ARK Invest's Cathie Wood has presented a counterintuitive argument about the impact of open-weight AI models on frontier AI companies. According to reports from ARK Invest, Wood argues that open-weight AI models are actually making companies like OpenAI and Anthropic richer, contrary to the popular narrative. An open-weight AI model allows anyone to download, run, and modify it on their own hardware without paying for access, creating a different competitive dynamic than traditional subscription-based models.
Wood's analysis centers on the security implications of open-weight models rather than their competitive threat. As reported by ARK Invest, the real threat is not that open-weight models replace frontier AI, but that bad actors use them to attack enterprises. This security pressure forces companies to maintain frontier-grade AI subscriptions as defense layers, creating sustained demand for premium services. The UK's AI Security Institute found that open-weight models now match frontier cyber capabilities from just four to seven months ago, highlighting the growing security gap.
Wood has specifically identified which companies will benefit most from this security-driven demand. According to ARK Invest, she named OpenAI, Anthropic, and SpaceXAI as the companies most likely to take the majority of model-driven revenue. ARK holds positions across all three companies, with Anthropic filing its S-1 at a near $1 trillion valuation and OpenAI eyeing a September 2026 debut. Wood stated that 'Ironically, contrary to the narrative, open weight models are becoming an important reason that OpenAI, Anthropic, and ultimately, in our view, SpaceXAI are likely to take the vast majority of model-driven AI revenue.'
The investment thesis reflects broader international discussions about AI governance and ecosystem design. Recent global AI governance discussions have emphasized the need for a protected open global AI ecosystem with both frontier and open-source models to widen access and support digital sovereignty. Multiple countries, including Russia, Egypt, and Costa Rica, have emphasized open-source models, lightweight models, non-discriminatory access and resistance to concentration. The debate highlights that countries with limited resources should avoid trying to win a frontier compute race and instead invest in enabling environments, leadership, workflows, local adaptation, lightweight or frugal models and practical applications tied to local needs. Recent discussions at international forums have reinforced this approach, with experts arguing that open, affordable and locally adaptable AI models are preferable to exclusive concentration and compute-heavy races, especially for developing countries.
