capability, not a specific product, is the scarce asset worth funding. In a market where companies like Cognition can achieve multi-billion dollar valuations with modest revenue, the potential upside from a genuine breakthrough in multi-agent coordination is seen as enormous. The inherent risk, however, is that foundational research is profoundly uncertain. Many architectures explored by neolabs, including diffusion models for reasoning, world models, and agent swarms, remain unproven outside controlled demonstrations. The chasm between a demo forecasting gold prices and a production system trusted by investment firms for real capital allocation is vast.

OpenAI’s involvement raises a strategic question. Why would a frontier lab invest in a startup founded by a former employee working on problems adjacent to its own research? The simplest answer is strategic optionality. If large-scale multi-agent coordination becomes a critical capability, OpenAI gains exposure to external development approaches. For a company raising capital at a $300 billion valuation and committed to building artificial general intelligence, a $94 million investment is relatively inexpensive insurance. There is also a crucial talent dimension at play. In the AI industry, top researchers are arguably the most valuable resource. By investing in Isara, OpenAI maintains a relationship with Zhang and his team, mitigating the total loss of their expertise to the competitive landscape. This dynamic mirrors investments by Google, Microsoft, and Amazon into smaller AI labs, where the cost of funding a startup is often lower than the cost of losing elite talent.

For Isara, the path forward involves translating a compelling research vision into a reliably functional product for paying customers. The agentic AI market is projected to grow from $7.8 billion in 2025 to over $52 billion by 2030, and every major platform is rapidly integrating multi-agent features. Isara’s fundamental bet is that coordination at the scale of thousands of agents requires an architecture wholly distinct from what incumbent platforms offer. The central question for the coming year is whether this specialized approach will prove correct, or whether the established players will incrementally solve the coordination problem as they scale their existing models.