China Just Made an Offer to the Global South. LatAm's AI Founders Are the Ones Who Should Negotiate.

The US export control regime was built on a specific theory: contain China's AI capabilities long enough for American companies to establish irreplaceable market positions globally. Bloomberg reported today that Beijing's answer is a program called "AI for All" — a systematic push to establish Chinese AI as the infrastructure layer for markets the US never prioritized.

The context matters. The US and China are now formally scheduling AI talks for September 2026, ahead of Xi Jinping's planned US visit on September 24. The talks, led on the American side by Treasury Secretary Bessent, will address military AI, cyberattack risk, open-weight model proliferation, and market access. That these two superpowers need a formal negotiating table to discuss AI governance is itself evidence of how contested the AI supply chain has become.

Chinese AI API pricing already runs at roughly one-third of US frontier model costs. DeepSeek V4 and Kimi K2.6 — both open-weight, both globally accessible — perform at frontier-competitive levels on coding, reasoning, and agentic tasks. The "AI for All" push goes further: subsidized infrastructure deployment, government-to-government agreements on AI technology access, and a deliberate positioning of Chinese models as the accessible alternative to US-controlled AI infrastructure.

For LatAm AI founders, this is not a threat. It is leverage. Brazil has no prohibition on Chinese AI infrastructure — ByteDance runs a data center in the country, and Chinese models are already in use across the startup ecosystem. Brazilian companies building at the application layer now have two competing clouds of AI capability competing for their infrastructure spend. That means lower costs, more options, and — critically — the credibility to negotiate data sovereignty terms with either provider.

The White House is finalizing a 30-day national security review window for frontier model releases, expected before August 1. If implemented, this will create a systematic delay between US-based enterprise access to frontier AI and international access. Brazilian fintechs building on Chinese open-weight models face no such window.

The application-layer thesis has always argued that the model is commodity and the value is in the workflow. The US-China AI competition is now accelerating that timeline. Founders who treat model selection as a fixed architecture decision rather than a continuous cost-and-capability optimization are making a strategic error. The geopolitical race has an unintended beneficiary: the application-layer companies with flexible infrastructure and a local market nobody else wants to serve.