OpenAI Doubles Agentic User Base as Cheaper Chinese Models Threaten US Dominance

On Tuesday, Bloomberg reported that the mega AI startup is set to announce that its two agentic tools, Codex and ChatGPT Work, are now serving about 10 million users, or double the number from earlier this month. The timing of the big reveal isn’t exactly surprising; the company’s much-awaited IPO has seemingly stalled as competitive open-weight models from overseas grab market share in the US. Whether the US market will remain open forever is another, suddenly pressing, question.

In March, leaders at OpenAI reportedly urged staff to abandon “side quests” (RIP, video-generator Sora) in favor of money-making agentic coding and enterprise tools amid rising competition. Back then, the company’s main concern was the rapidly rising Anthropic, whose Claude Code and Claude Cowork tools fueled tokenmaxxing.

Whether the new figures place OpenAI in the lead is unclear, but both companies are facing a much more existential threat: open-weight and open-source Chinese models from the likes of Moonshot, Alibaba and DeepSeek, which have sucked up roughly half of US enterprise token usage by proving nearly as powerful and far less expensive. The AI Race is now sparking a policy debate over AI protectionism in Washington … and Beijing:

Meet in the Lobby: Reuters also reported that Beijing and Washington are still planning an AI-focused summit in September. In the meantime, OpenAI and Anthropic are spending big to make sure their voices are heard. Together, the two spent nearly $3.2 million on DC lobbyists in the second quarter, up 23% from the previous quarter, according to recently released federal filings.