The internet's best case and worst case are not separated by a generation or a continent. They are separated by a connection, an investment or a safety guardrail – often by a single missing condition.

At this year's World Internet Conference Wuzhen Summit, the best case was concrete. Alibaba's Joseph Tsai argued that only open source can make AI inclusive, citing a Ugandan nonprofit that used the company's Qwen model to cover 68 African languages. Liu Dayiheng, who leads the model, added that open weights keep prices competitive, preventing a few firms from monopolizing access.

BrainCo offered the human version of the same story: neural-controlled prosthetics have helped more than 10,000 disabled people, including 2,046 in Zhejiang who received free limbs in 2025.

The worst case was equally concrete. UN Under-Secretary-General Li Junhua said about 2.2 billion people are still offline, for whom the promise of AI and digital transformation remains far off and prospects are vanishingly small. Meaningful connectivity, Li said, requires digital literacy, skills and a safe online environment – not just a signal. The International Telecommunication Union has warned that in the AI era, being offline means losing access to education, jobs and services, not merely a slower connection.

Read more: Inclusion, AI literacy and accessibility focus of China internet reports

Where best and worst nearly meet

The one-step distance is clearest in open-source safety. John Higgins of the International Association for AI Governance said responsibility in open-source communities cannot be left to chance or a few clever academics, and called for credible cross-border evaluation bodies. Liu Dayiheng was more specific: once an open model is released, it cannot be recalled, and third-party developers can bypass guardrails.

DeepSeek engineer Gao Xi conceded that open source "may amplify safety risks," but argued that closed models carry the same dangers. ZTE chairman Fang Rong offered the industry formula: the faster AI runs, the better the seatbelt. However, a seatbelt only works if someone is required to wear it.

That is the missing step. The best case studies – open models, cheap prosthetics, AI-assisted healthcare – all presuppose users who are already connected, financed and served by functioning institutions. The worst data – 2.2 billion offline, a widening AI divide, deepfake fraud and algorithmic harm – describe people and risks outside those preconditions.

GSMA chairman John Hoffman framed the choice as bridges instead of walls. Yet the summit produced no verifiable metric for whether the least-connected are actually being reached, nor a clear answer on who bears first-order liability when an open model is misused downstream.

Watch: John Hoffman says China well-positioned to lead in the 6G era

The internet is not simply getting better or worse. It is splitting into two speeds: one where AI becomes a public utility, and one where connectivity remains a distant promise. The gap between them is one step – accountable inclusion. That step is the story still unresolved.