On September 24, 2026, the BBC said Presidents Donald Trump and Xi Jinping would meet in Washington with artificial intelligence high on the agenda, framing the encounter as a contest to “win” the AI race. That sets the stage. The better question is who is actually ahead, and where, in the US-China AI race — and what that means for policy and business.
What the US-China AI race looks like in hard numbers
Training and deploying advanced models hinges on compute. In a visual explainer published on September 24, 2026, Al Jazeera reported that the United States accounts for nearly three-quarters of global AI computing power, citing estimates from research group Epoch AI, while China holds just over 14 percent. That gap shapes everything from model development speed to costs.
Money matters as well. The same analysis shows the US leading AI spending, which supports massive data center buildouts and access to top-tier chips. China, by contrast, remains constrained by export controls and supply bottlenecks, pushing investments toward domestic accelerators and efficiency techniques.
Research output is the area where the picture flips. Al Jazeera says China is ahead on raw publication counts and is narrowing the gap on advanced model capabilities. Quantity is not the same as frontier breakthroughs, but it signals a large and active pipeline of researchers and applied work. For companies and universities, that suggests a growing pool of talent and ideas on both sides, with the US holding the compute advantage and China pressing on research scale.
The scoreboard is mixed: the US leads in compute and spending; China leads in publications and is closing on model quality. That is the balance sheet the two leaders carry into talks.
Policy signals before the summit: standards, guardrails, and an AI incident hotline
Policy cues arrived hours before the meeting. The BBC Technology desk reported on September 24, 2026 that the US rejected pleas from OpenAI and Anthropic to adopt global AI standards drafted by industry. That stance suggests Washington wants to keep room for government-led rules, and to avoid letting vendor-written norms harden into de facto global policy.
At the same time, there is a nod to practical risk reduction. According to Al Jazeera, US Treasury Secretary Scott Bessent said Washington proposed an AI “notification mechanism” with China — a hotline to alert each other to incidents that could threaten national security. Think of it as a crisis phone for model failures, data poisoning, or runaway automation that crosses red lines.
This dual track — resisting industry-led standards while courting a bilateral safety valve — reflects the power math of the US-China AI race. The US wants to preserve its compute edge and export controls. China wants predictable channels to manage risks and avoid accidental escalation. A narrowly scoped hotline could be the first piece both sides can live with, even if they remain far apart on rules for data, chips, or content moderation.
The wider governance debate is also going public. On September 23, 2026, The Guardian reported on AI leaders addressing the UN Security Council, underscoring how pressure is building for international coordination, even as major governments diverge on who should write the rules.
What the numbers mean for business and researchers
If you build or buy AI, expect the policy tug-of-war to hit your roadmap. US firms are likely to keep priority access to high-end GPUs and cloud capacity thanks to the home-court advantage in compute power. That can shorten training cycles and cut inference latency for enterprise deployments.
Chinese firms may move faster on model distillation, quantization, and domain-specific accelerators to stretch scarce hardware. For researchers, China’s lead in publication counts means a large stream of papers and datasets, though access to top-tier compute may remain uneven.
Cross-border model licensing will stay tricky. The US rejection of industry-drafted global standards, reported by the BBC, points to a patchwork of compliance obligations rather than a single global playbook. Multinationals should map model training locations, data flows, and vendor chains in detail, and prepare for audits in multiple jurisdictions.
Vendors on both sides should factor in export controls and local security reviews when planning chip purchases or foundation model releases. A bilateral incident hotline would help, but it won’t remove legal exposure if a system causes harm or mishandles sensitive data. Using frameworks like the NIST AI Risk Management Framework can standardize internal checks while governments argue over external rules.
What to watch at the summit and after
Three signals will show whether Washington and Beijing can turn rhetoric into progress. First, any concrete language on an AI notification mechanism — scope, points of contact, and triggers — would mark a practical start. Second, references to testing and red-teaming protocols, even if nonbinding, could narrow the gap between national approaches. Third, clarity on chip export rules or narrow carve-outs would tell developers how stable the hardware supply will be over the next year.
Beyond communiqués, track the scoreboard. If the US increases its share of global compute or accelerates data center timelines, that advantage compounds. If China announces credible, large-scale domestic accelerators or partnerships that raise effective compute, the balance shifts. Watch for the release pace of top-tier models, not just paper counts, and for fresh funding tied to hardware and energy buildouts.
The summit won’t settle the contest. But it can set habits for handling risk while the US-China AI race grinds on. Whether you run a lab, a startup, or an IT budget, the numbers — and any new crisis channel — will decide how fast you can build and how safely you can operate. For more on this, see bloomberg.com and nytimes.com.
Related reading: NVIDIA • Meta AI • AI & Big Tech