Why the rise of open source AI isn’t hurting Anthropic … yet

In a recent commentary, Decagon's CEO Jesse Zhang presented a bold perspective on the evolving dynamics of the AI market. His article, titled "Everyone is wrong about open source AI in the enterprise," examines a compelling paradox within today's AI landscape. According to Zhang, many companies are transitioning to lighter AI models, including his own, yet the expenditure on high-end models remains largely unchanged. Zhang argues that open-source and frontier models should not be viewed as competitors. Instead, he believes they represent different phases within the same developmental cycle. High-cost frontier models serve to validate use cases, which can then be adapted for more affordable open-source options as those cases mature. As some applications shift towards these lighter alternatives, new use cases continue to emerge, leading to a stable overall investment in frontier models. Supporting his claims, recent data from Vercel’s AI gateway dashboard reveals that DeepSeek has recently surged ahead in token volume, now processing over one-third of the tokens within the company's infrastructure. Notably, Z.ai, known for its popular GLM-5.2 model, has also made significant gains. Despite these shifts, Anthropic still represents a substantial portion of spending on AI, accounting for more than half of the total outlay on the platform. Although Anthropic's share has slightly decreased due to rising prices, its dominance remains intact. OpenRouter’s statistics tell a similar story, with DeepSeek V4 Flash emerging as the top performer in token usage, processing an impressive 5.3 trillion tokens weekly, while the leading frontier model, Opus 4.8, manages just over 2 trillion. The economic landscape also reveals that the average cost of using Opus 4.8 is approximately 23 times higher than that of V4 Flash, suggesting that despite the rise of open-source solutions, premium providers continue to capture a significant share of market spending. Emerging models like Nvidia’s Nemotron are expected to make a significant impact, potentially reshaping the competitive landscape further. While these statistics do not completely validate Zhang's theory about the lifecycle of AI models, they illustrate that frontier labs such as Anthropic are not currently facing severe consequences from the rise of open-source alternatives. One potential reason for this resilience is the rapid expansion of AI tasks that can be addressed, allowing top-tier models to maintain their edge by dominating in early deployments. As Zhang succinctly states, "The frontier labs will keep owning discovery. Open source will increasingly own production.” Even as clients explore open-source options, many complex use cases remain challenging to replace with cheaper alternatives. This dual structure of the AI economy may well become a lasting characteristic, reflecting a stable coexistence between high-end and open-source models in the future.