China-US AI Cooperation Is a Necessity, Not a Choice

When Gary Dvorchak sat down with CGTN this week, the managing director of the San Francisco-based investor relations firm The Blueshirt Group had a message for the two most powerful capitals on Earth: when it comes to artificial intelligence, Washington and Beijing do not have the luxury of going it alone. “When machines get smarter than us, doing it alone won't cut it,” he argued in an interview published Thursday, framing cooperation not as a diplomatic courtesy but as a strategic imperative.

His timing could hardly be better. The interview lands in the same week that Washington and Beijing are quietly preparing their first official AI dialogue under President Donald Trump's second administration — high-level talks expected to take place in September, ahead of Chinese President Xi Jinping's planned visit to the United States on September 24. After more than a year of escalating export controls, tariffs and accusations of intellectual property theft, the question is no longer whether China and the United States will compete. It is whether they can compete without destroying the shared technological ecosystem both depend on.

The AmCham Case for Cooperation

Dvorchak is not an academic theorist or a policy think-tank fellow. He is a practitioner with decades of China experience, and he carries a rare piece of personal history: his family hosted a young Xi Jinping during the future Chinese president's first visit to the United States in 1985, when Xi was a relatively junior local official from Hebei. That connection has made Dvorchak a familiar voice on CNBC and in Chinese state media, where he is frequently asked to explain Chinese tech policy to Western investors.

His argument on AI is disarmingly simple: the technology is too powerful, and its risks too global, for unilateral control. At the 2026 World AI Conference in Shanghai in July, Dvorchak made the same case to a global audience, urging a people-centered approach to AI governance. “We're competitors, but we're also human beings. We're actually on the same team,” he said, remarks amplified by CGTN and the Chinese Embassy in Australia. The CGTN interview published Thursday expands on that theme, with Dvorchak walking through the state of AI advances, offering career advice for young people entering an AI-driven job market, and reflecting on the people-to-people ties that have kept the US-China relationship alive through its most difficult years.

A Rivalry Measured in Months, Not Years

Dvorchak's insistence on cooperation comes at a moment when the technological gap between the two AI superpowers is narrowing faster than most Western policymakers assumed. At the Aspen Security Forum in mid-July, Tarun Chhabra, Anthropic's head of national security policy, put a number on it: the United States leads China by roughly six to nine months in frontier models — “kind of an average if you look at models,” in his words. Without China's use of distillation, the practice of training cheaper models to mimic the outputs of stronger ones, Chhabra estimated the American lead could stretch to a year or 18 months.

Independent data tells a similar story. Stanford University's 2026 AI Index found that the US-China model performance gap has effectively closed, with US and Chinese models trading the lead multiple times since early 2025. As of March 2026, Anthropic's top model led the best Chinese model by just 2.7 points on the index's benchmark aggregate. That is not the comfortable moat that defined the early years of the AI boom; it is the difference between one release cycle.

The narrowing gap is reshaping the debate in Washington. The question is no longer whether to restrict China's access to AI technology, but what to restrict — and whether some restrictions backfire.

September Talks: A Diplomatic Opening

The planned September dialogue marks a significant escalation from the working-level AI discussions the Biden administration held with China in Geneva in 2024. According to a July 22 report from Modern Diplomacy, the talks are one of the most significant outcomes of the Trump-Xi summit held in May, though the agenda, location and final dates have yet to be confirmed. US Treasury Secretary Scott Bessent is expected to lead the American delegation, with the identities of other participants still being finalized.

Beijing has signaled its willingness to engage. China's Foreign Ministry said the country remains willing to work with Washington to implement the consensus on artificial intelligence reached by Presidents Xi and Trump, describing AI as one of the major challenges facing the international community. For a relationship that spent much of 2025 in a cold-war-style spiral of chip embargoes and retaliatory export bans, the very existence of a scheduled high-level dialogue is itself a signal: both governments recognize that AI competition, left unmanaged, could become an existential security problem.

The Watermark War

But the path to September is littered with landmines, and the sharpest one right now is intellectual property. Speaking on Fox Business Network this week, Bessent stopped short of confirming the planned meeting but made Washington's concerns explicit. “We are finding watermarks of our U.S. large language models on many of the Chinese models, and that's unacceptable,” he said. “We're going to be looking at that in the coming days or weeks.”

The watermark issue cuts to the heart of the cooperation debate. American labs argue that Chinese developers are extracting the most valuable intellectual property from US models through distillation, and Washington has begun to act. On June 12, 2026, the US Department of Commerce ordered Anthropic to deny access to its most powerful models — Mythos and Fable — to any non-US persons. Because Anthropic could not reliably verify user citizenship, it took both models down for all users globally. Fable's access was restored on July 1 with stricter guardrails; Mythos remains limited to American institutions.

The episode exposed the limits of the US approach. Chinese labs largely ship open-weight models that users can download and run locally, meaning no provider can revoke access after the fact. When Washington abruptly cuts off a closed American model with no clear public criteria, analysts at the Peterson Institute for International Economics argued, it risks handing China an opening rather than closing one — a self-inflicted wound in a race measured in months.

Export Controls and Their Limits

Beyond model access, the US toolkit has expanded dramatically through 2026. The Commerce Department's Bureau of Industry and Security shifted advanced chips equivalent to Nvidia's H200 and AMD's MI325X from a presumption of denial to case-by-case review in a final rule published January 13. A 25 percent value-based tariff on covered advanced AI chips not bound for the US supply chain took effect January 15. Congress approved the Chip Security Act on March 26, requiring location-verification technology embedded in advanced AI chips to detect illegal diversion. A more aggressive measure, the GAIN AI Act, which would have forced chipmakers to prioritize US customers over export markets, was left out of the final 2026 National Defense Authorization Act after industry pushback.

Yet the controls are not stopping China's advance. On July 29, Chinese memory chipmaker CXMT surged 466 percent on its STAR Market debut after raising the equivalent of 12.5 trillion won, lifting its global DRAM share toward 10 percent and rattling Korean and US chip stocks. The message to Washington is uncomfortable: restrictions may be slowing China's access to cutting-edge hardware, but they are accelerating the domestic substitution that Beijing has been demanding for years.

What This Means

For investors and technologists watching from Tokyo, Singapore and Silicon Valley, the Dvorchak interview and the September talks point in the same direction: the era of pure decoupling is over, replaced by a more complicated game of managed interdependence. The United States still holds the edge in frontier models, compute and the world's most valuable AI companies. China holds the edge in scale, manufacturing, open-weight distribution and the political will to subsidize domestic champions.

Neither side can win by refusing to talk. AI systems that can strengthen military capabilities, facilitate cyberattacks on critical infrastructure and disrupt labor markets are precisely the kind of shared risks that the 2025 Detroit consensus — the US-China dialogue agreement that paired America's AI Action Plan with China's AI Plus initiative — recognized as “systemic risks that no single country can address alone.” The September talks are the first test of whether that logic survives contact with Bessent's watermark warnings and Commerce's model-access orders.

The Human Dimension

Dvorchak's third theme — career advice for a job market being reshaped by AI — is easy to overlook in the geopolitical drama, but it may be the part that matters most to ordinary people on both sides of the Pacific. His message to young people, distilled in the CGTN interview, is that the machines are coming for routine cognitive work, and the humans who thrive will be those who learn to work alongside the models rather than compete with them.

That advice carries a geopolitical corollary. The people-to-people ties Dvorchak embodies — a 1985 dinner in Iowa, a CNBC segment, a Shanghai conference hall — are the connective tissue that keeps the US-China relationship from snapping entirely. When governments stop talking, businesspeople and students and researchers still find ways to cooperate. When they do cooperate, the shared infrastructure of global AI — the chips, the talent, the open models, the standards — keeps working for everyone.

September will test whether that fragile logic holds. The talks may fail; the watermarks may poison the well; another export order could land next week. But Dvorchak's core argument stands on the evidence: in a race where the leader's advantage is measured in months, the only sustainable strategy is to manage the rivalry — not to pretend the other side doesn't exist. As he put it, cooperation isn't a choice. It's a necessity.

By Kenji Tanaka, Global1.News

This article was produced with AI-assisted research and editorial support. Reporting is based on sources cited in the article.